BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Department of Electrical & Computer Engineering - ECPv6.15.18//NONSGML v1.0//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
X-WR-CALNAME:Department of Electrical & Computer Engineering
X-ORIGINAL-URL:https://ece.northeastern.edu
X-WR-CALDESC:Events for Department of Electrical & Computer Engineering
REFRESH-INTERVAL;VALUE=DURATION:PT1H
X-Robots-Tag:noindex
X-PUBLISHED-TTL:PT1H
BEGIN:VTIMEZONE
TZID:America/New_York
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
DTSTART:20220313T070000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:20221106T060000
END:STANDARD
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
DTSTART:20230312T070000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:20231105T060000
END:STANDARD
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
DTSTART:20240310T070000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:20241103T060000
END:STANDARD
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
DTSTART:20250309T070000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:20251102T060000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20240415T133000
DTEND;TZID=America/New_York:20240415T143000
DTSTAMP:20260414T035601
CREATED:20240411T010005Z
LAST-MODIFIED:20240411T010005Z
UID:6893-1713187800-1713191400@ece.northeastern.edu
SUMMARY:Lin Deng PhD Dissertation Defense
DESCRIPTION:Name:\nLin Deng \nTitle:\nFunction Capacity Expansion of Nano-Optics via Multiplexed Metasurfaces \nDate:\n4/15/2024 \nTime:\n1:30:00 PM \nLocation:\nSL 011 \nCommittee Members:\nProf. Yongmin Liu (advisor)\nProf. Hossein Mosallaei\nProf. Sunil Mittal \nAbstract:\nThroughout history\, the exploration of light has been fundamental to our understanding of the world and has driven advancements in technology and communication. Metasurfaces\, composed of rationally designed nanostructures\, offer a revolutionary means to control light in a prescribed manner. Metasurfaces can operate in conventional free space\, and the emerging integrated photonics domain. Maximizing functionality and degrees of freedom (DOFs) in both arenas is paramount. My thesis aims to push the limit of metasurface capabilities by leveraging multiplexing strategies across input/output parameters such as polarization\, incidence angle\, and waveguide mode. I will present three novel metasurfaces as follows. \n(1) We aim to expand nano-printing multiplexing capacity using the Polarization-Encoded Lenticular Nano-Printing (Pollen) method. When employing three input/output polarization pairs and varying detection angles\, a single metasurface device enables the observation of up to 49 high-resolution nano-printing images. \n(2) By integrating metasurfaces with waveguides\, we can couple guided modes to free space while controlling wavefront and polarization. Our research exploits the multiplexed on-chip metasurface\, which could generate multiple functions depending on the polarization states and waveguide mode propagation directions. \n(3) We investigated mode division multiplexing (MDM) for high-volume optical transmission\, enabling multiple waveguide modes to coexist without interference. By manipulating the orientations of individual nanoantennas\, we have achieved on-demand mode conversion and focusing effects\, demonstrating promising results in various scenarios.  \nIn conclusion\, my research seeks to push the boundaries of metasurface functionalities through innovative multiplexing approaches. The research findings allow us to unlock new possibilities in optical display\, communication\, manipulation\, and beyond by integrating multiple functionalities into single free-space and on-chip metasurfaces.
URL:https://ece.northeastern.edu/event/lin-deng-phd-dissertation-defense/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20240412T123000
DTEND;TZID=America/New_York:20240412T140000
DTSTAMP:20260414T035601
CREATED:20240403T221805Z
LAST-MODIFIED:20240403T221805Z
UID:6855-1712925000-1712930400@ece.northeastern.edu
SUMMARY:Peng Wu PhD Dissertation Defense
DESCRIPTION:Announcing:\nPhD Dissertation Defense \nName:\nPeng Wu \nTitle:\nBayesian Data Fusion for Distributed Learning \nDate:\n4/12/2024 \nTime:\n12:30:00 PM \nLocation:\nISEC 532 \nCommittee Members:\nProf. Pau Closas (Advisor)\nProf. Deniz Erdogmus\nProf. Lili Su \nAbstract:\nThe necessity for distributed data fusion arises from the increasing demand to integrate diverse and voluminous data sources\, especially in applications where large numbers of users are collaborating to perform inference and learning tasks. This integration is crucial when data is available in a distributed manner or originates from various sensor types\, aiming to deduce specific quantities of interest accurately. Moreover\, the importance of privacy cannot be overstated\, particularly in scenarios where sensitive information\, such as location data\, is involved. Federated learning emerges as a pivotal solution in this context\, enabling model training on local datasets without the need to exchange the data itself\, thus preserving user privacy. However\, the deployment of these technologies encounters significant challenges\, including the multiple counting problem in data fusion\, where data may be redundantly used across different estimations without user awareness\, and the non-IID problem in federated learning\, where the non-identically distributed nature of data across clients can severely hamper the model’s performance. \nTo address these challenges\, this dissertation explores the intersection of data fusion\, federated learning\, and Bayesian methods\, with a focus on applied problems in indoor localization\, satellite-based navigation\, and image processing that spans both theoretical analysis and practical application. In the realm of data fusion\, we delve into the Bayesian framework to offer a solution that not only facilitates the optimal integration of sensor data with prior knowledge but also navigates the intricacies of feature fusion effectively. This approach mitigates the multiple counting issue by ensuring that the fusion of local estimates accounts for the overuse of prior knowledge. In tackling the problems inherent to federated learning\, particularly the non-IID issue\, we introduce novel frameworks and algorithms designed to enhance model training and performance in a privacy-preserving manner. We explore personalized and clustered federated learning as methods to customize the learning process to individual client characteristics and to group clients with similar data traits\, respectively. A number of practical problems are explored using those federated methodologies\, including indoor fingerprinting\, jamming interference classification\, or image classification tasks. Noticeably\, this thesis proposes a novel Bayesian clustered federated learning framework that generalizes existing clustered federated learning schemes by leveraging Bayesian data association modeling. By implementing a Bayesian perspective within these frameworks\, the dissertation proposes practical algorithms that achieve a balance between performance and computational efficiency\, ultimately advancing the application of distributed data fusion and federated learning in privacy-sensitive fields.
URL:https://ece.northeastern.edu/event/peng-wu-phd-dissertation-defense/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20240412T120000
DTEND;TZID=America/New_York:20240412T130000
DTSTAMP:20260414T035601
CREATED:20240403T222325Z
LAST-MODIFIED:20240403T222325Z
UID:6861-1712923200-1712926800@ece.northeastern.edu
SUMMARY:Baolin Li PhD Dissertation Defense
DESCRIPTION:Announcing:\nPhD Dissertation Defense \nName:\nBaolin Li \nTitle:\nMaking Machine Learning on HPC Systems Cost-Effective and Carbon-Friendly \nDate:\n4/12/2024 \nTime:\n12:00:00 PM
URL:https://ece.northeastern.edu/event/baolin-li-phd-dissertation-defense/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20240408T153000
DTEND;TZID=America/New_York:20240408T170000
DTSTAMP:20260414T035601
CREATED:20240403T222632Z
LAST-MODIFIED:20240403T222632Z
UID:6865-1712590200-1712595600@ece.northeastern.edu
SUMMARY:Jinkun Zhang PhD Dissertation Defense
DESCRIPTION:Announcing:\nPhD Dissertation Defense \nName:\nJinkun Zhang \nTitle:\nLow-latency Forwarding\, Caching and Computation Placement in Data-centric Networks \nDate:\n4/8/2024 \nTime:\n3:30:00 PM \nLocation:\nEXP-459\, \nCommittee Members:\nProf. Edmund Yeh (Advisor)\nProf. Stratis Ioannidis\nProf. Kaushik Chowdhury \nAbstract:\nWith the exponential growth of data- and computation-intensive network applications\, such as real-time augmented reality/virtual reality rendering and large-scale language model training\, traditional cloud computing frameworks exhibit inherent limitations. To address these challenges\, dispersed computing has emerged as a promising next-generation networking paradigm. By enabling geographically distributed nodes with heterogeneous computation capabilities to collaborate\, dispersed computing overcomes the bottlenecks of traditional cloud computing and facilitates in-network computation tasks\, including the training of large models. In data-centric networks\, communication and computation are resolved around data names instead of host addresses. The deployment of network caches\, by enabling data reuse\, offers substantial benefits for data-centric networks. For instance\, consider a scenario where multiple machine learning applications seek to train different models simultaneously. This application could (partially) share data samples and/or computational results. Optimal caching of data and/or results can significantly reduce the overall training cost\, compared to each application independently gathering and transmitting data. \nThis dissertation aims to minimize average user delay in a general cache-enabled computing network. We introduce a low-latency framework that jointly optimizes packet forwarding\, storage deployment\, and computation placement. The proposed framework effectively supports data-intensive and latency-sensitive computation applications in data-centric computing networks with heterogeneous communication\, storage\, and computation capabilities. To minimize user latency in congestible networks\, we model delays caused by link transmissions and CPU computations using traffic-dependent nonlinear functions. We consider a series of related network resource allocation problems in a unified network model. \nWe first investigate the joint forwarding and computation placement problem\, then the joint forwarding and elastic caching problem. Despite the non-convexity of the former subproblem\, we provide a set of sufficient optimality conditions that lead to a distributed algorithm with polynomial-time convergence to the global optimum. For the latter subproblem\, we demonstrate its NP-hardness and non-submodularity\, even after continuous relaxation. We propose a set of conditions that provide a finite bound from the optimum. To the best of our knowledge\, our method represents the first analytical progress in addressing the joint caching and forwarding problem with arbitrary topology and non-linear costs. Upon solving the above two subproblems\, we formally propose the low-latency joint forwarding\, caching\, and computation placement framework. We formulate the mixed-integer NP-hard total cost minimization problem jointly over forwarding\, caching\, and computation offloading variables. Developing on the established result for both subproblems\, we propose two methods\, each with an analytical guarantee. The first method achieves a 1/2 approximation guarantee by exploiting the “submodular + concave” structure of the problem\, leading to an offline distributed algorithm. In real scenarios\, however\, request patterns and network status are not known prior and can be time-varying. To this end\, our second method leads to an online adaptive algorithm exploiting its “convex + geodesic-convex” nature\, with a proven bounded gap from the optimum. \nThe proposed solutions are followed by a few extension problems. Specifically\, we generalize the computation from “single-step” to “service chain” applications. We also generalize the solution to incorporate congestion control by considering an “extended graph”. Furthermore\, several network resource allocation optimization problems related to data-centric networking are introduced\, expanding the scope of this dissertation. For example\, we investigate joint caching and transmission power allocation in wireless heterogeneous networks\, where the total transmission energy is minimized subject to constraints for SINR lower bounds\, cache capacities\, and total power budget at each node. We also study the optimal multi-commodity pricing with finite menu length\, where novel asymptotic bounds on quantization errors are devised.
URL:https://ece.northeastern.edu/event/jinkun-zhang-phd-dissertation-defense/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20240404T153000
DTEND;TZID=America/New_York:20240404T170000
DTSTAMP:20260414T035601
CREATED:20240319T181622Z
LAST-MODIFIED:20240319T181622Z
UID:6843-1712244600-1712250000@ece.northeastern.edu
SUMMARY:Nicolas Bohm Agostini PhD Proposal Review
DESCRIPTION:Announcing:\nPhD Proposal Review \nName:\nNicolas Bohm Agostini \nTitle:\nHardware/Software Codesign and Compiler Techniques for Efficient Hardware Acceleration of Dense Linear Algebra Kernels and Machine Learning Applications \nDate:\n4/4/2024 \nTime:\n3:30:00 PM \nLocation: Zoom \nCommittee Members:\nProf. David Kaeli (Advisor)\nProf. Gunar Schirner\nProf. José Luis Abellán (University of Murcia)\nAntonino Tumeo (PNNL) \nAbstract:\nToday’s linear algebra and machine learning applications (ML) continue to grow in size and complexity\, placing rapidly increasing demands on the underlying hardware and software systems. To address these issues\, hardware designers have proposed using custom accelerators explicitly designed for accelerating these demanding workloads. What needs to be improved is the ability to perform efficient hardware/software (HW/SW) co-design in order to reap the full benefits from these platforms. This thesis presents an integrated solution to facilitate HW/SW accelerator design. We also address issues in accelerator deployment\, enabling rapid prototyping\, integrated benchmarking\, and comprehensive performance analysis of custom accelerators. \nIn this thesis\, we demonstrate the value of a lightweight system modeling library integrated into the build/execution environment\, leveraging TensorFlow~Lite for deployment. We also explore efficient design space exploration of different classes of accelerators while considering the impact of parameters. Secondly\, we employ the Multi-Level Intermediate Representation (MLIR) compiler framework to automatically partition host code from accelerator code\, pre-optimizing the latter for improved high-level synthesis designs and high-quality accelerated kernels. Lastly\, we propose compiler extensions to automate the generation and optimization of communication between the host CPU and AXI-based accelerators. \nWe present novel solutions that enable more efficient and effective design space exploration\, optimization\, and deployment of custom accelerators. The utility of these approaches is demonstrated through experiments with specific accelerator designs and key linear algebra and ML workloads. Most importantly\, these solutions empower high-level language users\, such as domain scientists\, to participate in the design of new accelerator features.
URL:https://ece.northeastern.edu/event/nicolas-bohm-agostini-phd-proposal-review/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20240404T130000
DTEND;TZID=America/New_York:20240404T140000
DTSTAMP:20260414T035601
CREATED:20240403T222208Z
LAST-MODIFIED:20240403T222208Z
UID:6859-1712235600-1712239200@ece.northeastern.edu
SUMMARY:Anu Jagannath PhD Dissertation Defense
DESCRIPTION:Announcing:\nPhD Dissertation Defense \nName:\nAnu Jagannath \nTitle:\nDeep Learning at the Edge for Future G Networks: RF Signal Intelligence for Comprehensive Spectrum Awareness \nDate:\n4/4/2024 \nTime:\n1:00:00 PM \nCommittee Members:\nProf. Tommaso Melodia (Advisor)\nProf. Kaushik Chowdhury\nProf. Yanzhi Wang \nAbstract:\nFuture communication networks must address the scarce spectrum to accommodate extensive growth of heterogeneous wireless devices. Efforts are underway to address spectrum coexistence\, enhance spectrum awareness\, and bolster authentication schemes. Wireless signal recognition is becoming increasingly more significant for spectrum monitoring\, spectrum management\, secure communications\, among others. Consequently\, comprehensive spectrum awareness at the edge has the potential to serve as a key enabler for the emerging beyond 5G (fifth generation) networks. State-of-the-art studies in this domain have (i) only focused on a single task – modulation or signal (protocol) classification or radio frequency fingerprinting – which in many cases is insufficient information for a system to act on\, (ii) consider either radar or communication waveforms (homogeneous waveform category)\, and (iii) does not address edge deployment during neural network design phase. In this dissertation\, deep learning is applied to the various signal recognition problems from  a multi-task perspective with an emphasis on edge deployment. To address edge deployment\, various techniques are applied to solve the signal recognition problem under consideration (modulation\, wireless protocol\, emitter fingerprint recognition) to design scalable and computationally efficient framework. While designing the edge deployable architectures\, the generalization capability of the architectures are evaluated under various circumstances to quantify their performance under real-world settings such as emissions from actual emitters (commercial emissions wherever applicable)\, training with a different propagation scenario and testing under a never-before-seen setting. \nThe study was sectioned into different stages where multi-task learning is first applied to solving wireless standard and modulation recognition\, followed by applying deep compression for CBRS radar waveform classification\, next radio frequency fingerprinting for commercial WiFi and Bluetooth emissions were studied utilizing novel multi-task attentional architectures\, and finally the multi-task learning together with deep compression was employed to deploy the architectures in a real-time streaming radio testbed for real-time inferencing of wireless standard and modulation recognition. The feasibility of employing deep compression techniques are carefully evaluated in a real-world deployment setting to quantify the performance from a computational and inference capacity perspective.
URL:https://ece.northeastern.edu/event/anu-jagannath-phd-dissertation-defense/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20240403T153000
DTEND;TZID=America/New_York:20240403T170000
DTSTAMP:20260414T035601
CREATED:20240319T181441Z
LAST-MODIFIED:20240319T181441Z
UID:6841-1712158200-1712163600@ece.northeastern.edu
SUMMARY:Kaustubh Shivdikar PhD Dissertation Defense
DESCRIPTION:Announcing:\nPhD Dissertation Defense \nName:\nKaustubh Shivdikar \nTitle:\nEnabling Accelerators for Graph Computing \nDate:\n4/3/2024 \nTime:\n3:30 PM \nLocation: Zoom \nCommittee Members:\nProf. David Kaeli (Advisor)\nProf. Devesh Tiwari\nProf. Ajay Joshi (Boston University)\nProf. John Kim (KAIST)\nProf. José Luis Abellán (University of Murcia) \nAbstract:\nThe advent of Graph Neural Networks (GNNs) has revolutionized the field of machine learning\, offering a novel paradigm for learning on graph-structured data. Unlike traditional neural networks\, GNNs are capable of capturing complex relationships and dependencies inherent in graph data\, making them particularly suited for a wide range of applications including social network analysis\, molecular chemistry\, and network security. The impact of GNNs in these domains is profound\, enabling more accurate models and predictions\, and thereby contributing significantly to advances in these fields. \nGNNs\, with their unique structure and operation\, present new computational challenges compared to conventional neural networks. This requires comprehensive benchmarking and a thorough characterization of GNNs to obtain insight into their computational requirements and to identify potential performance bottlenecks. In this thesis\, we aim to develop a better understanding of how GNNs interact with the underlying hardware and will leverage this knowledge as we design specialized accelerators and develop new optimizations\, leading to more efficient and faster GNN computations. \nA pivotal component within GNNs is the Sparse General Matrix-Matrix Multiplication (SpGEMM) kernel\, known for its computational intensity and irregular memory access patterns. In this thesis\, we address the challenges posed by SpGEMM by implementing a highly optimized hashing-based SpGEMM kernel tailored for a custom accelerator. This optimization is crucial to enhancing the performance of GNN workloads\, ensuring that the acceleration potential of custom hardware is fully realized. \nSynthesizing these insights and optimizations\, we design state-of-the-art hardware accel-erators capable of efficiently handling various GNN workloads. Our accelerator architectures are built on our characterization of GNN computational demands\, providing clear motivation for our approaches. Furthermore\, we extend our exploration to emerging GNN workloads in the domain of graph neural networks. This exploration into novel models underlines our comprehensive approach\, as we strive to enable accelerators that are not just performant\, but also versatile\, able to adapt to the evolving landscape of graph computing.
URL:https://ece.northeastern.edu/event/kaustubh-shivdikar-phd-dissertation-defense/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20240403T110000
DTEND;TZID=America/New_York:20240403T123000
DTSTAMP:20260414T035601
CREATED:20240403T222458Z
LAST-MODIFIED:20240403T222458Z
UID:6863-1712142000-1712147400@ece.northeastern.edu
SUMMARY:Batool Salehihikouei PhD Dissertation Defense
DESCRIPTION:Announcing:\nPhD Dissertation Defense \nName:\nBatool Salehihikouei \nTitle:\nLeveraging Deep Learning on Multimodal Sensor Data for Wireless Communication: From mmWave Beamforming to Digital Twins \nDate:\n4/3/2024 \nTime:\n11:00:00 AM \nLocation: EXP-601A \nCommittee Members:\nProf. Kaushik Chowdhury (Advisor)\nProf. Hanumant Singh\nProf. Josep Jornet\nDr. Mark Eisen \nAbstract:\nWith the widespread Internet of Things (IoT) devices\, a wide variety of sensors are now present in different environments. For example\, self-driving vehicles and automated warehouses depend on sensor information for navigation and management of the robots\, respectively. In this dissertation\, we present methods\, where these sensors are re-purposed to assist network management in wireless communication\, especially when classic approaches fall short to provide the required quality of service (QoS). This thesis presents data-driven and AI-based methods\, where the multimodal sensor information is used for two applications: (i) beamforming at the mmWave band and (ii) joint optimization of the navigation and network management in warehouse environments. In the first part\, we study multimodal beamforming methods for mmWave vehicular networks. First\, we present deep learning fusion algorithms\, where the inputs from a multitude of sensor modalities such as GPS (Global Positioning System)\, camera\, and LiDAR (Light Detection and Ranging) are combined towards predicting the optimum beam at the mmWave band. We prove that fusing the multimodal sensor data improves the prediction accuracy\, compared to using single modalities. Second\, we study the trade-off between the accuracy and cost of different learning strategies and demonstrate that federated learning is the most successful learning strategy\, with respect to the communication overhead. Third\, we propose algorithms to further optimize the communication overhead by incorporating a pruning strategy tailored to the disturbed nature of the federated learning systems. Fourth\, we propose a modality-agnostic deep learning paradigm that operates on any possible combination of sensor modalities. In part two\, we propose using digital twins to overcome the challenges of scarcity of data and close-world assumption in deep learning algorithms. A digital twin is a replica of a real world entity\, which is typically used for studying the impact of any configuration settings in a safe\, digital environment. In this dissertation\, we propose a framework that operates by harmonic usage of the DL models and running emulations in the twin. Moreover\, we use digital twins to generate training labels and fine-tune the models for unseen scenarios. Finally\, we study a robotic industrial setting\, where the path planning policy is continuously updated by monitoring the dynamics of the real world\, constructing the digital twin\, and updating the policy. The constructed twin captures the features of both physical and RF environments in the digital world and includes a reinforcement learning algorithm that jointly optimizes navigation and network resource management.
URL:https://ece.northeastern.edu/event/batool-salehihikouei-phd-dissertation-defense/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20240401T103000
DTEND;TZID=America/New_York:20240401T113000
DTSTAMP:20260414T035601
CREATED:20240319T180923Z
LAST-MODIFIED:20240319T180923Z
UID:6835-1711967400-1711971000@ece.northeastern.edu
SUMMARY:Reza Vafaee PhD Proposal Review
DESCRIPTION:Announcing:\nPhD Proposal Review \nName:\nReza Vafaee \nTitle:\nEfficient Algorithms for Sparse Sensor Scheduling in Large-Scale Dynamical Systems with Performance Guarantees \nDate:\n4/1/2024 \nTime:\n10:30:00 AM \nLocation: Zoom \nCommittee Members:\nProf. Milad Siami (Advisor)\nProf. Eduardo Sontag\nProf. Laurent Lessard\nProf. Alex Olshevsky (Boston University) \nAbstract:\nThis research proposal introduces innovative frameworks for sparse sensor scheduling in large-scale dynamical networks. The first framework addresses sensor scheduling in discrete-time linear time-invariant dynamical networks\, presenting a novel learning-based rounding method to convert weighted sensor schedules into sparse\, unweighted schedules while maintaining comparable observability performance. The second framework extends the approach to dynamically select sensors for linear time-varying systems\, utilizing an online sparse sensor scheduling framework with randomized algorithms to approximate fully-sensed systems with a constant average number of active sensors at each time step. Finally\, a myopic approach within a Kalman filtering framework is adopted in the third framework\, addressing non-submodular sensor scheduling in large-scale linear time-varying dynamics. A simple greedy algorithm is employed\, providing approximation bounds based on submodularity and curvature concepts. Simulation results validate the theoretical foundations and demonstrate the proposed approach’s superiority over existing methods.
URL:https://ece.northeastern.edu/event/reza-vafaee-phd-proposal-review/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20240329T100000
DTEND;TZID=America/New_York:20240329T120000
DTSTAMP:20260414T035601
CREATED:20240319T181258Z
LAST-MODIFIED:20240319T181258Z
UID:6839-1711706400-1711713600@ece.northeastern.edu
SUMMARY:Matthew Wallace MS Thesis Defense
DESCRIPTION:Announcing:\nMS Thesis Defense \nName:\nMatthew Wallace \nTitle:\nModel Predictive Planning \nDate:\n3/29/2024 \nTime:\n10:00:00 AM \nLocation:\nRoom: HS 204.  Link: Teams \nCommittee Members:\nProf. Laurent Lessard (Advisor)\nProf. Michael Everett\nProf. Derya Aksaray \nAbstract:\nThis thesis presents Model Predictive Planning (MPP)\, a trajectory planner for low-agility vehicles such as a fixed-wing aircraft to navigate obstacle-laden environments.  MPP consists of (1) a multi-path planning procedure that identifies candidate paths\, (2) a raytracing procedure that generates linear constraints around these paths that enforce obstacle avoidance\, and (3) a convex quadratic program that finds a feasible trajectory within these constraint if one exists. Low-agility aircraft cannot track arbitrary paths\, so refining a given path into a trajectory that respects the vehicle’s limited maneuverability and avoids obstacles often leads to an infeasible optimization problem. The critical feature of MPP is that it efficiently considers multiple candidate paths during the refinement process\, thereby greatly increasing the chance of finding a feasible and trackable trajectory. I begin by presenting a background on path planning\, trajectory optimization\, and Model Predictive Control.  This is followed by a presentation of the MPP algorithm.  Finally\, I demonstrate the effectiveness of MPP on both a longitudinal and 3D aircraft model.
URL:https://ece.northeastern.edu/event/matthew-wallace-ms-thesis-defense/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20240328T110000
DTEND;TZID=America/New_York:20240328T120000
DTSTAMP:20260414T035601
CREATED:20240306T200314Z
LAST-MODIFIED:20240321T230531Z
UID:6816-1711623600-1711627200@ece.northeastern.edu
SUMMARY:Huan Wang PhD Dissertation Defense
DESCRIPTION:Announcing:\nPhD Dissertation Defense \nName:\nHuan Wang \nTitle:\nTowards Efficient Deep Learning in Computer Vision via Network Sparsity and Distillation \nDate:\n3/28/2024 \nTime:\n11:00:00 AM \nZoom \nCommittee Members:\nProf. Yun Fu (Advisor)\nProf. Octavia Camps\nProf. Zhiqiang Tao \nAbstract:\nAI\, empowered by deep learning\, has been profoundly transforming the world. However\, the excessive size of these models remains a central obstacle that limits their broader utility. Modern neural networks commonly consist of millions of parameters\, with foundation models extending to billions. The rapid expansion in model size introduces many challenges including training cost\, sluggish inference speed\, excessive energy consumption\, and negative environmental implications such as increased CO2 emissions. \nAddressing these challenges necessitates the adoption of efficient deep learning. The dissertation focuses on two overarching approaches\, network pruning and knowledge distillation\, to enhance the efficiency of deep learning models in the context of computer vision. Network pruning focuses on eliminating redundant parameters in a model while preserving the performance. Knowledge distillation aims to enhance the performance of the target model\, referred to as the “student\,” by leveraging guidance from a stronger model\, known as the “teacher”. This approach leads to performance improvements in the target model without reducing its size. \nIn this defense presentation\, I will start with the background and major challenges of leveraging these techniques to improve the efficiency of deep neural networks. Then\, I shall present the proposed solutions for various vision tasks\, including image classification\, single-image super-resolution\, novel view synthesis / neural rendering / NeRF / NeLF\, text-to-image generation / diffusion models\, and photorealistic head avatars. Extensive results and analyses will justify the efficacy of the proposed approaches\, demonstrating that pruning and distillation make a generic and complete framework for efficient deep learning in various domains. Finally\, a comprehensive summary (with takeaways) and outlook of the future work will conclude the presentation.
URL:https://ece.northeastern.edu/event/human-wang-phd-dissertation-defense/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20240321T120000
DTEND;TZID=America/New_York:20240321T133000
DTSTAMP:20260414T035601
CREATED:20240319T181109Z
LAST-MODIFIED:20240319T181109Z
UID:6837-1711022400-1711027800@ece.northeastern.edu
SUMMARY:Julian Gutierrez PhD Dissertation Defense
DESCRIPTION:Announcing:\nPhD Dissertation Defense \nName:\nJulian Gutierrez \nTitle:\nTowards Real-Time Safe Flight Paths for Urban Air Mobility \nDate:\n3/21/2024 \nTime:\n12:00:00 PM \nLocation: Zoom \nCommittee Members:\nProf. David Kaeli (Advisor)\nProf. Pau Closas\nDr. Evan Dill (NASA)\nDr. Natasha Neogi (NASA) \nAbstract:\nThe emergence and development of advanced technologies and vehicle types have created a growing demand for introducing new forms of flight operations. These new and increasingly complex operational paradigms\, such as Advanced and Urban Air Mobility (AAM/UAM)\, present regulatory authorities and the aviation community with the challenge of finding methods to integrate these emerging operations without significant additional risk to pedestrians and infrastructure. Predictive and autonomous risk mitigation capabilities become critical to meet this challenge. However\, urban environments experience effects that are computationally expensive to model\, limiting conventional aviation concepts\, policy\, and risk prediction tools from being effectively translated into this space. With the emergence of High-Performance Computing (HPC) ecosystems in the last two decades\, we can use these software and hardware capabilities to help bridge the gap between real-time predictive responses and modeling accuracy. \nIn this dissertation we first present a simulation framework to estimate the quality of Global Navigation Satellite System (GNSS) performance for autonomous aircraft in urban environments. We propose a new algorithm designed for HPC to accelerate modeling the characteristic effects of dense urban canyons on GNSS\, allowing the extension of established GNSS integrity techniques into urban navigation. Additionally\, we provide a thorough validation of the simulator\, which proves high-accuracy modeling when compared to sensors in the real world. Second\, we use this simulation framework to provide situational awareness when processing the raw output of a GNSS sensor. This effort focuses on multipath mitigation\, which reduces the error in the estimated position solution. Third\, we use this simulation framework as the input into a new 4D path-planning algorithm based on an adaptation of the Bellman-Ford algorithm. HPC techniques are employed to accelerate the algorithm to produce flight paths that minimize exposure to GNSS risks. We evaluate the computational cost of satellite availability fluctuations by prioritizing events when satellite availability changes as triggers for these updates.
URL:https://ece.northeastern.edu/event/julian-gutierrez-phd-dissertation-defense/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20231220T140000
DTEND;TZID=America/New_York:20231220T150000
DTSTAMP:20260414T035601
CREATED:20231215T231627Z
LAST-MODIFIED:20231215T231627Z
UID:6662-1703080800-1703084400@ece.northeastern.edu
SUMMARY:Xiang Zhang PhD Proposal Review
DESCRIPTION:Title:Confidentiality and Privacy Preserving:  Intertwining Deep Learning and  Side-channel Analysis \nMeeting ID: 976 4324 8925 Passcode: 779251 \nCommittee Members:\nProf. Yunsi Fei (Advisor)\nProf. Adam Ding\nProf. Lili Su \nAbstract:\nIn the past decade\, deep learning-empowered technologies have significantly permeated our daily lives\, revolutionizing diverse application domains with superb performance.  In hardware security\, deep learning has been employed for power or electromagnetic side-channel analysis (SCA) and protection\, and the security of deep learning implementations starts gaining traction. \nThis dissertation delves into the intertwining deep learning techniques and side-channel analysis.  It addresses two critical questions: how to extend deep learning to other types of SCAs; what confidentiality and privacy vulnerabilities deep learning models have. \nOur research work first explores deep learning-assisted cache side-channel attacks and introduces innovative countermeasures grounded in the principles of adversarial samples against deep learning. We first design a novel high-frequency cache monitor\,  which runs concurrent to the victim execution and collects run-time timing traces\, while previous cache monitors are only able to collect timing samples. Such timing traces facilitate follow-on non-profiled Differential Deep Learning Analysis (DDLA) for secret retrieval. We also propose a novel countermeasure against the new DDLA\, leveraging the concept of adversarial examples\, which deliberately introduces obfuscation operations in the victim program so as to generate ‘adversarial’ timing traces and therefore circumvent the follow-on DDLA. \nThe second part of the dissertation addresses the vulnerability of deep neural network (DNN) implementations and presents novel methodologies for enhancing user privacy. It introduces a technique for extracting deep learning models through software-based power side channels. By manipulating model inputs and leveraging the on-chip Intel Running Average Power Limit (RAPL) sensors reporting\, the entire model parameters can be extracted when the model inference is executed on modern processors. To protect both the model confidentiality and the input privacy\, this dissertation proposes to obfuscate the model inputs while preserving the end-to-end functionality. It introduces an encoder to transform the inputs before feeding the DNN model\, and appends a decoder after the model outputs to recover the intended results. The approach\, compared to traditional encryption or masking techniques\, is more efficient and can effectively protect both user privacy and model confidentiality. \nThe overall goal of the dissertation is to further investigate the power of deep learning in SCA and countermeasure and safeguard secure DNN implementations.
URL:https://ece.northeastern.edu/event/xiang-zhang-phd-proposal-review/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20231212T110000
DTEND;TZID=America/New_York:20231212T170000
DTSTAMP:20260414T035601
CREATED:20231208T195540Z
LAST-MODIFIED:20231208T195540Z
UID:6658-1702378800-1702400400@ece.northeastern.edu
SUMMARY:Deepak Prabhala MS Thesis Defense
DESCRIPTION:Title: “Smart Microwave Devices with Programable Printed Circuit Board (PPCB): Design with Liquid Crystal Elastomer Polymers in Transmission Lines and Circulators” \nCommittee Members:\n1) Professor Nian X. Sun (Advisor)\n2) Professor Marvin Onabajo\n3) Professor Yongmin Liu \nAbstract:\nThis study explores the innovative application of liquid crystal elastomer (LCE) polymers in the design and implementation of microwave transmission lines and circulators. Liquid crystal elastomers\, known for their unique combination of liquid crystalline and elastomeric properties\, offer a unique approach to developing flexible and tunable microwave devices. The research focuses on a thorough study of the electro-mechanical properties of LCEs to achieve novel functionalities in the design of transmission lines and circulators for microwave communication systems in HFSS simulations. The first part of the study delves into the characterization of the dielectric and mechanical properties of the chosen LCE polymer. Subsequently\, the design and fabrication of a flexible and tunable transmission line using LCE are discussed. The LCE-based transmission line aims to measure the insertion loss and return loss with different widths\, lengths\, and thicknesses of the LCE polymer. The study investigates the impact of temperature on the transmission line’s performance\, offering insights into potential applications in reconfigurable microwave systems. The second phase of the research explores the utilization of LCE in the development of a microwave circulator\, a vital component in microwave communication networks. The circulator design incorporates the unique properties of LCE by using a stepped dielectric variation approach for broadband isolation. This innovation holds promise for enhancing the efficiency and adaptability of microwave systems in communication and radar applications. The findings of this research contribute to offering a pathway for integrating liquid crystal elastomers into flexible and reconfigurable microwave devices. This thesis aims to advance the understanding of smart microwave devices and inspire further exploration into the application of liquid crystal elastomer polymers in cutting-edge technologies.
URL:https://ece.northeastern.edu/event/deepak-prabhala-ms-thesis-defense/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20231212T090000
DTEND;TZID=America/New_York:20231212T100000
DTSTAMP:20260414T035601
CREATED:20231208T195140Z
LAST-MODIFIED:20231208T195212Z
UID:6654-1702371600-1702375200@ece.northeastern.edu
SUMMARY:Durga Suresh PhD Proposal Review
DESCRIPTION:Title: Network Security Management and Threat Mitigation in the Open Cloud\n \nCommittee Members:\nProf. Miriam Leeser (Advisor)\nProf. Michael Zink\nProf. Xiaolin Xu \nAbstract:\nCloud computing and advanced cyberinfrastructures are increasingly vital to the functioning of Internet systems. Every day\, more devices are added to the cloud\, to provide greater resource utilization\, availability\, and scalability. Due to the expanding reliance on cloud computing\, securing the cloud is paramount. Tackling the issue of securing the cloud is crucial not only for preserving the functionality and reliability of cloud-based systems but also for protecting the critical data and services that depend on these platforms. \nCloud computing models include public clouds\, private clouds\, community clouds\, and hybrid clouds. Private\, community\, and hybrid clouds provide security\, but with an important trade-off; namely\, user access restriction in the cloud. The proposed research uses the Open Cloud Testbed (OCT) which is part of the National Science Foundation’s (NSF) Computer and Information Systems Engineering(CISE) Community Research Infrastructure(CRI) program. OCT is an example of a public cloud that allows users two things: 1) an isolated set of nodes to perform experiments with bare metal access\, which can potentially lead to security issues\, and 2) the ability to test out the solutions for both using the cloud and adding security to it. The proposed research aims to target a system like the OCT\, specifically targeting a public cloud environment. \nThis system will be designed to allow access to the switch\, enabling control and management of traffic within the cloud network. This research aims to mitigate network security threats in the public cloud network. The aim of this research is multifold. First\, we identify and classify the behavior of users in the cloud. We then provide an approach to creating a network security management policy that will deal with 1)detecting network intruders that scan the cloud network and remove their access to the network\, and  2) managing heavy hitters that can cause Denial of Service (DOS) and Distributed Denial of Service (DDOS) attacks in the cloud network by using the heavy hitter detection system and prevent them from putting more traffic on the network. Both network intruder detection and heavy hitter management systems use Access Control Lists (ACL)as a means to prevent the user from putting traffic on the cloud network. Lastly\, we perform experiments to handle these threats and measure the success of the experimental setup concerning network attacks. The proposed approach will ensure network security by creating a framework for network security management policy to minimize threats in the cloud network and other resources directly attached to the network. The proposed research aims to enhance cybersecurity by employing network intruder detection techniques to identify potential threats\, implementing heavy hitter management to mitigate threats effectively\, and developing and enforcing a network security management policy to prevent future threats.
URL:https://ece.northeastern.edu/event/durga-suresh-phd-proposal-review/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20231208T120000
DTEND;TZID=America/New_York:20231208T123000
DTSTAMP:20260414T035601
CREATED:20231208T195358Z
LAST-MODIFIED:20231208T195358Z
UID:6656-1702036800-1702038600@ece.northeastern.edu
SUMMARY:Ate Darabi PhD Proposal Review
DESCRIPTION:Title:\nComplex Delayed Networks and Their Application in Epidemic Analysis: Modeling\, Analysis\, and Strategic Management \nCommittee Members:\nProf. Milad Siami (Advisor)\nProf. Bahram Shafai\nProf. Rozhin Hajian \nAbstract:\nIn the face of crowd-related disasters like pandemics and mass attacks\, the complex dynamics of human interactions demand comprehensive modeling approaches. This proposal adopts a network-based perspective\, leveraging the delayed Susceptible-Infected-Susceptible (SIS) model for epidemics and the Predator-Swarm-Guide (PSG) model for crowd movement\, to gain insights into the dynamics of these critical situations. \nIn epidemic networks\, time delays and uncertainties can significantly change the epidemic behavior and result in successive echoing waves of the spread between various population clusters. We examine these effects on linear SIS dynamics\, evaluating network stability and performance loss. We prove that network performance loss is correlated with the structure of the underlying graph\, intrinsic time delays\, epidemic characteristics\, and external shocks. This performance measure is then used to develop an optimal traffic restriction algorithm for network performance enhancement\, resulting in reduced infection in the metapopulation.   An epidemic-based centrality index is also proposed to evaluate the impact of every subpopulation on network performance\, and its asymptotic behavior is investigated. This index converges to local or eigenvector centralities under specific parameters. Moreover\, given that epidemic-based centrality depends on the epidemic properties of the disease\, it may yield distinct node rankings as the disease characteristics slowly change over time or as different types of infections spread. This unique characteristic of epidemic-based centrality enables it to adjust to various epidemic features. The derived centrality index is then adopted to improve the network robustness against external shocks on the epidemic network. \nThe PSG model addresses mass attack scenarios\, considering individuals’ efforts to evade adversaries and seek guidance. Environmental factors like impermeable walls and psychological elements are incorporated into this model. The preliminary results highlight the role of coordinated cooperation in minimizing casualties. The objective is to reduce casualties through a hybrid motion optimization approach for individuals and the guiding agent.
URL:https://ece.northeastern.edu/event/ate-darabi-phd-proposal-review/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20231207T140000
DTEND;TZID=America/New_York:20231207T150000
DTSTAMP:20260414T035601
CREATED:20231205T000346Z
LAST-MODIFIED:20231205T000346Z
UID:6643-1701957600-1701961200@ece.northeastern.edu
SUMMARY:Yisi Liu MS Thesis Defense
DESCRIPTION:Title: Experimental research on the Nonlinear Magnetoelectric Effect of the VLF ME Antennas \nCommittee Members:\nProf. Nian Sun (Advisor)\nProf. Yongmin Liu\nProf. Xufeng Zhang \nAbstract:\nMagnetoelectric (ME) coupling effects in ferromagnetic and piezoelectric composites involve the control of electric polarization (P) by applying a magnetic field (H) (direct ME effect)\, or the manipulation of magnetization (M) through an electric field (E) (converse ME effect) . These effects are facilitated by the mechanical deformation in the ferroic phases resulting from the combination of magnetostriction and piezoelectricity. In single-phase materials\, the breakthrough in achieving large ME coefficients has further advanced the development of ME materials and devices. Consequently\, numerous multifunctional ME devices\, such as mechanical antennas\, magnetic sensors\, tunable inductors\, and filters\, have been developed. This thesis has provides a summary and categorization of these devices based on their physical mechanism and type of ME effects. The inclusion of mechanical ME antennas based on piezoelectric/magnetostrictive heterostructures with acoustic actuation reflects the significant interest in this topic. Notably\, a maximum communication distance of 120 m for a very low frequency (VLF) communication system has been achieved using a pair of mechanical ME antennas. Subsequently\, we will focus on introducing and reviewing the materials and devices related to the ME effect\, as well as the application of ME mechanical antennas in very low frequency (VLF) communication systems. \nIn addition to that\, we developed a transmitter with a Metglas/PZT/Metglas structure antenna. Our study focuses on investigating the transmission effects of this antenna when employing direct antenna modulation techniques to enhance data transmission. Through our research\, we have introduced a novel modulation method by modulating the antenna. We observed that this modulation method produces a more stable and stronger signal. \n 
URL:https://ece.northeastern.edu/event/yisi-liu-ms-thesis-defense/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20231207T100000
DTEND;TZID=America/New_York:20231207T110000
DTSTAMP:20260414T035601
CREATED:20231204T235514Z
LAST-MODIFIED:20231204T235514Z
UID:6639-1701943200-1701946800@ece.northeastern.edu
SUMMARY:Mauro Belgiovine PhD Proposal Review
DESCRIPTION:Title: Wireless Intelligence: A Comprehensive Exploration of AI-Driven Solutions in Channel Estimation\, Beam Refinement\, and Protocol Classification for Next Generation Networks \nCommittee Members:\nProf. Kaushik Chowdhury (advisor)\nProf. Stratis Ioannidis\nDr. Chris Dick \nAbstract:\nhis thesis explores the transformative impact of artificial intelligence (AI) on wireless systems through model-driven simulations and real-world datasets\, with a focus on enhancing both local and cellular wireless networks through the deployment of highly customized deep learning solutions that target specific bottlenecks affecting traditional signal processing based communication. \nThe research delves into three key areas that address critical challenges in the current wireless landscape. The first focal point of the investigation involves channel estimation using deep learning techniques to denoise pilots and expedite the accurate estimation of Channel State Information (CSI). By leveraging deep learning methodologies\, the proposed solution aims to enhance the reliability and computation for MIMO and massive MIMO channel estimation\, thereby contributing to improved communication efficiency and reduced errors. The second major topic encompasses the application of reinforcement learning for 5G New Radio (NR) millimeter-wave (mmWave) beam refinement. The study aims to develop a Deep Reinforcement Learning algorithm capable of adjusting beamsteering angles\, starting from a coarse beam scanning procedure and further refining them for higher transmission efficiency. This innovation is expected to substantially decrease traffic overhead while simultaneously enhancing beam steering precision\, thus optimizing the performance of mmWave communication. The third and final area of focus introduces a transformer-based WiFi multi-protocol classifier\, strategically deployed on a DeepWave Air-T edge device\, which is equipped with Module on Chip (MoC) low power CPU-GPU and programmable Software Defined Radio (SDR). This classifier outperforms existing modulation classification models and legacy methods under lower SNR conditions\, leveraging TensorRT’s model compression capabilities to efficiently process extended sequences of raw IQ samples\, ensuring high performance at a low computational cost. The proposed solution addresses the growing demand for efficient and adaptable wireless communication systems\, paving the way for advancements in edge-based processing and intelligent protocol classification. \nThis work seeks to contribute significantly to the ongoing AI revolution in wireless systems by addressing crucial issues in channel estimation\, beam refinement\, and protocol classification. The outcomes of this research hold the potential to redefine the landscape of wireless communication\, offering enhanced performance\, reduced overhead\, and increased adaptability in both local and cellular networks.
URL:https://ece.northeastern.edu/event/mauro-belgiovine-phd-proposal-review/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20231206T150000
DTEND;TZID=America/New_York:20231206T170000
DTSTAMP:20260414T035601
CREATED:20231204T235947Z
LAST-MODIFIED:20231204T235947Z
UID:6641-1701874800-1701882000@ece.northeastern.edu
SUMMARY:Suyash Pradhan MS Thesis Defense
DESCRIPTION:Title: COPILOT: Cooperative Perception using Lidar for Handoffs between Road Side Units \nCommittee Members:\nProf. Kaushik Chowdhury (Advisor)\nProf. Stratis Ioannidis\nProf. Jennifer Dy \nAbstract:\nThis thesis presents COPILOT\, a ML-based approach that allows vehicles requiring ubiquitous high bandwidth connectivity to identify the most suitable road side units (RSUs) through proactive handoffs. By cooperatively exchanging the data obtained from local 3D Lidar point clouds within adjacent vehicles and with coarse knowledge of their relative positions\, COPILOT identifies transient blockages to all candidate RSUs along the path under study. Such cooperative perception is critical for choosing RSUs with highly directional links required for mmWave bands\, which majorly degrade in the absence of LOS. COPILOT proposes three modules that operate in an inter-connected manner: (i) As an alternative to sending raw Lidar point clouds\, it extracts and transmits low-dimensional intermediate features to lower the overhead of inter-vehicle messaging; (ii) It utilizes an attention-mechanism to place greater emphasis on data collected from specific vehicles\, as opposed to nearest neighbor and distance-based selection schemes\, and (iii) it experimentally validates the outcomes using an outdoor testbed composed of an autonomous car and Talon AD7200 60GHz routers emulating the RSUs\, accompanied by the public release of the datasets. Results reveal COPILOT yields upto 69.8% and 20.42% improvement in latency and throughput compared to traditional reactive handoffs for mmWave networks\, respectively
URL:https://ece.northeastern.edu/event/suyash-pradhan-ms-thesis-defense/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20231204T103000
DTEND;TZID=America/New_York:20231204T113000
DTSTAMP:20260414T035601
CREATED:20231127T213905Z
LAST-MODIFIED:20231127T213905Z
UID:6620-1701685800-1701689400@ece.northeastern.edu
SUMMARY:Cheng Gongye PhD Dissertation Defense
DESCRIPTION:Title:\nHardware Security Vulnerabilities in Deep Neural Networks and Mitigations \nDate:\n12/4/2023 \nTime:\n10:30:00 AM \nCommittee Members:\nProf. Yunsi Fei (Advisor)\nProf. Aidong Ding\nProf. Xue Lin\nProf. Xiaolin Xu \nAbstract:\nIn the past decade\, Deep Neural Networks (DNNs) have become pivotal in numerous fields\, including security-sensitive autonomous driving and privacy-critical medical diagnosis. This Ph.D. dissertation delves into the hardware security of DNNs\, discovering their vulnerabilities to fault and side-channel attacks and exploring novel countermeasures essential for their safe deployment in critical applications. \nFault attacks disrupt computation or inject faults into parameters\, compromising the integrity of targeted applications. This dissertation demonstrates a power-glitching fault injection attack on FPGA-based DNN accelerators\, common in cloud environments\, which exploits vulnerabilities in the shared power distribution network and results in model misclassification. In response to these threats\, we introduce a novel\, lightweight defense mechanism to protect DNN parameters from adversarial bit-flip attacks. The proposed framework incorporates a dynamic channel-shuffling obfuscation scheme coupled with a logits-based model integrity monitor. The approach effectively safeguards various DNN models against bit-flip attacks\, without necessitating retraining or structural changes to the models. Furthermore\, our research expands the scope of fault analysis beyond just the parameters of DNN models. We thoroughly examine the entire implementation of commercial products\, defying the prevailing assumption that quantized DNNs are inherently resistant to bit-flips. \nSide-channel attacks exploit information leakage of system implementations\, such as power consumption and electromagnetic emanations\, to reveal system secrets and therefore compromise confidentiality. This dissertation makes significant contributions to side-channel assisted model extraction of DNNs. We present a floating-point timing side-channel attack on x86 CPUs that reverse-engineers DNN model parameters in software implementations. For hardware accelerators\, we target the state-of-the-art AMD-Xilinx deep-learning processor unit (DPU)\, a reconfigurable engine dedicated to convolutional neural networks (CNNs) and representing the most complex commercial FPGA accelerator with encrypted IPs. Our work demonstrates that electromagnetic analysis can be leveraged to recover the data flow and scheduling of the DNN accelerators\, facilitating follow-on architecture and parameter extraction attacks. To mitigate EM side-channel model extraction attacks\, we introduce a novel defense mechanism that devises a random importance-aware activation mask on input pixels to disrupt the operation alignment on EM traces\, with minimal performance and efficiency impacts. \nOverall\, this dissertation significantly deepens the understanding of hardware security of DNN models. It makes important contributions in discovering novel and critical vulnerabilities of DNN inference pertaining to system implementations\, and proposing effective and practical solutions for securing DNNs in mission-critical environments. The research work marks a substantial step forward in the development of resilient and secure AI systems.
URL:https://ece.northeastern.edu/event/cheng-gongye-phd-dissertation-defense/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20231129T163000
DTEND;TZID=America/New_York:20231129T170000
DTSTAMP:20260414T035601
CREATED:20231127T214136Z
LAST-MODIFIED:20231127T214136Z
UID:6622-1701275400-1701277200@ece.northeastern.edu
SUMMARY:Aria Masoomi PhD Proposal Review
DESCRIPTION:Title:\nMaking Deep Neural Network Transparent \nDate:\n11/29/2023 \nTime:\n4:30:00 PM \nCommittee Members:-\nProf. Jennifer Dy\nProf. Eduardo Sontag\nProf. Mario Sznaier\nProf. Peter Castaldi \nAbstract:\nAs machine learning algorithms are deployed ubiquitously to a variety of domains\, it is imperative to make these often black-box models transparent. The ability to interpret and comprehend the reasoning behind machine learning models plays a pivotal role in increasing user trust. It not only offers insights into how a model functions but also opens avenues for model enhancements. \nThis research delves into the realm of interpretability\, focusing on the dichotomy between ‘intrinsic’ and ‘post hoc’ interpretability. Intrinsic interpretability involves constraining the complexity of the machine learning model itself\, resulting in models inherently interpretable due to their simplicity\, such as decision trees or sparse linear regression. On the other hand\, post hoc interpretability employs techniques that assess the model’s behavior after training\, offering insights into the model’s outcomes. Examples of post hoc techniques include permutation feature importance and the Shapley value method for feature importance. \nThe core contribution of this Thesis proposal lies in the development of novel methods to enhance both intrinsic and post hoc interpretability. These methods aim to advance the field by offering new perspectives on understanding machine learning models\, thereby contributing to the ongoing discourse on model transparency and user trust.
URL:https://ece.northeastern.edu/event/aria-masoomi-phd-proposal-review-2/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20231129T150000
DTEND;TZID=America/New_York:20231129T160000
DTSTAMP:20260414T035601
CREATED:20231127T213756Z
LAST-MODIFIED:20231127T213756Z
UID:6618-1701270000-1701273600@ece.northeastern.edu
SUMMARY:Aria Masoomi PhD Proposal Review
DESCRIPTION:Title:\nMaking Deep Neural Network Transparent \nDate:\n11/29/2023 \nTime:\n3:00:00 pm \nCommittee Members:\nProf. Jennifer Dy (Advisor)\nProf. Mario Sznaier\nProf. Eduardo Sontag\nProf. Peter Castaldi \nAbstract:\nAs machine learning algorithms are deployed ubiquitously to a variety of domains\, it is imperative to make these often black-box models transparent.\nThe ability to interpret and comprehend the reasoning behind machine learning models plays a pivotal role in increasing  user trust. It not only offers insights into how a model functions but also opens avenues for model enhancements. \nThis research delves into the realm of interpretability\, focusing on the dichotomy between ‘intrinsic’ and ‘post hoc’ interpretability. Intrinsic interpretability involves constraining the complexity of the machine learning model itself\, resulting in models inherently interpretable due to their simplicity\, such as decision trees or sparse linear regression. On the other hand\, post hoc interpretability employs techniques that assess the model’s behavior after training\, offering insights into the model’s outcomes. Examples of post hoc techniques include permutation feature importance and the Shapley value method for feature importance. \nThe core contribution of this Thesis proposal lies in the development of novel methods to enhance both intrinsic and post hoc interpretability. These methods aim to advance the field by offering new perspectives on understanding machine learning models\, thereby contributing to the ongoing discourse on model transparency and user trust.
URL:https://ece.northeastern.edu/event/aria-masoomi-phd-proposal-review/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20231127T080000
DTEND;TZID=America/New_York:20231127T170000
DTSTAMP:20260414T035601
CREATED:20231127T213640Z
LAST-MODIFIED:20231127T213640Z
UID:6616-1701072000-1701104400@ece.northeastern.edu
SUMMARY:Bruno Souto Maior Muniz Morais PhD Dissertation Defense
DESCRIPTION:Title:\nEnabling Domain Platform Design for Streaming Applications: A Holistic Approach \nCommittee Members:\nGunar Schirner (Advisor)\nProf. David Kaeli\nProf. Hamed Tabkhi (UNCC) \nTime:\n10:00:00 AM \nLocation: ISEC 601 \nAbstract:\nIn recent years\, more demanding streaming applications make striking a balance between high compute performance and efficiency paramount in platforms designs for edge computing. In addition\, designing a platform that is optimized for a single application is costly due to non-recurring engineering (NRE) costs. In contrast\, multiple applications can be grouped in domains\, e.g. computer vision\, software-defined radio. Leveraging shared characteristics of similar applications within a domain\, e.g. structural composition/computation patterns\, a single domain platform that caters to these similarities and accelerates applications can be generated\, thus benefiting multiple applications at once and dramatically improving NRE and time-to-market (TTM). \nThis dissertation introduces methodologies atvarious abstraction levels to enable streamlined domain platform design for streaming applications. Thrust 1 introduces high level DSE methods based on integer linear programming (ILP)\, Tile-based Synchronization Aware ILP (TSAR-ILP). Initially\, single-application platform allocations are considered using TSAR-ILP. While TSAR-ILP only focuses on applications in isolation\, its formulation lays the foundations for DmTSAR-ILP\, a method that performs domain DSE with multiple applications\, obtaining an optimal unified platform allocation that and achieving an increase of 22.5% in throughput\, while being 70x faster when compared to previous methods (MG-DmDSE). However\, DmTSAR-ILP aims to aggregate all applications fairly. This presents a challenge when the designer wishes to focus on a subset of applications. To enable ultimate flexibility in a product-oriented setting\, modeled after a market analysis process\, this dissertation introduces ProdDSE. ProdDSE enables application prioritization while also introducing concurrent application modeling and a multi-objective optimization (area\, performance) approach. This enables up to a 3.4x boost in performance depending on use case\, while also providing gains in DSE runtime (4.3x faster). \nThrust 2 introduces Sedona\, a domain-specific language (DSL) and exploration enviroment that captures parametric dataflow application descriptions with language features dedicated to streaming applications. A design identified by Thrust 1 can be further refined using the tools in Thrust 2\, by capturing the connectivity of a design using Sedona. Then\, automatic wiring is performed for target outputs such as timing-aware simulations or RTL-level code\, enabling structural manipulation at a high-level description without the burden of low-level manual integration. \nFinally\, to better guide the high-level decisions performed in Thrust 1 and further exploration/integration in Thrust 2\, Thrust 3 considers the implications of HWACC topology choices in an HWACC-rich SoC. The ACTAR flow is introduced to explore different topologies in a RISC-V based SoC and the side-effects of topology and memory sizing choices on the system-wide performance and synchronization burdens due computation offloading to HWACCs. This produces valuable and actionable insights for designers to make informed choices on system-level compositions depending on application communication and computation demands.
URL:https://ece.northeastern.edu/event/bruno-souto-maior-muniz-morais-phd-dissertation-defense/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20230803T100000
DTEND;TZID=America/New_York:20230803T110000
DTSTAMP:20260414T035601
CREATED:20230508T193859Z
LAST-MODIFIED:20230508T193859Z
UID:6321-1691056800-1691060400@ece.northeastern.edu
SUMMARY:Yu Yin's PhD Dissertation Defense
DESCRIPTION:“Synthetic Data Generator: Understanding Human Face & Body via Image Synthesis” \nCommittee Members:\nProf. Yun Fu (Advisor)\nProf. Sarah Ostadabbas\nProf. Ming Shao \nAbstract:\nThe community has long enjoyed the benefits of synthesizing data\, as it provides a reliable and controllable source for training machine learning models while reducing the need for data collection from the real world. Human face and body synthesis are especially appealing to research communities\, where model fairness and ethical deployment are critical concerns. However\, generating digit humans that are convincing\, realistic-looking\, identity-preserving\, and high-quality are still challenging in 2D and 3D image synthesis. \nThis dissertation investigates the potential for understanding human behavior by recreating it\, and can be broadly divided into three sections. (1) In Section one\, we explore the 2D image generation models and their interaction with face applications (i.e.\, landmark localization and face recognition tasks). Specifically\, super-resolution (SR) and landmark localization of tiny faces are highly correlated tasks. To this end\, we propose joint frameworks that enable face alignment and SR to benefit from one another\, hence enhancing the performance of both tasks. Moreover\, we demonstrate that face frontalization provides an effective and efficient way for face data augmentation and further improves face recognition performance in extreme pose scenarios. (2) In Section two\, we explore the 3D parametric generation models and how they support human body pose and shape estimation. Advancing technology to monitor our bodies and behavior while sleeping and resting is essential for healthcare. However\, keen challenges arise from our tendency to rest under blankets. To mitigate the negative effects of blanket occlusion\, we use an attention-based restoration module to explicitly reduce the uncertainty of occluded parts by generating uncovered modalities\, which further update the current estimation via a cyclic fashion. (3) In Section three\, we explore the 3D Nerf-based Generative models in generating high-quality images with consistent 3D geometry. We propose a universal method to surgically fine-tune these NeRF-GAN models in order to achieve high-fidelity animation of real subjects only by a single image.
URL:https://ece.northeastern.edu/event/yu-yins-phd-dissertation-defense/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20230721T140000
DTEND;TZID=America/New_York:20230721T153000
DTSTAMP:20260414T035601
CREATED:20230718T175222Z
LAST-MODIFIED:20230718T175222Z
UID:6433-1689948000-1689953400@ece.northeastern.edu
SUMMARY:Daniel Uvaydov's PhD Dissertation Defense
DESCRIPTION:Title: Real-Time Spectrum Sensing for Inference and Control \nCommittee Members: \nProf. Tommaso Melodia (Advisor) \nProf. Kaushik Choudhury \nProf. Francesco Restuccia \nAbstract: \nThrough growing cellular innovations\, the usage and congestion of the wireless spectrum is increasing at incredible speeds. High demand and limited supply pose a resource issue known as the “spectrum crunch”. With the high diversity of users sharing a large portion of the spectrum to request and receive diverse services\, spectrum coordination becomes very difficult. Large scale device synchronization for spectrum coordination requires high overhead and more wireless transmissions further reducing spectrum resources. However\, by monitoring the spectrum\, otherwise known as spectrum sensing\, we can develop mechanisms where users can opportunistically take action based on the current state of the spectrum\, without need for direct coordination between devices. Spectrum sensing can enable the next generation of wireless applications ranging from opportunistic spectrum access to cognitive radio networks. The key unaddressed challenges of spectrum sensing are that (i) it requires very extensive and diverse datasets; (ii) it has to be performed with extremely low latency over varying bandwidths and must guarantee strict real-time processing constraints; (iii) its underlying algorithms need to be extremely accurate\, and flexible enough to work with different wireless bands and protocols to find application in real-world settings. This dissertation focuses on addressing these challenges in multiple wireless applications by utilizing Deep Learning (DL) techniques as the main vehicle of spectrum sensing for both inference and control. Algorithmic spectrum sensing has generally been model-based which limits its performance in diverse settings and environments\, for this reason we explore data-driven spectrum sensing algorithms. Mainly\, this work takes a holistic approach to address spectrum sensing problems from multiple directions with the overarching goal of developing the core building blocks for the next generation of intelligent\, AI-driven\, efficient spectrum sharing systems. By leveraging mechanisms such as data augmentation\, channel attention\, voting\, and segmentation we are able to push beyond the capabilities of existing DL techniques and create generalizable spectrum sensing algorithms. Furthermore we deploy different spectrum sensing solutions in real testbeds for over the air evaluations and applicable proof-of-concepts. The contributions of this work includes (i) multiple datasets and implementations for DL enabled spectrum sensing with applications in radio frequency and underwater; (ii) a method for tackling the core issue of dataset generation in supervised learning algorithms for spectrum sensing via a novel data augmentation technique; (iii) a study into one of the first ever semi-unsupervised approaches for wideband multi-class spectrum sensing.
URL:https://ece.northeastern.edu/event/daniel-uvaydovs-phd-dissertation-defense/
LOCATION:432 ISEC\, 360 Huntington Ave\, Boston\, MA\, 02115\, United States
GEO:42.3396156;-71.0886534
X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=432 ISEC 360 Huntington Ave Boston MA 02115 United States;X-APPLE-RADIUS=500;X-TITLE=360 Huntington Ave:geo:-71.0886534,42.3396156
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20230720T130000
DTEND;TZID=America/New_York:20230720T140000
DTSTAMP:20260414T035601
CREATED:20230711T180015Z
LAST-MODIFIED:20230711T180015Z
UID:6425-1689858000-1689861600@ece.northeastern.edu
SUMMARY:Qing Jin's PhD Dissertation Defense
DESCRIPTION:Title:Decoupling Efficiency-Performance Optimization for Modern Neural Networks \nDate: \n7/20/2023 \nCommittee Members: \nYanzhi Wang (Advisor); Prof. David Kaeli; Prof. Sunil Mittal; Prof. Jennifer Dy \nAbstract: \nDeep learning has achieved remarkable success in a variety of modern applications\, but this success is often accompanied by inefficiency in terms of storage and inference speed\, which can hinder their practical use on resource-constrained hardware. Developing highly efficient neural networks that maintain high prediction accuracy is crucial and challenging. This dissertation explores the potential for simultaneously achieving high efficiency and high prediction accuracy in neural networks\, and can be broadly divided into three sections. (1) In Section One\, we explore the implementation of highly efficient generative adversarial networks (GANs) capable of generating high-quality images within a predefined computational budget. The key challenge lies in identifying the optimal architecture for the generative model while simultaneously preserving the quality of the generated images from the compressed model\, despite its reduced computational cost. To achieve this\, we propose a novel neural architecture search (NAS) algorithm and a new knowledge distillation technique. (2) In Section Two\, we explore the challenge of quantizing discriminative models without relying on high-precision multiplications. To address this issue\, we present an innovative approach to determine the optimal fixed-point formats for both weights and activations based on their statistical properties. Our results demonstrate that high accuracy in quantized neural networks can be achieved without the need for high-precision multiplications. (3) In Section Three\, we delve into the challenge of training neural networks for innovative computing platforms\, specifically processing-in-memory (PIM) systems. Through a detailed mathematical derivation of the backward propagation algorithm\, we facilitate the training of quantized models on these platforms. Additionally\, through a thorough theoretical analysis of training dynamics\, we ensure convergence and propose a systematic solution for quantizing neural networks on PIM systems.
URL:https://ece.northeastern.edu/event/qing-jins-phd-dissertation-defense/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20230629T170000
DTEND;TZID=America/New_York:20230629T173000
DTSTAMP:20260414T035601
CREATED:20230626T213009Z
LAST-MODIFIED:20230626T213036Z
UID:6407-1688058000-1688059800@ece.northeastern.edu
SUMMARY:Zifeng Wang's PhD Dissertation Defense
DESCRIPTION:Title: Effective and Efficient Continual Learning \nCommittee Members:\nProf. Jennifer Dy (Advisor)\nProf. Stratis Ioannidis\nProf. Yanzhi Wang \nAbstract:\nContinual Learning (CL) aims to develop models that mimic the human ability to learn continually without forgetting knowledge acquired earlier. While traditional machine learning methods focus on learning with a certain dataset (task)\, CL methods adapt a single model to learn a sequence of tasks continually. \nIn this thesis\, we target developing effective and efficient CL methods under different challenging and resource-limited settings. Specifically\, we (1) leverage the idea of sparsity to achieve cost-effective CL\, (2) propose a novel prompting-based paradigm for parameter-efficient CL\, and (3) utilize task-invariant and task-specific knowledge to enhance existing CL methods in a general way. \nWe first introduce our sparsity-based CL methods. The first method\, Learn-Prune-Share (LPS)\, splits the network into task-specific partitions\, leading to no forgetting\, while maintaining memory efficiency. Moreover\, LPS integrates a novel selective knowledge sharing scheme\, enabling adaptive knowledge sharing in an end-to-end fashion. Taking a step further\, we present Sparse Continual Learning (SparCL)\, a novel framework that leverages sparsity to enable cost-effective continual learning on edge devices. SparCL achieves both training acceleration and accuracy preservation through the synergy of three aspects: weight sparsity\, data efficiency\, and gradient sparsity. \nSecondly\, we present a new paradigm\, prompting-based CL\, that aims to train a more succinct memory system that is both data and memory efficient. We first propose a method that learns to dynamically prompt (L2P) a pre-trained model to learn tasks sequentially under different task transitions\, where prompts are small learnable parameters maintained in a memory space. We then improve L2P by proposing DualPrompt\, which decouples prompts into complementary “General” and “Expert” prompts to learn task-invariant and task-specific instructions\, respectively. \nFinally\, we propose DualHSIC\, a simple and effective CL method that generalizes the idea of leveraging task-invariant and task-specific knowledge. DualHSIC consists of two complementary components that stem from the so-called Hilbert Schmidt independence criterion (HSIC): HSIC-Bottleneck for Rehearsal (HBR) lessens the inter-task interference and HSIC Alignment (HA) promotes task-invariant knowledge sharing. \nComprehensive experimental results demonstrate the effectiveness and efficiency of our methods over the state-of-the-art methods on multiple CL benchmarks.
URL:https://ece.northeastern.edu/event/zifeng-wangs-phd-dissertation-defense/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20230626T083000
DTEND;TZID=America/New_York:20230626T093000
DTSTAMP:20260414T035601
CREATED:20230624T220848Z
LAST-MODIFIED:20230624T220848Z
UID:6402-1687768200-1687771800@ece.northeastern.edu
SUMMARY:Deniz Unal's PhD Proposal Review
DESCRIPTION:Title:\nSoftware-Defined Underwater Acoustic Networks \nCommittee Members:\nProf. Tommaso Melodia (Advisor)\nProf. Stefano Basagni\nProf. Kaushik Chowdhury\nDr. Emrecan Demirors \nAbstract:\nThe exploration\, monitoring\, and understanding of oceans play a crucial role in addressing climate change\, overseeing underwater pipelines\, and preventing maritime warfare attacks. To achieve these significant objectives\, it is vital to utilize networks of cost-effective and flexible underwater devices capable of efficiently collecting and transmitting information to the shore. However\, the progress of underwater networks heavily relies on underwater acoustic modems\, which currently face limitations such as low data rates and inflexible hardware designs\, limiting their usability to specific scenarios. To overcome these limitations\, we propose a modular software-defined acoustic networking platform built on the Zynq system-on-chip architecture that can be easily deployed in a compact form factor. Our platform distinguishes itself from existing solutions in several ways. Firstly\, it possesses the capability to adapt to varying conditions by adjusting protocol parameters at all layers of the networking stack. Secondly\, it achieves high data rate connections\, particularly over short distances. Additionally\, it seamlessly integrates with other sub-sea platforms\, including underwater drones. We demonstrate the capabilities and the performance of our platform with tasks\, such as channel estimation and characterization\, establishing high data rate Orthogonal Frequency-Division Multiplexing (OFDM) links\, and running third-party software to implement JANUS standard. In addition\, we introduce the enabling technologies for the development and implementation of underwater networks. These technologies facilitate the establishment of connectivity between underwater networks and the shore\, as well as the integration of modems with underwater vehicles. Lastly\, we provide a demonstration of the algorithmic development conducted on our platform. We mainly consider high-rate\, wideband\, adaptive links and perform experimental evaluations at sea. In particular\, we demonstrate multicarrier communications with mobile platforms with the presence of Doppler and compare the performance of forward error correction methods\, and demonstrate dataset recording for artificial intelligence research.
URL:https://ece.northeastern.edu/event/deniz-unals-phd-proposal-review/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20230623T100000
DTEND;TZID=America/New_York:20230623T110000
DTSTAMP:20260414T035601
CREATED:20230606T193237Z
LAST-MODIFIED:20230606T193237Z
UID:6395-1687514400-1687518000@ece.northeastern.edu
SUMMARY:Cooper Loughlin's PhD Dissertation Defense
DESCRIPTION:“Deep Generative Models for High Dimensional Spatial and Temporal Data Analysis” \nCommittee Members:\nProf. Vinay Ingle (Advisor)\nDr. Dimitris Manolakis\nProf. Purnima Ratilal-Makris \nAbstract:\nData analysis and exploitation in practical applications is challenging when observations are the result of many interacting natural and man-made phenomena. We address two important problems for which traditional methods of analysis are insufficient. One problem of practical interest is the identification of particular materials from remotely sensed hyperspectral imagery. This is traditionally accomplished by comparing image pixel spectra to those from a known material library. Such techniques are limited by spectral variability\, background interference\, and imperfect compensation of atmospheric components. Established methods address these limitations with statistical techniques. Simple probability models result in tractable methods; however\, analyses are limited by errors due\, in particular\, due to false alarms. \nAnalysis of complex time series is another challenging problem\, particularly when data are high dimensional. This arises in air quality monitoring\, where atmospheric concentration measurements of multiple pollutants are taken over time. Two analysis goals in this context are forecasting and anomaly detection. Both tasks are enabled by an accurate model for the temporal dynamics and interaction between pollutants. Air quality data are complex due to long term temporal dependencies\, non-linear dependence between pollutants\, and missing observations. Traditional multivariate time series analysis approaches\, such as the vector autoregression and linear dynamical system models\, fail to capture those characteristics necessary for a sufficient probabilistic model. \nWe use deep generative models to develop practical solutions that address these problems. This is made possible through the application of deep latent variable models. The modeling approach follows the philosophy that complex data can typically be explained by simpler underlying factors of variation. Variational autoencoders (VAEs) are deep latent variable models that emulate data generation by transforming simple\, low dimensional\, latent random vectors through a deep neural network. VAEs are trained to produce samples that resemble the training data\, thus capturing a manifold on which complex data are distributed. This philosophy is extended to time series data\, where we consider sequences of latent vectors. \nWe utilize VAEs develop a flexible generative model for hyperspectral imagery. Based on that model\, we develop a novel material identification framework which localizes target material spectra along the manifold. Through experiments on real data\, we show that the \ac{VAE} approach is better able to reject false alarms from materials with similar spectra when compared to established methods alone. We additionally develop a novel dynamical \ac{VAE} model for time series of air quality data. Using that model\, we develop practical methods for computing forecast distributions using Monte Carlo integration. We evaluate forecast distributions against real air quality data and demonstrate the ability to predict temporal dynamics and forecast uncertainty. The primary contribution of this work is to develop practical solutions to challenging data analysis problems through the use of deep generative models.
URL:https://ece.northeastern.edu/event/cooper-loughlins-phd-dissertation-defense/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20230620T130000
DTEND;TZID=America/New_York:20230620T140000
DTSTAMP:20260414T035601
CREATED:20230522T212041Z
LAST-MODIFIED:20230522T212041Z
UID:6354-1687266000-1687269600@ece.northeastern.edu
SUMMARY:Chang Liu's PhD Dissertation Defense
DESCRIPTION:“Unleashing the Potential of Transfer Learning for Visual Applications” \nCommittee Members:\nProf. Raymond Fu (Advisor)\nProf. Sarah Ostadabbas\nProf. Zhiqiang Tao \nAbstract:\nThe recent flourish of deep learning in various tasks is largely accredited to the rich and accessible labeled data. Nonetheless\, massive supervision remains a luxury for many real-world applications. Further\, the domain shift problem has also seriously impeded large-scale deployments of deep-learning models. As a remedy\, Transfer learning aims at improving the performance of target learners on target domains by transferring the knowledge contained in different but related source domains. In this way\, the dependence on a large number of target domain data can be reduced for constructing target learners. \nIn this dissertation research\, I investigate two major problems in transfer learning\, domain adaptation (DG) and domain adaptation (DA)\, on various visual applications. (1) The challenge of DG lies in an over-simplified assumption\, that is\, the source and target data are independent and identically distributed (i.i.d.) while ignoring out-of-distribution (OOD) scenarios commonly encountered in practice. This issue is common in visual applications such as object recognition\, hyperparameter optimization\, and face recognition. We propose algorithms that are specifically designed for each task\, such as metric learning\, adversarial regularization\, feature disentanglement\, and meta-learning. (2) DA can be considered a special case of DG with unlabeled target data available. The major challenge is how to align the labeled source and unlabeled target data. We delve into the applications of image recognition and video recognition and propose algorithms to ensure domain-wise discriminativeness and class-wise closeness across domains. Experiments show that the proposed algorithms outperform the state-of-the-art methods on the commonly-used benchmark datasets.
URL:https://ece.northeastern.edu/event/chang-lius-phd-dissertation-defense/
END:VEVENT
END:VCALENDAR