BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Department of Electrical & Computer Engineering - ECPv6.15.20//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:20200308T070000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:20201101T060000
END:STANDARD
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
DTSTART:20210314T070000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:20211107T060000
END:STANDARD
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
DTSTART:20220313T070000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:20221106T060000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210423T130000
DTEND;TZID=America/New_York:20210423T140000
DTSTAMP:20260424T095938
CREATED:20210421T194056Z
LAST-MODIFIED:20210421T194056Z
UID:4866-1619182800-1619186400@ece.northeastern.edu
SUMMARY:ECE PhD Dissertation Defense: Lichen Wang
DESCRIPTION:PhD Dissertation Defense: Correlation Discovery for Multi-view and Multi-label Learning \nLichen Wang \nLocation: Zoom Link \nAbstract: Correlation indicates the interactions or connections across different instances. It exists in a wide range of real-world applications such as social network\, scene understanding\, and time-series data analysis. Correlation provides the unique and informative knowledge to reveal the connections across instances\, and it plays an essential and important role in machine learning field. However\, recovering and utilizing correlation is challenging. First\, it is hard to explicitly define and understand the correlations. Second\, there are not sufficient datasets which contain the well-labeled task-specific correlations. Third\, how to efficiently utilize the learned correlations for other down-stream tasks have not been well-explored.\nIn this dissertation research\, we investigate the techniques to effectively discover various kinds of correlations in machine learning tasks including multi-view learning\, multi-label learning\, image/scene understanding\, time-series data analysis\, human action recognition\, and graph representation learning. Specifically\, we propose algorithms from the following perspectives: (1) designing an advanced correlation discovery network to automatically explore the label correlations in multi-label scenarios\, (2) proposing a multi-view fusion strategy which effectively dig the latent correlations across different views\, (3) exploring the correlations and structural knowledge from graph structured objects in an inductive and unsupervised scenario. To demonstrate the effectiveness of the proposed algorithms\, various experiments on commonly used datasets have been implemented and the results shows the superiority of our algorithms over the other state-of-the-art methods.
URL:https://ece.northeastern.edu/event/ece-phd-dissertation-defense-lichen-wang/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210422T123000
DTEND;TZID=America/New_York:20210422T133000
DTSTAMP:20260424T095938
CREATED:20210414T213642Z
LAST-MODIFIED:20210414T213642Z
UID:4845-1619094600-1619098200@ece.northeastern.edu
SUMMARY:MS Thesis Defense: Duschia Bodet
DESCRIPTION:MS Thesis Defense: Modulations to Exploit the THz Band \nDuschia Bodet \nLocation: Zoom Link \nAbstract: Terahertz (THz)-band (0.1-10 THz) communication has been envisioned as a key technology to enable wireless Terabit-per-second (Tbps) links. At THz frequencies\, the path-loss is governed by the spreading loss and the molecular absorption loss. The latter also determines the available transmission bandwidth\, which drastically shrinks with distance. As a result\, traditional modulation schemes cannot fully take advantage the THz channel\, and new modulation schemes are needed if THz channel communications are going to reach their full potential. Several solutions have been presented including Hierarchical Bandwidth Modulations (HBM)\, which is the only presented work that not only compensates for molecular absorption losses but leverages those losses to improve the capabilities of the system. The focus of this thesis is two-fold. First the design of HBM is formalized\, exploring the trade-offs and its achievable performance as a function of different system parameters. Secondly\, these trade-offs and performance metrics are verified using a one-of-a-kind experimental testbed for ultrabroadband communication networks. The results show that with proper design HBM successfully achieves its goal of exploiting the distance-dependent characteristics of the THz channel.
URL:https://ece.northeastern.edu/event/ms-thesis-defense-duschia-bodet/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210422T103000
DTEND;TZID=America/New_York:20210422T113000
DTSTAMP:20260424T095938
CREATED:20210405T174659Z
LAST-MODIFIED:20210405T174659Z
UID:4823-1619087400-1619091000@ece.northeastern.edu
SUMMARY:ECE PhD Dissertation Defense: Linbin Chen
DESCRIPTION:PhD Dissertation Defense: Low Power Designs using Approximate Computing and Emerging Memories at Nanoscales \nLinbin Chen \nLocation: Zoom Link \nAbstract: A power efficient integrated circuit design is essential for mobile and embedded computer systems. This dissertation proposes several novel low power designs using approximate computing and emerging memories for computers with arithmetic circuits and large on-chip caches. Initially\, low power approximate designs are proposed both for fixed point radix-2 and high-radix division at circuit-level. Then\, an approximate parallel CORDIC algorithm and its hardware implementation are developed. Trade-offs between circuit metrics and error characteristics are pursued by simulation and analysis. The proposed approximate arithmetic designs have excellent performance for image processing applications while significantly reducing power consumption. Then\, hybrid cache designs integrating SRAM with emerging memories are also investigated. An intra-cell\, as well as inter-subarray and inter-bank hybrid caches with SRAM\, eDRAM and NVM (such as PCM or STT-MRAM) are proposed. Architectural level approaches such as special migration structures and policies are designed to address the eDRAM refresh requirements and the NVM large write latency issue. An analytical circuit-level model based on NVsim focusing on hybrid granularity and an architecture level model based on gem5 focusing on a migration policy are developed. To explore the hybrid cache’s benefits to main memory\, a combined-cache design for addressing endurance issues of multi-level non-volatile memory in embedded system is proposed. It is shown that these hybrid cache designs exhibit smaller area and lower leakage than conventional designs so with great potential to be used for large-capacity on-chip caches in mobile and embedded systems.
URL:https://ece.northeastern.edu/event/ece-phd-dissertation-defense-linbin-chen/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210422T100000
DTEND;TZID=America/New_York:20210422T110000
DTSTAMP:20260424T095938
CREATED:20210420T180653Z
LAST-MODIFIED:20210420T180709Z
UID:4855-1619085600-1619089200@ece.northeastern.edu
SUMMARY:ECE PhD Dissertation Defense: Seyedmehdi Sadeghzadeh
DESCRIPTION:PhD Dissertation Defense: Physical Layer Security in Multi-User Wireless Networks: Impact of Interference and Artificial Noise on Large-Scale Analysis \nSeyedmehdi Sadeghzadeh \nLocation: Zoom Link \nAbstract: In this thesis\, we study the physical layer security in downlink multi-user wireless networks. Traditionally\, security has been addressed by cryptography at the higher layers of the communication stack. Security at the physical layer has been a major research topic in recent years. We study two different precoder designs alongside artificial noise (AN) to mitigate multi-user interference and deteriorate reception at the eavesdropper (Eve). We study the large scale analysis to calculate the secrecy sum-rate for these two cases and analyze the effect of AN on the system. First\, we consider the worst case scenario\, when eavesdropper’s (Eve’s) rate is not deteriorated by the interference caused by the legitimate users. Later\, we investigate how interference from legitimate users would affect the large scale security sum rate. At the end\, we assume more practical situation where the channel state information at the transmitter is not perfect due to feedback limitation and estimation error.
URL:https://ece.northeastern.edu/event/ece-phd-dissertation-defense-seyedmehdi-sadeghzadeh/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210422T100000
DTEND;TZID=America/New_York:20210422T110000
DTSTAMP:20260424T095938
CREATED:20210414T213301Z
LAST-MODIFIED:20210414T213301Z
UID:4843-1619085600-1619089200@ece.northeastern.edu
SUMMARY:ECE MS Thesis Defense: Tianhong Xu
DESCRIPTION:MS Thesis Defense: A novel simple power analysis (SPA) attack on Elliptic Curve Cryptography (ECC) \nTianhong Xu \nLocation: Zoom Link \nAbstract: Elliptic Curve Cryptography (ECC)\, as a widely used public-key cryptography\, is vulnerable to simple power analysis(SPA) attacks. There are many countermeasures against simple power analysis(SPA) attacks on ECC implementation\, the Always-add algorithm is one of the most popular countermeasures. This research proposes a new SPA attack which is effective to the ECC encrypting implemented with Always-add algorithm\, it uses deep-learning tools and statistical method to retrieve a secret key from only one EM trace collected from a ASIC circuit running ECC encryption.
URL:https://ece.northeastern.edu/event/ece-ms-thesis-defense-tianhong-xu/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210422T090000
DTEND;TZID=America/New_York:20210422T100000
DTSTAMP:20260424T095938
CREATED:20210420T215252Z
LAST-MODIFIED:20210420T215252Z
UID:4860-1619082000-1619085600@ece.northeastern.edu
SUMMARY:ECE MS Thesis Defense: Peter Kelly
DESCRIPTION:MS Thesis Defense: Design of a Thruster-assisted Bipedal Robot \nPeter Kelly \nLocation: Zoom Link \nAbstract: During the past few years\, legged robot technology has been rapidly advancing.\nHowever\, even the most advanced bipedal legged robots are susceptible to strong disturbances and slippery or impassible terrain. By introducing thrusters to enable hybrid legged-aerial locomotion\, these problems can be circumvented by increasing a robot’s stability and allowing it to jump over obstacles. Harpy is a bipedal robot with eight actuators and two thrusters that serves as a hardware platform for developing control algorithms to advance research in thruster assisted bipedal legged locomotion. This thesis explores the conception\, simulation\, and electromechanical design process of the robot\, which prioritizes thrust-to-weight ratio\, impact resistance\, power density\, and modularity. The fabrication process of actuators and the leg which enable the robot to be both light and strong and testing of the leg design and thrusters is also discussed.
URL:https://ece.northeastern.edu/event/ece-ms-thesis-defense-peter-kelly/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210421T173000
DTEND;TZID=America/New_York:20210421T173000
DTSTAMP:20260424T095938
CREATED:20210421T193821Z
LAST-MODIFIED:20210421T193821Z
UID:4864-1619026200-1619026200@ece.northeastern.edu
SUMMARY:ECE PhD Dissertation Defense: Muhamed Yildiz
DESCRIPTION:PhD Dissertation Defense: Interpretable Machine Learning for Retinopathy of Prematurity \nMuhamed Yildiz \nLocation: Zoom Link \nAbstract: Retinopathy of Prematurity (ROP)\, a leading cause of childhood blindness\, is diagnosed by clinical ophthalmoscopic examinations or reading retinal images. Plus disease\, defined as abnormal tortuosity and dilation of the posterior retinal blood vessels\, is the most important feature to determine treatment-requiring ROP. State-of-the-art ROP detection systems employ convolutional neural networks (CNNs) %\cite{brown2018automated} and achieve up to $0.947$ and $0.982$ area under the ROC curve (AUC) in the discrimination of \textit{normal} and \textit{plus} levels of ROP. However\, due to their black-box nature\, clinicians are reluctant to trust diagnostic predictions of CNNs.\nFirst\, we aim to create an interpretable\, feature extraction-based pipeline\, namely\, I-ROP ASSIST\, that achieves CNN like performance when diagnosing plus disease from retinal images. Our method segments retinal vessels\, detects the vessel centerlines. Then\, our method extracts features relevant to ROP\, including tortuosity and dilation measures\, and uses these features for classification via logistic regression\, support vector machines and neural networks to assess a severity score for the input. For predicting \textit{normal} and \textit{plus} levels of ROP on a dataset containing 5512 posterior retinal images\, we achieve $0.88$ and $0.94$ AUC\, respectively. Our system combining automatic retinal vessel segmentation\, tracing\, feature extraction and classification is able to diagnose plus disease in ROP with CNN like performance.\nThen\, we introduce a novel method for extracting tortuosity features. Current feature extraction pipelines of retinal image analysis systems extract tortuosity features based on the derivatives of vessel centerlines or a segment of a vessel. Our method eliminates the need for finding vessel centerlines by introducing a method for calculating curvature at each pixel in the fundus image. When calculating curvature\, we use the geometric interpretation of eigenvectors of the Hessian of an interpolation function. By selecting an appropriate interpolation function\, our method can be applied in many domains\, including corner detection\, noise removal and image registration. We present the results of our method on artificial images that contains curved structures such as circle\, sine waves as well as real images from MNIST and our retinal fundus image dataset. Experimental results shows that our model accurately captures the high curvature parts of the blood vessels. \nFurthermore\, we aim to address the interpretability problem of CNN-based ROP detection system. Incorporating visual attention capabilities in CNNs enhances interpretability by highlighting regions in the images that CNNs utilize for prediction. Generic visual attention methods do not leverage structural domain information such as tortuosity and dilation of retinal blood vessels in ROP diagnosis. We propose the Structural Visual Guidance Attention Networks (SVGA-Net) method\, that leverages structural domain information to guide visual attention in CNNs. SVGA-Net achieves $0.979$ and $0.987$ AUC to predict \textit{normal} and \textit{plus} levels of ROP. Moreover\, SVGA-Net consistently results in higher AUC compared to visual attention CNNs without guidance\, baseline CNNs\, and CNNs with structured masks.
URL:https://ece.northeastern.edu/event/ece-phd-dissertation-defense-muhamed-yildiz/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210421T100000
DTEND;TZID=America/New_York:20210421T170000
DTSTAMP:20260424T095938
CREATED:20210406T210701Z
LAST-MODIFIED:20210406T210701Z
UID:4831-1618999200-1619024400@ece.northeastern.edu
SUMMARY:MS Thesis Defense: Yuezhou Liu
DESCRIPTION:MS Thesis Defense: Optimizations of Caching Networks: Fairness and Application to Mobile Networks \nYuezhou Liu \nLocation: Zoom Link \nAbstract: In-network caching is playing a more and more important role in today’s network architectures\, because of the explosive growth of data traffic due to the proliferation of mobile devices and demands for high-volume media content\, as well as the development of low-latency applications\, such as VR/AR and cloud gaming. The replication of popular contents in the caches that located closer to end users than central servers\, can significantly reduce backbone traffic\, benefit request latency\, and balance the load of central servers. In this thesis\, we study two problems in the field of network caching. In the first part\, we consider fair caching policies in caching networks with arbitrary topology. We introduce a utility maximization framework to find a caching decision that reduces aggregate expected request routing cost in the network while taking fairness issues into consideration. The utility maximization problem is NP-hard\, and we propose two efficient approximation algorithms to solve it. In the second part\, we study how caching may affect user association in mobile networks. We jointly optimize the user association decision and caching at both base stations (BSs) and gateways (GWs). The resulting problem is also NP-hard. We propose a polynomial-time algorithm based on concave approximation and pipage rounding that produces a solution within a constant factor of 1-1/e from the optimal. Simulation results show that the proposed algorithm outperforms schemes that combine cache-independent user association methods with traditional caching strategies (e.g.\, LRU) in terms of minimizing the aggregate expected routing cost and backhaul traffic while achieving a high data sum rate in the access network.
URL:https://ece.northeastern.edu/event/ms-thesis-defense-yuezhou-liu/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210421T100000
DTEND;TZID=America/New_York:20210421T110000
DTSTAMP:20260424T095938
CREATED:20210420T180528Z
LAST-MODIFIED:20210420T180528Z
UID:4854-1618999200-1619002800@ece.northeastern.edu
SUMMARY:ECE MS Thesis Defense: Matin Raayai Ardakani
DESCRIPTION:MS Thesis Defense: A Framework for Denoising Two and Three-dimensional Monte CarloPhoton Transport Simulations Using Convolutional Neural Networks \nMatin Raayai Ardakani \nLocation: Zoom Link \nAbstract: The Monte Carlo (MC) method is considered to be the gold standard for modeling light propagation inside turbid media\, proving superior to other Radiative Transfer Equation (RTE) solvers relying on variational principles. However\, like most MC-based algorithm\, a large number of independently launched photons is needed for converging to the correct result and combating its inherent stochastic noise\, yielding longer computation times\, even when accelerated on GraphicProcessing Units (GPUs).\nTo remove this noise from the output without increasing the number of photons used for simulation\, modified versions of commonly used filters for image and volumetric data based on non-local self similarity has been used in the past. Current state-of-the-art denoising approaches rely on Convolutional Neural Networks (CNN) to remove spatially variant noise\, but the high dynamic range of MC simulations has hindered their adaptation to remove MC noise.\nIn this thesis\, we address this problem by presenting a supervised framework for using CNNs to denoise MC simulations. First\, a dataset is created with each entry comprising of a unique configuration simulated with different numbers of photons. The simulation configurations are generated using a simple generative model that introduces objects with both smooth and sharp edges into the volume. By selecting the group of fluence maps simulated with the maximum number of photons in the dataset as labels\, we train a range of CNN-based models to learn the underlying mapping between noisy and clean images. The CNN input is converted to log scale and normalized to reduce the high dynamic range\, and converted back after inference. The trained CNNs are then shown to have better performance compared to using an Adaptive Non-local Means filter\, in terms of mean square error (MSE)\, structural similarity index (SSIM)\, and peak signal-to-noise ratio (PSNR) in the image domain.\nFinally\, we purpose our own architecture that combines DnCNN and UNet\, a strategy that can learn both local and global residual noise maps\, achieving state-of-the-art performance compared to existing CNN methods. Future avenues of research and challenges for denoising 3D simulations are also discussed.
URL:https://ece.northeastern.edu/event/ece-ms-thesis-defense-matin-raayai-ardakani/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210421T090000
DTEND;TZID=America/New_York:20210421T100000
DTSTAMP:20260424T095938
CREATED:20210420T175730Z
LAST-MODIFIED:20210420T175730Z
UID:4851-1618995600-1618999200@ece.northeastern.edu
SUMMARY:ECE MS Thesis Defense: Rubens Lacouture
DESCRIPTION:MS Thesis Defense: GPUBLQMR: GPU-Accelerated Sparse Block Quasi-Minimum Residual Linear Solver \nRubens Lacouture \nLocation: Zoom Link \nAbstract: Solutions of linear systems of equations is the central point of many scientific and engineering research problems across a variety of domains. In many cases\, the solution of linear systems can even take most of the simulation time which presents a huge computational bottleneck issue. This can hinder the scalability of various scientific software hindering for larger problems. For large-scale simulations\, this can result in having to find the solutions of millions of unknowns\, making this an ideal problem to exploit parallelism to improve performance.\nPreconditioned Krylov subspace methods have proven effective and robust in various applications. The block Quasi-Minimum Residual (BLQMR) method as developed by Boyse et al. has been shown to be efficient for solving systems of equations with multiple righthand sides. This method is based on the conventional Quasi-Minimum Residual (QMR) method which is generalized using the block Lanczos algorithm to solve multiple solutions simultaneously. In particular\, it is shown that this method accelerates the convergence behavior based on the set number of righthand sides\, grouped to be solved simultaneously. Block iterative solver methods are often characterized by a high degree of parallelism.\nIn this thesis\, we show how BLQMR can be successfully implemented on a distributed memory computer taking advantage of Graphics Processing Units (GPU) accelerators. We leveraged the processing power of GPUs to show how the proposed GPU-accelerated BLQMR approach can out-perform state-of-the-art linear solvers and results in an ideal behavior for solving challenging linear algebra problems through data from various numerical experiments. The library code developed in this work is written using the CUDA framework. The performance of the parallel algorithm is optimized using several CUDA optimization strategies and the speedup of the parallel GPU implementation over the existing sequential CPU implementations is reported.
URL:https://ece.northeastern.edu/event/ece-ms-thesis-defense-rubens-lacouture/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210420T160000
DTEND;TZID=America/New_York:20210420T170000
DTSTAMP:20260424T095938
CREATED:20210414T213500Z
LAST-MODIFIED:20210414T213500Z
UID:4844-1618934400-1618938000@ece.northeastern.edu
SUMMARY:MS Thesis Defense: Hao Chen
DESCRIPTION:MS Thesis Defense: Reconstruction of Sulcal Geometry in Brain Stimulation Models using Spherical Harmonics \nHao Chen \nLocation: Zoom Link \nAbstract: Over the past few years\, there has been increasing interest in transcranial electrical stimulation (tCS) and thus it has been the subject of a growing number of simulation studies. Indeed\, some federal agencies in the US now require model-based simulations to be included as part of tCS grant proposal. In order to obtain more accurate simulation results and guide the relevant research\, it is of important to assess the impact of the accuracy of the anatomical 3D brain model that these studies depend on. However\, due to the partial volume problem\, many 3D reconstruction results based on MR images are inaccurate with respect to the details of the geometry of the sulci. Specifically\, when the sulci are on the scale of\, or even smaller than\, the voxel resolution of the MRI\, these models generally really in a binary approximation\, either making the sulcus wider in the model than in reality or eliminating it altogether. In this thesis\, we describe a method for modeling the 3D reconstruction of the brain that may facilitate controlled study of the effect of these approximations. The general approach is to model the brain surface using a spherical harmonic expansion\, then modify the expansion coefficients in an attempt to selectively and smoothly control sulcal width. In the first part of the thesis\, we describe and evaluate an approach in which we experimentally selected two groups of spherical harmonic coefficients within a specified range that could simultaneously affect a chosen sample of the gyri. For the coefficients in the first group\, the widths of all gyri in the sample were increased by enlarging the corresponding coefficients for each spherical harmonic. Conversely\, for each coefficient in the second group\, this adjustment caused the widths of the sampled gyri to decrease simultaneously. We evaluated the method by alternately increasing / decreasing the coefficients in the first group\, and decreasing / increasing those in the second\, by a chosen range of factors\, and observing the effects on the model cortical surface. Experimental results showed that the widths of most of the sulci and gyri were simultaneously adjusted according to the desired effect.\nIn the second part of the thesis\, we tried to build a volume mesh starting from the modified spherical harmonic surfaces. It turned out that this problem was particularly challenging because most of the surface models in our study had self-intersection points. We used a well-known software package for mesh processing\, iso2mesh\, to successfully remove the self-intersection points on all surfaces were removed finally\, but this process seemed to create small holes in the surfaces of the models. Despite these holes\, with a few exceptions\, the widths of most sulci (gyri) were still simultaneously increased (decreased) with the coefficient adjustments. This result provides a direction for further study towards controlled study of the influence of the partial volume problem on modeling of tCS.
URL:https://ece.northeastern.edu/event/ms-thesis-defense-hao-chen/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210420T140000
DTEND;TZID=America/New_York:20210420T150000
DTSTAMP:20260424T095938
CREATED:20210420T175556Z
LAST-MODIFIED:20210420T175556Z
UID:4850-1618927200-1618930800@ece.northeastern.edu
SUMMARY:ECE MS Thesis Defense: Griffin Knipe
DESCRIPTION:MS Thesis Defense: Unifying Performance and Security Evaluation for Microarchitecture Design Exploration \nGriffin Knipe \nLocation: Zoom Link \nAbstract: Computer architects develop microarchitectural features that boost instruction-level parallelism to improve CPU performance. While performance may be improved\, adding new features increases the CPU’s design complexity. This further compounds the effort required to complete design verification. Trustworthy design verification is paramount to microarchitecture design\, as silicon chips cannot easily be patched in the field.\nDespite the best efforts for security verification\, researchers have created transient execution side-channel attacks which can exploit microarchitecture performance features to leak data across ISA-prescribed security boundaries. This motivates the unification of performance evaluation and security verification techniques to ensure that new microarchitectural features are understood from multiple design perspectives.\nThis thesis presents Yori\, a RISC-V microarchitecture simulator that aims to enable computer architects to evaluate microarchitecture performance and security using a single framework. As Yori is a work-in-progress\, this thesis presents the work-to-date\, focusing on a detailed model of the reference microarchitecture and evaluation of the current model accuracy. We describe a viable methodology to interface between the Yori simulator and an existing security verification tool. We conclude the thesis\, laying out a plan to complete this marriage of performance and security.
URL:https://ece.northeastern.edu/event/ece-ms-thesis-defense-griffin-knipe/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210420T133000
DTEND;TZID=America/New_York:20210420T143000
DTSTAMP:20260424T095938
CREATED:20210420T175838Z
LAST-MODIFIED:20210420T175838Z
UID:4852-1618925400-1618929000@ece.northeastern.edu
SUMMARY:ECE PhD Dissertation Defense: Peng Chang
DESCRIPTION:PhD Dissertation Defense: Model-Based Manipulation of Linear Flexible Objects \nPeng Chang \nLocation: Teams Meetings \nAbstract: Manipulation of deformable objects plays an important role in various scenarios such as manufacturing\, service\, healthcare\, and security. Linear flexible objects such as cables\, wires\, and ropes are common in these scenarios. However\, the high dimensionality of the linear flexible objects brings challenges to the modeling and planning in manipulation tasks\, and automatic manipulation of these objects is computationally expensive due to their infinite degrees of freedom in the free spaces. In this dissertation\, we investigate model-based manipulation of linear flexible objects such as cables. We contribute to different models including geometrical and physical models to represent the linear flexible objects. With these models\, we then develop manipulation plans and strategies to achieve the automation of the linear flexible object manipulation tasks in both simulation and real-world. Besides\, we also investigate human-robot collaboration to complete a sample assembly task involving linear flexible object manipulation.
URL:https://ece.northeastern.edu/event/ece-phd-dissertation-defense-peng-chang/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210420T120000
DTEND;TZID=America/New_York:20210420T130000
DTSTAMP:20260424T095938
CREATED:20210414T213115Z
LAST-MODIFIED:20210414T213115Z
UID:4842-1618920000-1618923600@ece.northeastern.edu
SUMMARY:ECE MS Thesis Defense: Ashutosh Singh
DESCRIPTION:MS Thesis Defense: Variation is the Norm: Brain State Dynamics Evoked By Emotional Video Clips \nAshutosh Singh \nLocation: Zoom Link \nAbstract: Past affective neuroscience studies have attempted to identify a “biomarker” or consistent pattern of brain activity (as measured externally using\, for instance\, fMRI) to indicate the presence of a single pre-defined category of emotion (e.g.\, fear) that remains consistent throughout all instances of that category for an individual across contexts and even across individuals. In this thesis\, we investigated variation rather than consistency during emotional experiences. Using fMRI data acquired while individuals watched affect-invoking video clips that have been normed for their evoked emotion categories in prior population studies. Towards this end\, we developed a probabilistic model of the temporal dynamics associated with the hypothetical affect-related brain states\, fitted to the measured brain activity of the participants. We characterized brain states traversed while individuals watched these clips as distinct state occupancy periods between state transitions\, inferred by blood oxygen level-dependent (BOLD) signal patterns captured in fMRI measurements. We found substantial variability in the state occupancy probability distributions across individuals watching the same video\, hence supporting the hypothesis that when it comes to the brain correlates of emotional experience\, variation may indeed be the norm. Studying the mean activation pattern associated with each state\, as well as covariance (in the Gaussian conditional measurement model we assumed)\, we further improve our understanding of the variability between instances of these brain states. Additionally\, we analyzed the presence of potential clusters of brain state trajectories among participants who showed less divergence in their response to each of these videos and checked for their consistency throughout all the video clips.
URL:https://ece.northeastern.edu/event/ece-ms-thesis-defense-ashutosh-singh/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210420T110000
DTEND;TZID=America/New_York:20210420T120000
DTSTAMP:20260424T095938
CREATED:20210412T184906Z
LAST-MODIFIED:20210412T184906Z
UID:4834-1618916400-1618920000@ece.northeastern.edu
SUMMARY:ECE MS Thesis Defense: Yize Li
DESCRIPTION:MS Thesis Defense: Supervised Classification on Deep Neural Network Attack Toolchains \nYize Li \nLocation: Zoom Link \nAbstract: Deep learning\, while an important machine learning technique\, is susceptible to adversarial example attacks. Adversarial examples generated by adding perturbations on clean images/video frames can lead to mis-predictions of deep neural networks. Moreover\, deep learning/machine learning can be used to deceive humans by generating adversarial falsified media e.g.\, deepfake attacks. The thesis work will study the above two attack scenarios\, i.e.\, machine-centric adversary and human-centric adversary\, with targets to fool ML decisions and human decisions\, respectively. We aim to build a generalizable and scalable supervised learning system for classifying attack attributes behind the machine-centric attacks as well as the human-centric attacks. We start from building an integrated Attack Toolchain Library (ATL) with a broad coverage of both machine-centric and human-centric adversaries\, as well as through an integrated user interface for great flexibility and extensibility to serve our downstream tasks. Based on the developed ATL\, we further design a meta-classifier pipeline architecture for predicting attack attributes. The proposed overall meta-classifier shows effectiveness in dealing with false alarms and data distribution shift\, and generalization to both machine-centric and human-centric attacks.
URL:https://ece.northeastern.edu/event/ece-ms-thesis-defense-yize-li/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210420T100000
DTEND;TZID=America/New_York:20210420T110000
DTSTAMP:20260424T095938
CREATED:20210420T180838Z
LAST-MODIFIED:20210420T180838Z
UID:4857-1618912800-1618916400@ece.northeastern.edu
SUMMARY:ECE MS Thesis Defense: Danton Zhao
DESCRIPTION:MS Thesis Defense: LiDAR with a Silicon Photomultiplier for Applications in Adverse Weather \nDanton Zhao \nLocation: Zoom Link \nAbstract: As Light Detection and Ranging (LiDAR) integration becomes more widespread in the field of remote sensing for autonomous navigation\, the impact of degraded visual environments will quickly need to be addressed. The particles responsible for the degradation not only reduce the reflected signal from targets of interest but can also trigger false returns given sufficient density. Of particular interest for solutions to this problem are Geiger-mode avalanche photodiodes\, as these detectors provide high photon sensitivity and high time accuracy with a caveat. In this thesis\, I will be discussing the work that I have done in modeling and addressing artifacts that were generated in the data as a result of using Geiger-mode detectors.
URL:https://ece.northeastern.edu/event/ece-ms-thesis-defense-danton-zhao/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210419T173000
DTEND;TZID=America/New_York:20210419T183000
DTSTAMP:20260424T095938
CREATED:20210303T194634Z
LAST-MODIFIED:20210303T194634Z
UID:4773-1618853400-1618857000@ece.northeastern.edu
SUMMARY:ECE PhD Dissertation Defense: Ilkay Yildiz
DESCRIPTION:PhD Dissertation Defense: Spectral Ranking Regression \nIlkay Yildiz \nLocation: Zoom Link \nAbstract: We consider learning from ranking labels generated as follows: given a query set of samples in a dataset\, a labeler ranks the samples w.r.t.~her preference. Such ranking labels scale exponentially with query set size; most importantly\, in practice\, they often exhibit lower variance compared to class labels. \nWe propose a new neural network architecture based on siamese networks to incorporate both class and comparison labels\, i.e.\, rankings of sample pairs\, in the same training pipeline using Bradley-Terry and Thurstone loss functions. Our architecture leads to a significant improvement in predicting both class and comparison labels\, increasing classification AUC by as much as 35% and comparison AUC by as much as 6% on several real-life datasets. We further show that\, by incorporating comparisons\, training from few samples becomes possible: a deep neural network of 5.9 million parameters trained on 80 images attains a 0.92 AUC when incorporating comparisons. \nFurthermore\, we tackle the problem of accelerating learning over the exponential number of rankings. We consider a ranking regression problem in which we learn Plackett-Luce scores as functions of sample features. We solve the maximum likelihood estimation problem by using the Alternating Directions Method of Multipliers (ADMM)\, effectively separating the learning of scores and model parameters. This separation allows us to express scores as the stationary distribution of a continuous-time Markov Chain. Using this equivalence\, we propose two spectral algorithms for ranking regression that learn shallow regression model parameters up to 579 times faster than the Newton’s method. \nFinally\, we bridge the gap between deep neural networks (DNNs) and efficient spectral algorithms that regress rankings under the Plackett-Luce model. We again solve the ranking regression problem using ADMM\, and thus\, express scores as the stationary distribution of a Markov chain. Moreover\, we replace the standard l_2-norm proximal penalty of ADMM with Kullback-Leibler (KL) divergence. This is a more suitable distance metric for Plackett-Luce scores\, which form a probability distribution\, and significantly improves prediction performance. Our resulting spectral algorithm is up to 175 times faster than siamese networks over four real-life datasets comprising ranking observations. At the same time\, it consistently attains equivalent or better prediction performance than siamese networks\, by up to 26% higher Top-1 Accuracy and 6% higher Kendall-Tau correlation.
URL:https://ece.northeastern.edu/event/ece-phd-dissertation-defense-ilkay-yildiz/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210419T150000
DTEND;TZID=America/New_York:20210419T160000
DTSTAMP:20260424T095938
CREATED:20210420T180007Z
LAST-MODIFIED:20210420T180007Z
UID:4853-1618844400-1618848000@ece.northeastern.edu
SUMMARY:ECE MS Thesis Defense: Kaier Liang
DESCRIPTION:MS Thesis Defense: Rough-Terrain Locomotion and Unilateral Contact Force Regulations With a Multi-Modal Legged Robot \nKaier Liang \nLocation: Zoom Link \nAbstract: The study for legged locomotion has made lots of achievements. However\, the stability of the state-of-the-art bipedal robots are still vulnerable to external perturbation\, cannot negotiate extreme rough terrains\, and cannot directly regulate unilateral contact force.\nThis thesis will introduce a thruster-assisted bipedal walking robot called Harpy. The objective is to integrate the merits of legged and aerial robots in a single platform. The robot’s dynamics is simulated with simplifying assumptions. Furthermore\, this research will show that the employment of thruster allows to stabilize the robot’s frontal dynamics and apply model predictive control (MPC) to jump over obstacles to achieve multi-modal locomotion. In addition\, we will capitalize the thruster actions to demonstrate an optimization-free approach by regulating contact forces using an Explicit Reference Governor (ERG). Then\, we will focus on ERG-based fine-tuning of the joint’s desired trajectories to satisfy unilateral contact force constraints.
URL:https://ece.northeastern.edu/event/ece-ms-thesis-defense-kaier-liang/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210419T090000
DTEND;TZID=America/New_York:20210419T100000
DTSTAMP:20260424T095938
CREATED:20210412T185223Z
LAST-MODIFIED:20210412T185223Z
UID:4836-1618822800-1618826400@ece.northeastern.edu
SUMMARY:ECE PhD Dissertation Defense: Ahmet Oner
DESCRIPTION:PhD Dissertation Defense: Improving the Resilience of the Power Grid \nAhmet Oner \nLocation: Teams Meeting \nAbstract: The power grid constitutes one of the most critical infrastructures that have significant interdependencies with various others such as communication\, transportation\, emergency\, and health-care delivery systems. A disruption in the operation of the power grid may affect the operation of all others in an undesirable manner. Therefore\, improving the resiliency of power grids can also help increase the resiliency of other critical infrastructures. This dissertation presents methods to improve the resiliency of power grids against extreme events and/or system changes. \nFirst\, generation dispatch\, adaptable load shedding strategy\, and pro-active line switching are combined in order to maximize the resiliency of the overall power grid against extreme events. The moving event is monitored\, and the control actions are adjusted accordingly to improve the resilience under changing conditions affected by the natural disaster during its active period. Then\, that study is further extended and made it robust against voltage instability. The details of the methodology and its implementation are presented. \nTo reduce the probability of voltage problems and line flow limit violations\, and to improve power quality\, distributed generators (DG) are placed strategically ahead of the event using outage forecasts based on historical outage data. Therefore\, a possible set of outage scenarios is considered\, and a minimum number of required DG placements are determined to maintain system feasibility for all considered scenarios. \nLastly\, reactive power sources are placed to solve the voltage instability problems\, which are caused by the lack of reactive power in the system. The computational burden of optimal placement problem presents a practical limitation for applying it to very large scale systems considering multi-contingency cases. This part presents a practical and easily implementable solution that will address this limitation.
URL:https://ece.northeastern.edu/event/ece-phd-dissertation-defense-ahmet-oner/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210416T153000
DTEND;TZID=America/New_York:20210416T163000
DTSTAMP:20260424T095938
CREATED:20210412T185358Z
LAST-MODIFIED:20210412T185358Z
UID:4837-1618587000-1618590600@ece.northeastern.edu
SUMMARY:ECE MS Thesis Defense: Mehrshad Zandigohar
DESCRIPTION:MS Thesis Defense: Real-Time Grasp Type Estimation for a Robotic Prosthetic Hand \nMehrshad Zandigohar \nLocation: Zoom Link \nAbstract: For lower arm amputees\, prosthetic hands promise to restore most of physical interaction capabilities. This requires to accurately predict hand gestures capable of grabbing varying objects and execute them timely as intended by the user. Current approaches often rely on physiological signal inputs such as Electromyography (EMG) signal from residual limb muscles to infer the intended motion. However\, limited signal quality\, user diversity and high variability adversely affect the system robustness.\nInstead of solely relying on EMG signals\, our work enables augmenting EMG intent inference with physical state probability through machine learning and computer vision method.\nTo this end\, we: (i) study state-of-the-art deep neural network architectures to select a performant sources of knowledge transfer for the prosthetic hand; (ii) use a dataset containing object images and probability distribution of grasp types as a new form of labeling where instead of using absolute values of zero and one as the conventional classification labels\, our labels are a set of probabilities whose sum is 1. The proposed method generates probabilistic predictions which could be fused with EMG prediction of probabilities over grasps by using the visual information from the palm camera of a prosthetic hand.\nMoreover\, As robotic prosthetic hands are targeted for amputees with the goal of assisting them for their daily life activities\, it is crucial to have a portable and reliable system. Although embedded devices employed in such systems\, provide portability and comfort for the end user\, their limited computational resources comparing to a desktop or server computer impose longer latencies when executing such applications\, making them unreliable and generally impractical to use. Therefore\, it is critical to optimize the aforementioned applications especially DNNs to meet the specified deadline\, resulting in a real-time system. Therefore\, for real-time execution of grasp estimation we propose: (iii) the concept of layer removal as a means of constructing TRimmed Networks (TRNs) that are based on removing problem-specific features of a pretrained network used in transfer learning\, and (iv) NetCut\, a methodology based on an empirical or an analytical latency estimator\, which only proposes and retrains TRNs that can meet the application’s deadline\, hence reducing the exploration time significantly. We demonstrate that TRNs can expand the Pareto frontier that trades off latency and accuracy to provide networks that can meet arbitrary deadlines with potential accuracy improvement over off-the-shelf networks. Our experimental results show that such utilization of TRNs\, while transferring to a simpler dataset\, in combination with NetCut\, can lead to the proposal of networks that can achieve relative accuracy improvement of up to 10.43\% among existing off-the-shelf neural architectures while meeting a specific deadline\, and 27x speedup in exploration time.\nThe proposed methods in this work enable robust and realistic prediction of the grasp type as well as real-time execution of the detection pipeline\, resulting in the improved overall satisfaction of the targeted population.
URL:https://ece.northeastern.edu/event/ece-ms-thesis-defense-mehrshad-zandigohar/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210416T130000
DTEND;TZID=America/New_York:20210416T140000
DTSTAMP:20260424T095938
CREATED:20210414T005310Z
LAST-MODIFIED:20210414T005310Z
UID:4841-1618578000-1618581600@ece.northeastern.edu
SUMMARY:ECE PhD Dissertation Defense: Kai Li
DESCRIPTION:PhD Dissertation Defense: Robust Visual Learning with Limited Labels \nKai Li \nLocation: Zoom Link \nAbstract: The recent flourish of deep learning in various tasks is largely credited to the rich and accessible labeled data. Nonetheless\, massive supervision remains a luxury for many real-world applications: It is costly and time-consuming to collect and annotate a large amount of training data. Sometimes it is even infeasible to get large training datasets because for certain tasks only a few or even no examples are available\, or annotating requires expert knowledge.\nIn this dissertation research\, I investigate techniques systematically addressing the problem of learning with limited labels from the following three aspects. The first aspect is learning to generalize from limited label supervision. I develop few-shot learning algorithms that perform data augmentation in the feature space and that generate task-specific networks based on the limited supervision provided. The second aspect is learning to reuse label supervision from a relevant but different task. I propose domain adaptation algorithms that adapt label supervision from a richly-labeled source domain to a scarcely-labeled target domain with consistency learning\, data augmentation and adversarial learning. The last aspect is learning representations without label supervision. I develop algorithms that learn semantic-rich representations that allow to reliably establish relations among high-dimensional data. This is achieved by explicitly modeling the intrinsic relationship among data points during the representation learning process. \n 
URL:https://ece.northeastern.edu/event/ece-phd-dissertation-defense-kai-li/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210416T093000
DTEND;TZID=America/New_York:20210416T103000
DTSTAMP:20260424T095938
CREATED:20210412T185721Z
LAST-MODIFIED:20210412T185721Z
UID:4838-1618565400-1618569000@ece.northeastern.edu
SUMMARY:ECE MS Thesis Defense: Shanchuan Liang
DESCRIPTION:MS Thesis Defense: Design and Characterization of Flexible Neural Interface Connector for Large-scale Neuronal Recording \nShanchuan Liang \nLocation: Zoom Link \nAbstract: With the increasing demand of the electrically active implantable devices for studying neuroscience\, microelectrode arrays (MEAs) have been widely developed to measure extracellular neuronal activity. Multiple channels MEAs with electrodes embedded are designed to allow coupling time-resolved data simultaneously. In this process\, a well-designed PCB is also essential which use as a bridge to connect MEAs and back-end data acquisition system. This work developed an up to 256-channel flexible neural interface connector for neural signal recording. This thesis aims to introduce the detailed design and implementation procedures of the neural interface connector which consists of MEA\, PCB and amplifier. Considering the contact physics of the connector\, a contact model was established by using COMSOL to address the contact zone and figure out the displacement and pressure on the layer MEAs embedded. The simulation results were used for characterization and optimizing. Robustness tests reveal that the connector is stable up to 500 cycles with high yield. The following in vivo recordings by installed the device on mouse brain validate its excellent performance of recordings of spontaneous single-unit activity of neurons in which spikes in neurons were captured after signal processing.
URL:https://ece.northeastern.edu/event/ece-ms-thesis-defense-shanchuan-liang/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210412T150000
DTEND;TZID=America/New_York:20210412T170000
DTSTAMP:20260424T095938
CREATED:20210412T185039Z
LAST-MODIFIED:20210412T185039Z
UID:4835-1618239600-1618246800@ece.northeastern.edu
SUMMARY:ECE PhD Proposal Review: Murphy Wonsick
DESCRIPTION:PhD Proposal Review: Improving Human Robot Interaction through Extended Reality Technologies \nMurphy Wonsick \nLocation: Teams Link \nAbstract: Recent advancements in robotics have allowed robots to become capable enough to be used in a wide variety of domains\, such as manufacturing\, search-and-rescue\, and space exploration. However\, human-robot interaction with these systems are still primarily achieved using 2D devices\, such and laptops\, tablets\, and/or game controllers despite operating in a 3D world. And although these interfaces can be very capable in operating a robot\, they are often complex and require expert operators as well as extensive training. Extended reality technologies provide an opportunity to create more intuitive human-robot interaction by allowing operators to visualize and interact with 3D data in a 3D environment\, allowing for a more natural interaction. Usage of extended reality technologies in human-robot interaction though are still very limited. In this proposal\, I aim to investigate how to provide better experiences for humans in human-robot interaction using extended reality technologies. Focus will be spent on using virtual reality headset to create supervisory control interfaces for remote robot operation and augmented reality head-mounted displays to help facilitate communication in human-robot shared workspaces. The goal of this work is to move towards more intuitive and easy-to-use interfaces for human-robot interaction.
URL:https://ece.northeastern.edu/event/ece-phd-proposal-review-murphy-wonsick/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210412T150000
DTEND;TZID=America/New_York:20210412T160000
DTSTAMP:20260424T095938
CREATED:20210401T223643Z
LAST-MODIFIED:20210401T223643Z
UID:4822-1618239600-1618243200@ece.northeastern.edu
SUMMARY:ECE PhD Dissertation Defense: Neville Sun
DESCRIPTION:PhD Dissertation Defense: RF Magnetoelectric Devices for Communication\, Sensing\, and Power Electronics \nNeville Sun \nLocation: Zoom Link \nAbstract: A strong magnetoelectric (ME) coupling of layered magnetic/ferroelectric heterostructures can effectively convert energy between electric and magnetic fields. By utilizing strain mediated ME coupling\, it is possible to use an electric field to control magnetic film properties\, such as magnetization\, permeability\, and spin wave. Additionally\, an applied magnetic field can be used to control electric polarization. In this talk\, ME voltage tunable inductors and ME acoustically actuated mechanical antennas/sensors are demonstrated and analyzed with different heterostructure compositions and design considerations for improving device performance.\nThe first part examines a new class of voltage tunable magnetoelectric inductors with textured multiferroic cores consisting of a Metglas/piezoelectric laminate/Metglas composite for MHz adaptive power systems. These inductors demonstrate a large\, instantaneous\, and non-discrete tunable range with a wide operational frequency range from DC to 10 MHz. A tunable inductance range of up to 346% was achieved with an applied electric field of 24 kV/cm. However\, low voltage tunability is miniscule\, typically less than 6% at 30 V applied voltage. By optimizing the anisotropy of magnetoelastic stress\, a 50 um thick PMN-PT slab is shown to improve low voltage tuning by 6 times. These ME tunable inductors with low driving voltage provide adaptability for changing circuit conditions and are ideal for compact/lightweight power systems for electronic warfare and communication systems.\nThe second device of interest is a new MEMS ME antenna/sensor design based on the solidly mounted resonator (SMR) structure. The SMR replaces the freestanding membrane structure of a film-bulk acoustic resonator (FBAR) with a Bragg acoustic reflector for concentrated energy confinement while improving structural integrity and power handling. The antenna radiates using converse ME coupling physics while receiving and sensing EM waves by using direct ME coupling. A unique spin sprayed NiZn ferrite/AlN structure and performance characterization for arrayed resonators are presented. The acoustic resonance in the heterostructure films operates at UHF range for seamless on-chip integration with WiFi\, Bluetooth\, and GPS devices. The robust features of the sub-mm size SMR ME antenna are demonstrated in a miniature aerial drone communication system and provide a possible alternative for biomedical implantables for neurological studies.
URL:https://ece.northeastern.edu/event/ece-phd-dissertation-defense-neville-sun/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210412T090000
DTEND;TZID=America/New_York:20210412T100000
DTSTAMP:20260424T095938
CREATED:20210312T012401Z
LAST-MODIFIED:20210312T012401Z
UID:4780-1618218000-1618221600@ece.northeastern.edu
SUMMARY:ECE PhD Dissertation Defense: Armin Moharrer
DESCRIPTION:PhD Dissertation Defense: Leveraging Structural Properties for Large-Scale Optimization \nArmin Moharrer \nLocation: Zoom Link \nAbstract: Large scale optimization problems abound in data mining\, machine learning\, and system design. We address the challenges posed by such large scale optimization problems by providing efficient optimization algorithms. The scope of studied problems is quite broad; it includes applications such as experimental design\, computing graph distances (dissimilarity scores)\, training auto-encoders\, multi-target regression\, and the design of cache networks. We leverage the structural properties present in these problems\, e.g.\, sparsity or separability. In particular\, we introduce some structural properties under which the Frank-Wolfe algorithm (FW) can be distributed over a cluster of computers. We show that the distributed FW running over 350 workers (CPUs) solves an instance of experimental design problem with 20M variables in 79 minutes\, while the serial implementation takes 48 hours. Furthermore\, we study a variant of FW for the design of cache networks. The problem is NP-hard\, but we achieve a $1-1/e$ approximation ratio\, by optimizing a non-convex relaxation via FW. We also propose a distributed Alternating Direction Method of Multipliers (ADMM) algorithm for computing graph distances. We observe speedups of 153 times when running over a cluster with 448 CPUs\, in comparison with running over 1 CPU\, for graphs with 2.4K nodes. Moreover\, we study applications of ADMM in solving robust variants of risk minimization problems; in these variants we replace the typically chosen mean squared error loss with a general lp norm. We combine model based optimization with ADMM to minimize the resulting non-smooth and non-convex objectives. We show that a stochastic variant of ADMM converges with the rate O(log T/T) and is highly efficient for optimizing the corresponding model functions.
URL:https://ece.northeastern.edu/event/ece-phd-dissertation-defense-armin-moharrer/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210408T110000
DTEND;TZID=America/New_York:20210408T120000
DTSTAMP:20260424T095938
CREATED:20210406T210543Z
LAST-MODIFIED:20210406T210543Z
UID:4829-1617879600-1617883200@ece.northeastern.edu
SUMMARY:ECE Seminar: Mahdi Imani
DESCRIPTION:ECE Seminar: Reinforcement Learning Perspective to Data-Driven and Model-Based Experimental Design \nMahdi Imani \nLocation: Zoom Link \nAbstract: Design and decision-making are pervasive in most practical systems including smart grids\, transportation\, manufacturing\, healthcare\, and smart homes. Accurate system modeling is difficult in most systems/processes due to the complicated system dynamics\, multi-physics and multiple time scales involved in phenomena\, hybrid dynamics across cyber and physical layers\, and various sources of parametric and environmental uncertainties. Design and decision-making in these systems are fraught with choices\, choices that are often expensive\, complex\, and high-dimensional\, with interactions and uncertainties that make them difficult for individuals to reason about. This talk will mainly focus on the speaker’s latest research on providing a new unified reinforcement learning perspective for model-based and data-driven experimental design to enable scalable\, efficient\, and reliable design and decision-making under various sources of uncertainty. \nBio: Mahdi Imani is an Assistant Professor in the Department of Electrical and Computer Engineering at the George Washington University. He received his Ph.D. degree in Electrical and Computer Engineering from Texas A&M University in 2019\, and his M.Sc. degree in Electrical Engineering and his B.Sc. degree in Mechanical Engineering\, both from the University of Tehran in 2014 and 2012. His research interests include Machine Learning\, Control Theory\, and Signal Processing\, with a wide range of applications from computational biology to cyber-physical systems. He has been elevated to IEEE Senior Member grade in 2021. He is also the recipient of multiple awards\, including NSF SCH Aspiring PI Awardee in 2020 and 2021\, IBM Research Almaden Distinguished Speaker in 2019\, the Association of Former Students Distinguished Graduate Student Award for Excellence in Research-Doctoral in 2019\, the Best Ph.D. Student Award in ECE department and a single finalist nominee of ECE department for the Outstanding Graduate Student Award in the college of engineering at Texas A&M University in 2018\, and the best paper finalist award from the 49th Asilomar Conference on Signals\, Systems\, and Computers\, 2015.
URL:https://ece.northeastern.edu/event/ece-seminar-mahdi-imani/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210407T140000
DTEND;TZID=America/New_York:20210407T150000
DTSTAMP:20260424T095938
CREATED:20210323T213831Z
LAST-MODIFIED:20210323T213831Z
UID:4809-1617804000-1617807600@ece.northeastern.edu
SUMMARY:ECE PhD Dissertation Defense: Vikrant Shah
DESCRIPTION:PhD Dissertation Defense: Visual Navigation Applications in Low Contrast Environments: Multi Sensor Iceberg Mapping \nVikrant Shah \nLocation: Zoom Link \nAbstract: Most approaches to visual navigation make multiple assumptions about the scenes being imaged. There are implicit assumptions about the scene being predominantly static and the availability of well illuminated\, texture rich\, objects in the scene. In some cases these assumptions severely limit or eliminate the full applicability of visual Simultaneous Localization and Mapping (SLAM) and Structure from Motion (SfM) methodologies. This dissertation attempts to address problems where the assumptions of static scenes and texture rich objects are not valid. Motivated by the application of mapping rotating and translating icebergs\, we propose a system level solution for addressing the problem of mapping large\, low contrast\, moving targets with slow but complicated dynamics. \nOur approach leverages the complementary nature of multiple sensing modalities and utilizes a rigidly coupled combination of a subsurface multibeam sonar (a line scan sensor) and an optical camera (an area scan sensor). This allows the system to exploit the optical camera information to perform iceberg relative navigation\, which can be directly used by the multibeam sonar to map the iceberg underwater. To compensate for the effect of low contrast we conducted an in-depth analysis of features detectors and descriptors on end-to-end SfM algorithms to demonstrate and understand how methodologies such as Contrast Limited Adaptive Histogram Equalization (CLAHE) and Zernike Moment descriptors help improve the overall accuracy in these challenging applications. \nWe merge these approaches into an algorithmic framework that allows us to compute the scale of the navigation solution and iceberg centric navigation corrections. These corrections can then be used for accurate iceberg reconstructions. This enables a quantitative analysis of our iceberg mapping efforts including volume estimation and change detection. \nWe successfully demonstrate our approach on real field data from three of the icebergs surveyed multiple times during the 2018 and 2019 campaigns to the Sermilik fjord in Eastern Greenland. Availability of iceberg mounted Global Navigation Satellite System (GNSS) observations during these research expeditions also allowed for a comparison of this approach against ground truth\, providing additional confidence in the systems level mapping efforts. The accuracy of the reconstructions is demonstrated by estimating iceberg volumes\, calculating their ablation rates\, and performing change detection at a granular scale.
URL:https://ece.northeastern.edu/event/ece-phd-dissertation-defense-vikrant-shah/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210406T100000
DTEND;TZID=America/New_York:20210406T110000
DTSTAMP:20260424T095938
CREATED:20210401T223518Z
LAST-MODIFIED:20210401T223518Z
UID:4821-1617703200-1617706800@ece.northeastern.edu
SUMMARY:ECE PhD Proposal Review: Subhramoy Mohanti
DESCRIPTION:PhD Proposal Review: Distributed Data and Energy Beamforming with Unmanned Vehicles for Wireless IoT : A Systems Perspective \nSubhramoy Mohanti \nLocation: Teams Meeting \nAbstract: The pervasive deployment of the wireless Internet of Things (IoT) has given rise to heterogeneous sensors and small form-factor computing devices in homes\, offices\, public spaces\, manufacturing floors\, among others. Such large number of connected devices require (i) simple ways of charging\, so that they remain operationally available\, and (ii) effective ways of sharing wireless spectrum\, so that they continue to transmit and receive data amidst competing and interfering signals. This thesis focuses on the link and physical layer of the protocol stack to enable distributed beamforming as a key enabler for these two objectives. Specifically\, we experimentally demonstrate how beamforming capability can address both wireless power transfer (WPT) needs and resilient communication in interference-challenged environments.\nThis thesis proposes a method for accessing and sharing the wireless channel for both regular data communication and WPT. This is the first work that accomplishes these dissimilar tasks within the constraints of the standard compliant IEEE 802.11 protocol\, resulting in a practical and so called ‘WiFi-friendly Energy Delivery’ (WiFED). First\, WiFED exploits the IEEE 802.11 supported protocol features to request energy and for energy transmitters to participate in energy transfer via beamforming. Second\, it devises a controller-driven bipartite matching algorithm\, assigning appropriate number of energy transmitters to sensors for efficient energy delivery. Thirdly\, it detects outlier sensors\, which have limited power reception from static energy transmitters and utilizes mobile energy transmitters to satisfy their charging cycles.\nFrom a communication-only perspective that relies on distributed beamforming\, this thesis presents AirBeam\, a software-based approach that runs on Unmanned Aerial Vehicles (UAVs) to deliver on-demand data to sensors deployed in infrastructure constrained environments. We first show why this problem is difficult given the continuous hovering-related channel fluctuations\, synchronizing the distributed transmit streams without a wired clock reference\, the need to ensure timely feedback from the ground receiver due to the channel coherence time\, and the size\, weight\, power\, and cost (SWaP-C) constraints for UAVs. This work is extended further to consider realistic traffic patterns and packet arrival thresholds\, involving dynamic grouping of transmitters to beamform towards target receivers at any given time. Again\, we evaluate outcome both experimentally and in a virtual environment in Colosseum\, the world’s largest RF emulator.\nSince beamforming requires the action of multiple devices not directly connected to each other by wire\, we introduce a security framework called AirID\, which identifies authorized beamforming UAVs by learning their so called ‘RF fingerprints’. This step requires applying deep learning techniques on their received signals\, with the goal of identifying discriminative features introduced by the transmitter due to process variations. Our approach involves intentionally inserting ‘signatures’ in the signals from each known UAV\, which are detected through a deep convolutional neural network (CNN) at the physical layer\, without affecting the ongoing UAV data communication process.\nIn the proposed work\, we will explore optimized placement of UAVs\, while also considering battery limits\, to enhance beamforming performance. We will validate these outcomes in a testbed of 4-5 UAVs.
URL:https://ece.northeastern.edu/event/ece-phd-proposal-review-subhramoy-mohanti/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210326T130000
DTEND;TZID=America/New_York:20210326T140000
DTSTAMP:20260424T095938
CREATED:20210322T180141Z
LAST-MODIFIED:20210322T180141Z
UID:4805-1616763600-1616767200@ece.northeastern.edu
SUMMARY:ECE Seminar: Sara Dean
DESCRIPTION:ECE Seminar: Reliable Machine Learning in Feedback Systems \nSara Dean \nLocation: Zoom Link \nAbstract: Machine learning techniques have been successful for processing complex information\, and thus they have the potential to play an important role in data-driven decision-making and control. However\, ensuring the reliability of these methods in feedback systems remains a challenge\, since classic statistical and algorithmic guarantees do not always hold. In this talk\, I will provide rigorous guarantees of safety and discovery in dynamical settings relevant to robotics and recommendation systems. I take a perspective based on reachability\, to specify which parts of the state space the system avoids (safety) or can be driven to (discovery). For data-driven control\, we show finite-sample performance and safety guarantees which highlight relevant properties of the system to be controlled. For recommendation systems\, we introduce a novel metric of discovery and show that it can be efficiently computed. In closing\, I discuss how the reachability perspective can be used to design social-digital systems with a variety of important values in mind. \nBio: Sarah is a PhD candidate in the Department of Electrical Engineering and Computer Science at UC Berkeley\, advised by Ben Recht. She received her MS in EECS from Berkeley and BSE in Electrical Engineering and Math from the University of Pennsylvania. Sarah is interested in the interplay between optimization\, machine learning\, and dynamics in real-world systems. Her research focuses on developing principled data-driven methods for control and decision-making\, inspired by applications in robotics\, recommendation systems\, and developmental economics. She is a co-founder of a transdisciplinary student group\, Graduates for Engaged and Extended Scholarship in computing and Engineering\, and the recipient of a Berkeley Fellowship and a NSF Graduate Research Fellowship.
URL:https://ece.northeastern.edu/event/ece-seminar-sara-dean/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20210326T120000
DTEND;TZID=America/New_York:20210326T130000
DTSTAMP:20260424T095938
CREATED:20210323T180009Z
LAST-MODIFIED:20210323T180009Z
UID:4808-1616760000-1616763600@ece.northeastern.edu
SUMMARY:ECE PhD Dissertation Defense: Mo Han
DESCRIPTION:PhD Dissertation Defense: Human Grasp Intent Inference and Multimodal Control in Prosthetic Hands \nMo Han \nLocation: Zoom Link \nAbstract: Upper limb and hand functionality is critical to many activities of daily living and the amputation of one can lead to significant functionality loss for individuals. From this perspective\, advanced prosthetic hands of the future are anticipated to benefit from improved shared control between a robotic hand and its human user\, but more importantly from the improved capability to infer human intent from multimodal sensor data to provide the robotic hand perception abilities regarding the operational context. Such multimodal data may be collected from various environment sensors such as camera providing visual information\, as well as easily-accessed human physiologic sensors including electromyographic (EMG) sensors. A fusion methodology for environmental state and human intent estimation can combine these sources of evidence in order to help prosthetic hand motion planning and control. \nAs part of a multi-disciplinary project\, i.e. HANDS project\, which aims at designing a robotic hand as an upper limb prosthetic device\, we developed two independent prosthetic control systems (HANDS V1 and HANDS V2) integrating multimodal sources of EMG and visual evidences into the control loop. Multiple grasps required for activities of daily living can be performed by both robotic systems which were developed in a lighter and cheaper semi-autonomous manner. The HANDS V1 system was first developed to provide an easy and convenient prosthesis with a portable EMG armband and a built-in palm camera\, and hereafter the HANDS V2 was constructed as an upgraded solution of HANDS V1 to achieve more difficult tasks with more identified grasp types\, more EMG channels and more complicated visual information involved. Both systems depend on multimodal signals from EMG and vision\, where the EMG could reflect the physiologic features related to user intents\, while the robustness and adaptability to different users could be retained by the visual information relying more on surrounding environments. We collected two datasets for the initialization of each system\, and the developments of the EMG-control\, visual-control\, and joint-control algorithms were conducted for both systems. We exploited efficient computer vision and physiological signal processing methodologies to decrease the system complexity as well as improve the user comfort\, in order to provide smarter and cheaper prosthetic hands to the audience. Online experiments were executed and evaluated on both HANDS V1 and HANDS V2 systems\, implemented by the Robot Operating System (ROS) system.
URL:https://ece.northeastern.edu/event/ece-phd-dissertation-defense-mo-han/
END:VEVENT
END:VCALENDAR