BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Department of Electrical & Computer Engineering - ECPv6.16.2//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: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
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
DTSTART:20230312T070000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:20231105T060000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20220707T130000
DTEND;TZID=America/New_York:20220707T140000
DTSTAMP:20260623T035621
CREATED:20221103T183228Z
LAST-MODIFIED:20221103T183228Z
UID:5908-1657198800-1657202400@ece.northeastern.edu
SUMMARY:Sara Garcia Sanchez's PhD Dissertation Defense
DESCRIPTION:“Learning and Shaping the Wireless Environment: An Integrated View of Sensing\, Computing and Communication” \nAbstract: \nThe explosive growth in Internet of Things (IoT) deployments and anticipated data volumes that will be generated within future autonomous devices require collecting and processing large amounts of data\, generally transmitted over the wireless channel. Rigid infrastructure deployment that does not adapt to the changing wireless environment is not well suited to handle these new demands. To address this limitation\, this dissertation takes a hands-on approach to equip communication systems with technology to learn from\, interact with and actuate within the environment. Specifically\, we build (i) accurate physics-based predictive models and multimodal sensing techniques to gain awareness of the existing channel\, as well as (ii) novel multidisciplinary approaches to intelligently shape the wireless channel towards enhancing the communication link. \nWe first prove that combining wireless channel modeling\, multimodal sensing and robotics provides significant link performance gains. To this extent\, we adopt a systems approach to study how millimeter wave (mmWave) radio transmitters on Unmanned Aerial Vehicles (UAVs) provide high throughput links under typical hovering conditions. Based on sensing and modeling efforts\, we propose techniques to exploit the information contained in the spatial and angular domains of empirically collected data from GPS\, cameras and RF signals. We demonstrate how to mitigate the impact of hovering by (i) selecting near-to-optimum transmission parameters as compared to the mmWave standard IEEE 802.11ad\, and (ii) proposing corrective coordinated actions at the UAVs from the robotic controls. These methods achieve mmWave beam-tracking and robust link deployment under event(s) impacting link performance\, such as hovering or blockage in the light of sight between transmitter and receiver.\nFinally\, we experimentally demonstrate how the wireless environment can be interactively shaped through the use of Reconfigurable Intelligent Surfaces (RIS). First\, we propose AirNN\, a system capable of partially offloading computation into the wireless domain by realizing analog convolutions with over-the-air computation. We demonstrate that such computation is accurate enough to substitute its digital equivalent in a Convolutional Neural Network (CNN). Second\, we propose a RIS-based spatio-temporal signal modification approach for channel hardening (i.e.\, ensure low power fluctuations in the received signal) in a Single-Input Single-Output link and under rich multipath\, which is common for IoT 5G+ deployments. We prove that our approach achieves channel hardening similar to a classical Single-Input Multiple-Output (SIMO) system while only using a single antenna element at the receiver end. \nAll the above theoretical advances are validated with rigorous analysis and experimentation. \nCommittee: \nProf. Kaushik Chowdhury (Advisor) \nProf. Stefano Basagni \nProf. Josep Jornet
URL:https://ece.northeastern.edu/event/sara-garcia-sanchezs-phd-dissertation-defense/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20220707T130000
DTEND;TZID=America/New_York:20220707T140000
DTSTAMP:20260623T035621
CREATED:20221103T183515Z
LAST-MODIFIED:20221103T183515Z
UID:5912-1657198800-1657202400@ece.northeastern.edu
SUMMARY:Alexandria Will-Cole's PhD Proposal Review
DESCRIPTION:“Morphology\, Magnetism\, and Transport in Nanomaterials and Nanocomposites” \nAbstract: \nMagnetic thin film materials and bilayer composites enable unprecedented new applications\, ranging from magnetic-based microelectromechanical systems (magnetoelectric sensors\, ultracompact magnetoelectric antennas\, etc.)\, terahertz emitters\, to spin-orbit-torque driven magnetic memories. Here we focus on two subdisciplines within magnetics – magnetoelectrics and spintronics heterostructures. \nThe first aspect of the talk is focused on magnetoelectrics. Strain-mediated magnetoelectric coupling (i.e.\, voltage/electric field control of magnetism\, or magnetic field control of electrical polarization) in bilayer composites has received heightened attention in the research community for applications in memory\, motors\, sensors\, communication etc. The composite ME effect is dependent on the magnetostrictive effect (magnetic-mechanical coupling) and the piezoelectric effect (electrical-mechanical coupling)\, and therefore to improve the composites each constituent phase needs to be optimal. Here we demonstrate the feasibility of machine learning\, specifically Bayesian Optimization methods\, to optimize ferromagnetic materials\, specifically (Fe100−y Gay)1−xBx (x=0–21 & y=9–17) and (Fe100−y Gay)1−xCx (x=1–26 and y=2–18) to demonstrate optimization of structure-property relationships\, specifically the compositional effect on magnetostriction and ferromagnetic resonance linewidth. Following the materials optimization study\, we present voltage control of ultrafast demagnetization in ME heterostructure of (Fe81Ga19)88B12/ Pb(Mg1/3Nb2/3)O3–PbTiO3. Previous studies implement multiple strategies to tune ultrafast demagnetization namely via the laser pump wavelength\, fluence\, polarization\, and pulse duration as these control the total absorbed energy into the film. Here we present an alternate strategy to tune ultrafast demagnetization with application of an electric field in the ME heterostructure to induce magnetic axis rotation. Additionally\, we studied magnetic anisotropy changes and E-field tuning behavior following ultrafast demagnetization. \nThe second aspect of this talk is focused on spintronics heterostructures\, namely ferromagnetic (FM)/topological insulator (TI) or ferrimagnetic insulator (FI)/topological insulator (TI) bilayer composites\, and TI sputter growth and characterization. Bilayer FM/TI and FI/TI heterostructures are promising for spintronic memory applications due to their low switching energy and therefore power efficiency. TIs have been grown with molecular beam epitaxy (oriented\, epitaxial films) and RF magnetron sputtering (amorphous to crystalline oriented films) and have demonstrated large spin-to-charge conversion efficiencies. However\, the reactivity of TIs with FM films is often overlooked in the spin-orbit-torque literature\, even though there are reports that it is energetically favorable for topological insulators to react with metals and form interfacial layers. Here we present the interfacial reaction and antiferromagnetic phase formation between MBE-grown Sb2Te3 and sputtered Ni80Fe20 films. Since FM/TI interfaces are highly reactive and form novel interfacial phases\, which can encourage spin memory loss\, it is critical to explore heterostructures with cleaner interfaces. Recently\, we synthesized chemically stable Y3Fe5O12/Bi2Te3 films\, which should have a chemically sharp interface. We present preliminary structural and magnetic characterization\, followed by proposed experiments to study proximity induced magnetization in these bilayer composites. Concurrent to our investigation spintronic heterostructures\, we seek to optimize sputter deposition of TIs. However\, sputtering TIs requires enhanced control over defects/stoichiometry as these influence bulk transport. We present preliminary results and propose experiments to elucidate structure-transport relationships\, such that we can provide strategies to controllably suppress bulk conduction to access topologically protected surface states. \nCommittee:\nProf. Nian X. Sun (advisor)\nProf. Don Heiman (co-advisor)\nProf. Yongmin Liu\nDr. A. Gilad Kusne\nDr. Todd Monson
URL:https://ece.northeastern.edu/event/alexandria-will-coles-phd-proposal-review/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20220711T100000
DTEND;TZID=America/New_York:20220711T230000
DTSTAMP:20260623T035621
CREATED:20221103T183012Z
LAST-MODIFIED:20221103T183012Z
UID:5904-1657533600-1657580400@ece.northeastern.edu
SUMMARY:Bengisu Ozbay's PhD Proposal Review
DESCRIPTION:“Fast Identification via Subspace Clustering and Applications to Dynamic and Geometric Scene Understanding” \nAbstract: \nMore and more data is needed in order to build new machine learning and computer vision techniques. Using human operators to identify these vast datasets would be too expensive\, hence the use of unsupervised learning has grown more common. Piecewise linear or affine models can be used in a broad range of applications connected to system identification and computer vision.\nIn this proposal\, we suggest an efficient method that only requires singular value decomposition of matrices whose size is unaffected by the total number of points. This method only has to be performed (number of clusters) times. We discovered that it is feasible to find the polynomials that represent the hyperplanes by doing a singular value decomposition (SVD) on the empirical moments matrix containing the data. In this approach\, the notion of using polynomials and Christoffel functions to conduct SVDs in order to partition data into sets\, each of which originates from a different cluster\, is central. Data may be segmented and then the parameters of each group can be extracted using application-specific techniques. In particular\, the problems that are taken into consideration in this proposal include identification of Auto-regressive with Extra Input (SARX) models\, affine linear subspace clustering\, two-view motion segmentation\, and identification of a group of nonlinear systems known as Wiener systems.\nThis proposal is structured as follows: to begin with\, we offer a semi-algebraic clustering framework for locating reliable subsets from the data\, which belongs in a union of varieties and segments the data sequentially using Christoffel polynomials. We employ this strategy for switched system identification and affine subspace clustering challenges. In both instances\, the data resides in linear affine varieties. To expand the given approach beyond linear affine arrangements\, we reformulate it for quadratic surfaces and further apply it to the two-view motion segmentation task. Finally\, using this suggested semi-algebraic formulation\, we are able to detect a class of nonlinearities\, namely Wiener systems with an even nonlinearity\, which is indeed an NP-hard issue.
URL:https://ece.northeastern.edu/event/bengisu-ozbays-phd-proposal-review/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20220712T153000
DTEND;TZID=America/New_York:20220712T163000
DTSTAMP:20260623T035621
CREATED:20221103T183319Z
LAST-MODIFIED:20221103T183319Z
UID:5910-1657639800-1657643400@ece.northeastern.edu
SUMMARY:Zulqarnain Qayyum Khan's PhD Dissertation Defense
DESCRIPTION:“Interpretable Machine Learning for Affective Psychophysiology and Neuroscience” \nAbstract: \nIn this thesis\, we leverage existing Machine Learning (ML) models where appropriate and develop novel models to advance the understanding of affective psychophysiology and neuroscience. Additionally\, considering the increased use of ML as a toolbox\, we highlight underlying assumptions and limitations of basic ML methods to help better contextualize the conclusions drawn from application of ML in this domain. Similarly\, given the increasingly opaque ML models\, the resulting rise of methods to explain these models\, and the importance of explainability to interdisciplinary research\, we investigate theoretical properties of these explainers.\nAffective pyschophysiology research typically uses supervised analyses which leave little room for exploration. Studies of motivated performance tasks often focus on two states of threat and challenge\, exhibiting somewhat inconsistent physiological properties. Using unsupervised analysis of physiology data\, we find evidence for the presence of a third state for the first time\, that may help explain these inconsistencies. Similarly\, prototypical view of emotion often searches for consistency and specificity\, as opposed to constructionist account of emotion which proposes emotion categories as populations of situation-specific variable instances. In results supportive of this constructionist view\, we find large variability in both the number and nature of clusters in unsupervised analyses of ambulatory physiological data. Similarly\, in functional neuroimaging a largely unsolved challenge is to develop models that appropriately account for the commonalities and variations among participants and stimuli\, scale to large amounts of data\, and reason about uncertainty in an unsupervised manner. Such models are needed to investigate important neuroscientific phenomena such as individual variation and degeneracy. We develop Neural Topographic Factor Analysis (NTFA)\, a novel ML model for fMRI data with a deep generative prior that teases apart participant and stimulus driven variation and commonalities\, and demonstrate its potential in investigating individual variation and degeneracy.\nWe further utilize this interdisciplinary research experience to shed light on assumptions and limitations of some of the basic ML methods commonly used in the sciences (especially psychological science). These methods are often used as software packages. We argue that researchers need to be more mindful of their underlying assumptions when drawing conclusions. Along the same lines\, ML methods themselves are becoming increasingly blackbox\, making it harder to reason about underlying assumptions. This has led to an increased focus on explainers\, which provide interpretability to ML methods that is critical for interdisciplinary research. The theoretical properties of these explainers\, however\, remain understudied. We further the research in this direction by defining explainer astuteness as a measure of robustness and theoretically demonstrate that smooth classifiers lend themselves to more astute explanations. \nCommittee: \nProf. Jennifer Dy (Advisor)\nProf. Lisa Feldman Barrett\nProf. Dana Brooks\nProf. Karen Quigley\nProf. Octavia Camps
URL:https://ece.northeastern.edu/event/zulqarnain-qayyum-khans-phd-dissertation-defense/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20220713T110000
DTEND;TZID=America/New_York:20220713T120000
DTSTAMP:20260623T035621
CREATED:20221103T183115Z
LAST-MODIFIED:20221103T183115Z
UID:5906-1657710000-1657713600@ece.northeastern.edu
SUMMARY:Leonardo Bonati's PhD Dissertation
DESCRIPTION:“Softwarized Approaches for the Open RAN of NextG Cellular Networks” \nAbstract: \nThe 5th and 6th generations of cellular networks (5G and 6G)\, also known as NextG\, will bring unprecedented flexibility to the wireless cellular ecosystem. Because of a typically closed and rigid market\, the telco industry has incurred high costs and non-trivial obstacles for delivering new services and functionalities that satisfy the requirements and the demands of NextG networks. To break this trend the industry is now moving toward open architectures based on softwarized approaches\, which afford network operators flexible control and unprecedented adaptability to heterogeneous conditions\, including traffic and application requirements. Now\, by simply expressing a high-level intent\, operators will be able to instantiate bespoke services on-demand on a generic hardware infrastructure\, and to adapt such services to the current network conditions. Through disaggregation\, network elements will split their functionalities across multiple components—possibly provided by different vendors—interconnected through well-defined open interfaces. The separation of control functions from the hardware fabric\, and the introduction of standardized control interfaces\, will ultimately enable the definition and use of softwarized control loops\, which will bring embedded intelligence and real-time analytics to effectively realizing the vision of autonomous and self-optimizing networks.\nThis dissertation work focuses on the design\, prototyping and experimental evaluation of softwarized approaches for the Open Radio Access Network (RAN) of NextG cellular networks. We analyze the architectural enablers\, challenges\, and requirements for a programmatic zero-touch control of the very many network elements and propose practical solutions for its realization. We prototype solutions by leveraging open-source software implementations of cellular protocol stacks and frameworks\, and heterogeneous virtualization technologies\, including the srsRAN and OpenAirInterface cellular implementations\, and the O-RAN framework. The contributions of this work include (i) the first demonstration of O-RAN data-driven control loops in a large-scale experimental testbed using open-source\, programmable RAN and RAN Intelligent Controller (RIC) components through xApps of our design; (ii) CellOS\, a zero-touch cellular operating system that automatically generates and executes distributed control programs for simultaneous optimization of heterogeneous control objectives on multiple network slices starting from a high-level intent expressed by the operators; (iii) OpenRAN Gym\, the first publicly-available research platform for the design\, prototyping\, and experimentation at scale of data-driven O-RAN solutions\, and (iv) OrchestRAN\, a network intelligence orchestration framework for Open RAN that automates the deployment of data-driven inference and control solutions. The effectiveness of our solutions in achieving superior control and performance of the RAN is demonstrated at scale on state-of-the-art experimental facilities\, including software-defined radio-based laboratory setups and open access experimental wireless platforms\, such as Colosseum\, Arena\, and the POWDER and COSMOS platforms from the U.S. PAWR program.
URL:https://ece.northeastern.edu/event/leonardo-bonatis-phd-dissertation/
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:20220721T090000
DTEND;TZID=America/New_York:20220721T100000
DTSTAMP:20260623T035621
CREATED:20221103T182650Z
LAST-MODIFIED:20221103T182650Z
UID:5900-1658394000-1658397600@ece.northeastern.edu
SUMMARY:Abhimanyu Sheshashayee's PhD Dissertation Defense
DESCRIPTION:Location: 532 ISEC \n“Wake-up Radio-enabled Wireless Networking: Measurements and Evaluation of Data Collection Techniques in Static and Mobile Scenarios” \nAbstract: \nMulti-hop wireless networks such as Wireless Sensor Networks and in general\, networks without the support of a fixed infrastructure\, which enable most applications of the Internet of Things\, are comprised of wirelessly communicating nodes that are often powered by batteries. In many relevant scenarios—ranging from precision agriculture to oceanographic surveillance—it is inconvenient or impossible to replenish or replace the energy systems of these nodes\, which limits the operational lifespan of the network. One of the most significant sources of power consumption comes from idle listening on the node’s wireless transceiver (main radio). This consumption can be reduced by endowing the nodes with Wake-up Radio (WuR) technology: Nodes keep their main radio off while listening for a signal via an ultra-low-power auxiliary radio used only for wake-up purposes. When the appropriate signal is received\, the node turns its main radio on\, conducts the necessary exchange of packets\, and then turns off its main radio. This strategy allows for a considerable reduction in power consumption.This dissertation investigates data collection approaches that leverage WuR technology to maximize the lifespan of multi-hop networks for data gathering\, via routing and via a Mobile Data Collector (MDC). We analyze contemporary WuR technology\, isolating the main criticalities of the state-of-the-art\, including range and data rates. We use WuR prototypes with highly desirable characteristics to conduct experiments to measure effective communication ranges\, in both static and mobile scenarios. We then examine the application of WuR technology to data collection based on multi-hop routing. We devise new techniques and evaluate the effects of different WuR characteristics on the performance of routing\, considering for the first time what the network performance could be if we could overcome the limitation of current WuRs.The culmination of this dissertation focuses on mobile data collection protocols and approaches. We conduct a comprehensive survey of mobile data collection studies and protocols. We develop a robust taxonomy to set the framework for our analyses of various methodologies and elements of mobile data collection. Guided by our review of the literature\, we define two collection strategies: a simple naïve strategy\, and a novel AI-driven adaptive strategy. Both strategies leverage WuR technology to minimize the amount of time SNs remain awake. Considering both duty cycle-based and WuR based scenarios\, we conduct extensive experiments with a quad-rotor UAV-MDC and a network of WuR-enabled wireless sensor motes. We replicate these experiments in our simulator\, informed by the parameters and characteristics observed in our real-world experiments. Having validated our simulations\, we proceed to execute exhaustive simulation-based experiments. We evaluate the effects of scale (namely\, network size and deployment region size) on the performance of the naïve and adaptive strategies\, and we contrast the energy efficiency. The WuR-based scenarios experience considerably lower time spent awake\, which gives rise to longer network lifespan. The adaptive strategy minimizes the time taken for each collection cycle\, thereby reducing the amount of time spent awake in the duty cycle-based scenarios. The adaptive strategy also results in a noticeable reduction in both the awake duration and latency for the WuR-based scenarios. \nCommittee: \nProfessor Stefano Basagni (Advisor)Professor Kaushik ChowdhuryProfessor Tommaso Melodia
URL:https://ece.northeastern.edu/event/abhimanyu-sheshashayees-phd-dissertation-defense/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20220721T143000
DTEND;TZID=America/New_York:20220721T153000
DTSTAMP:20260623T035621
CREATED:20221103T182855Z
LAST-MODIFIED:20221103T182855Z
UID:5902-1658413800-1658417400@ece.northeastern.edu
SUMMARY:Siyue Wang's PhD Dissertation Defense
DESCRIPTION:“Towards Robust and Secure Deep Learning Models and Beyond” \nAbstract: \nModern science and technology witness the breakthroughs of deep learning during the past decades. Fueled by the rapid improvements of computational resources\, learning algorithms\, and massive amounts of data\, deep neural networks (DNNs) have played a dominant role in many real world applications. Nonetheless\, there is a spring of bitterness mingling with this remarkable success – recent studies have revealed the limitations of DNNs which raise safety and reliability concerns of its widespread usage: 1) the robustness of DNN models under adversarial attacks and facing instability problems of edge devices\, and 2) the protection and verification of intellectual properties of well-trained DNN models.In this dissertation\, we first investigate how to build robust DNNs under adversarial attacks\, where deliberately crafted small perturbations added to the clean inputs can lead to wrong prediction results with high confidence. We approach the solution by incorporating stochasticity into DNN models. We propose multiple schemes to harden the DNN models when facing adversarial threats\, including Defensive Dropout (DD)\, Hierarchical Random Switching (HRS)\, and Adversarially Trained Model Switching (AdvMS). Besides\, we also propose a stochastic fault-tolerant training scheme that can generally improve the robustness of DNNs when facing the instability problem on DNN accelerators without focusing on optimizations for individual devices.The second part of this dissertation focuses on how to effectively protect the intellectual property for DNNs and reliably identify their ownership. We propose Characteristic Examples (C-examples) for effectively fingerprinting DNN models\, featuring high-robustness to the well-trained DNN and its derived versions (e.g. pruned models) as well as low-transferability to unassociated models. To better perform functionality verification of DNNs implemented on edge devices for on-device inference applications\, we also propose Intrinsic Examples. Intrinsic Examples as fingerprinting of DNN can detect adversarial third-party attacks that embed misbehaviors through re-training. The generation process of our fingerprints does not intervene with the training phase and no additional data are required from the training/testing set. \nCommittee: \nProf. Xue Lin (Advisor)Prof. Yunsi FeiProf. Yanzhi Wang
URL:https://ece.northeastern.edu/event/siyue-wangs-phd-dissertation-defense/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20220727T143000
DTEND;TZID=America/New_York:20220727T153000
DTSTAMP:20260623T035621
CREATED:20221103T182600Z
LAST-MODIFIED:20221103T182600Z
UID:5898-1658932200-1658935800@ece.northeastern.edu
SUMMARY:Kimia Shayestehfard's PhD Proposal Review
DESCRIPTION:“Permutation Invariant Graph Learning” \nAbstract:\nGraphs are widely used in many areas such as biology\, engineering\, and social sciences to model sets of objects and their interactions and relationships. Tasks addressed by applying machine learning to graphs\, known as graph learning\, include node and graph classification\, edge prediction\, transfer learning\, and generative modeling/distribution sampling\, to name a few.\nDue to high prevalence and multitude of applications of graphs across different fields\, graph neural networks have been developed in the past few years. Graph neural networks have shown tremendous success at producing node embeddings that capture structural and relational information of a graph and are discriminative for downstream tasks. However\, graph learning algorithms still deal with a major challenge\, namely\, the lack of permutation invariance: In a dataset of sampled graphs\, nodes may be ordered arbitrarily\, and aligning them is combinatorial and computationally expensive. Moreover\, many graph distance algorithms do not satisfy metric properties\, which can significantly hamper the fidelity of the downstream tasks. In this work we address the challenges posed by permutation invariance via combining fast and tractable metric graph alignment methods with graph neural networks. We propose a tractable\, non-combinatorial method for solving the graph transfer learning problem by combining classification and embedding losses with a continuous\, convex penalty motivated by tractable graph distances. We demonstrate that our method successfully predicts labels across graphs with almost perfect accuracy; in the same scenarios\, training embeddings through standard methods leads to predictions that are no better than random. Furthermore\, we propose a framework that combines fast and tractable graph alignment methods with a family of deep generative models and are thus invariant to node permutations. These models can be learned by solving convex optimization problems. Our experiments demonstrate that our models successfully learn graph distributions\, outperforming competitors by at least 66% in two relevant performance scores and improve the computation time up to 20 times over existing metric graph alignment methods. \nCommittee: \nProf. Stratis Ioannidis (Advisor) \nProf. Dana Brooks (Advisor) \nProf. Tina Eliassi-Rad
URL:https://ece.northeastern.edu/event/kimia-shayestehfards-phd-proposal-review/
END:VEVENT
END:VCALENDAR