April 2023 - Sanghani-faculty - mailman.cs.vt.edu

Meghana Holla's MS Thesis Defense (Monday May 1st, 1:00-3:00pm)
by Ismini Lourentzou 29 Apr '23

29 Apr '23

Hello everyone, You are invited to join us for Meghana's MS thesis defense this Monday. *Title:* Commonsense for Zero-Shot Language Video Localization *Date:* Monday, May 1st *Time:* 1:00-3:00 pm *Location:* 3160A Torgersen Hall and Zoom https://virginiatech.zoom.us/j/87360587457 <https://www.google.com/url?q=https://virginiatech.zoom.us/j/87360587457&sa=…> *Committee:* Ismini Lourentzou (chair), Narendran Ramakrishnan, Lifu Huang *Abstract: *Zero-shot Natural Language-Video Localization (NLVL) is designed to train an NLVL model with only raw video data by dynamically generating video segment proposals and pseudo-query annotations. However, pseudo-queries generated by prior work are not fully grounded in the source video. Additionally, given the unstructured nature of the pseudo-queries, the video-query annotations lack common ground. To this end, we propose CORONET, a zero-shot NLVL framework that leverages commonsense information to bridge the gap between the video and generated pseudo-queries through a commonsense enhancement module. We employ Graph Convolution Networks (GCN) to encode commonsense information extracted from a knowledge graph, conditioned on the video, and cross-attention to enhance the encoded video and pseudo-query vectors prior to localization. Empirical evaluations on two datasets show that our model outperforms several zero-shot and weakly supervised baselines, highlighting the importance of commonsense reasoning in multimodal understanding. Best, Ismini -- *Ismini Lourentzou* <https://isminoula.github.io/> Assistant Professor Virginia Tech | Computer Science Email: ilourentzou(a)vt.edu

1 0

Manoj Prabhakar's MS thesis defense - EdgeFn: A Lightweight Customizable Data Store for Serverless Edge Computing
by Liting Hu 20 Apr '23

20 Apr '23

Hello all, You are all invited to attend Manoj Prabhakar's MS thesis defense on April 21 2pm. *Title*:- EdgeFn: A Lightweight Customizable Data Store for Serverless Edge Computing. *Location*:- Zoom (https://virginiatech.zoom.us/j/9775852036 <https://www.google.com/url?q=https://virginiatech.zoom.us/j/9775852036&sa=D…> ) *Time:- *2:00 - 3:00 pm EST *Abstract*: Serverless Edge Computing is an extension of the serverless computing paradigm that enables the deployment and execution of modular software functions on resource-constrained edge devices. However, it poses several challenges due to the edge network's dynamic nature and serverless applications' latency constraints. In this work, we introduce EdgeFn, a lightweight distributed data store for the serverless edge computing system. While serverless computing platforms simplify the development and automated management of software functions, running serverless applications reliably on resource-constrained edge devices poses multiple challenges. These challenges include a lack of flexibility, minimum control over management policies, high data shipping, and cold start latencies. EdgeFn addresses these challenges by providing distributed data storage for serverless applications and allows users to define custom policies that affect the life cycle of serverless functions and their objects. First, we study the challenges of existing serverless systems to adapt to the edge environment. Second, we propose a distributed data store on top of a Distributed Hash Table (DHT) based Peer-to-Peer (P2P) Overlay, which achieves data locality by co-locating the function and its data. Third, we implement programmable callbacks for storage operations which users can leverage to define custom policies for their applications. We also define some use cases that can be built using the callbacks. Finally, we evaluate EdgeFn scalability and performance using industry-generated trace workload and real-world edge applications.

1 0

Gaurang's Karwande MS Defense - Geometric Deep Learning for Healthcare Applications
by Ismini Lourentzou 17 Apr '23

17 Apr '23

Hello all, You are invited to join us this Wednesday for Gaurang's Karwande MS thesis defense. *Title:* Geometric Deep Learning for Healthcare Applications *Date:* Wednesday, April 19 *Time:* 3:00-4:30 pm *Location:* Whittmore 457B and Zoom https://virginiatech.zoom.us/j/85081753808 *Committee: *Ismini Lourentzou (Co-chair), Creed Jones (Co-chair), Lynn Abbot *Abstract: *This thesis explores the application of Geometric Deep Learning or Graph Neural Networks (GNNs) for medical image analysis and causal structure learning. Tracking the progression of pathologies in chest radiography poses several challenges in anatomical motion estimation and image registration as this task requires spatially aligning the sequential X-rays and modeling temporal dynamics in change detection. The first part of this thesis proposes a novel approach for change detection in sequential Chest X-ray (CXR) scans using GNNs. The proposed model CheXRelNet utilizes local and global information in CXRs by incorporating intra-image and inter-image anatomical information and showcases an increased downstream performance for predicting the change direction for a pair of CXRs. The second part of the thesis focuses on using GNNs for causal structure learning. The proposed method introduces the concept of intervention on graphs and attempts to relate belief propagation in Bayesian Networks (BN) to message passing in GNNs. Specifically, the proposed method leverages the downstream prediction accuracy of a GNN-based model to infer the correctness of Directed Acyclic Graph (DAG) structures given observational data. Our experimental results do not reveal any correlation between the downstream prediction accuracy of GNNs and structural correctness and hence indicate the harms of directly relating message passing in GNNs to belief propagation in BNs. Overall, this thesis demonstrates the potential of GNNs in medical image analysis and highlights the challenges and limitations of applying GNNs to causal structure learning. Best, Ismini -- *Ismini Lourentzou* <https://isminoula.github.io/> Assistant Professor Virginia Tech | Computer Science Email: ilourentzou(a)vt.edu

1 0

Sanghani Center Faculty Meeting - Friday, April 7 at 12 pm
by Shalaby, Wanawsha 07 Apr '23

07 Apr '23

Hi All, Our next Sanghani Center Faculty meeting is scheduled for Friday, April 7 at 12:00 - 1:30 pm. As usual, if you have any recent news or updates since our last faculty meeting, please send them my way so that I may add them to our meeting slides. Updates include awards, proposals, grants, conference presentations, events, etc. Please send all updates by Wednesday, April 5. I'm awaiting confirmation of a conference room for those of you in Blacksburg who would like to join in person. I'll provide the location as soon as the room is booked. Those of us in the DC area will still be joining via zoom. Join Zoom Meeting Link: https://virginiatech.zoom.us/j/82892613942 Meeting ID: 828 9261 3942 One tap mobile +13017158592,,82892613942# US (Washington DC) +13052241968,,82892613942# US Dial by your location +1 301 715 8592 US (Washington DC) Meeting ID: 828 9261 3942 Thanks, Wanawsha Wanawsha Shalaby Manager of Operations UrbComp Program Coordinator Sanghani Center for AI and Data Analytics - Virginia Tech 900 N. Glebe Road, Arlington, VA 22203 Phone: (571) 858-3361 E-Mail: wanah92(a)vt.edu<mailto:wanah92@vt.edu>

1 3

Fwd: Seeking advice on pending call for participation in TLOS Community of Practice
by Chris North 07 Apr '23

07 Apr '23

Looking for faculty partners for generative AI models, LLMS, … > Begin forwarded message: > > From: "Pike, Dale" <dalepike(a)vt.edu> > Subject: Seeking advice on pending call for participation in TLOS Community of Practice > Date: April 6, 2023 at 2:14:43 PM EDT > To: "Ramakrishnan, Narendran" <naren(a)vt.edu>, "North, Christopher" <chnorth1(a)vt.edu> > > The VT TLOS Professional Development Network (PDN) will soon be inviting faculty to consider participating in one of five communities of practice (CoPs) focused on emerging learning technologies that may directly affect how we work, do research, and learn. As part of this call, we are looking for partners who can guide us in exploring the implications of these cutting-edge transformative technologies as well as document and promote best practices. Participation is open to all teaching positions at VT including faculty, instructors, professors of practice, etc. > > I am writing to inquire as to whether you might have any suggestions faculty who might be interested in helping me coordinate a community of practice focused on Exploring Generative AI for faculty. The overall goal is to increase awareness of generative AI tools and practices to improve our collective capacity to determine appropriate use of these technologies for teaching, research and productivity. > > This effort will involve three phases and the call for participation we’re preparing solicits involvement in Phase 1. The overview of each Phase is as follows: > > Phase 1 Planning Activities: > Faculty engagement in one of the five emergent technology CoPs, beginning in late April 2023 to June 30, 2023. A stipend of $1,000 per faculty member involved in the planning cycle (with a $2500 stipend for coordinating faculty) will support engagement in guided discussion and planning activities during dedicated blocks of time that will lead to the submission of a project proposal that if approved will allow for participation in Phase 2. In addition, participation will count for 12 PDN credits toward the computer refresh. > > Phase 2 Planning and Development Grants: > Grants in Phase 2 will be awarded up to $25,000 and are expected to allow for deeper planning and development of the CoP over the 23-24 academic year. Awards are expected to support continued shared exploration of the CoP topic. Activities might include identifying and purchasing training materials, workshops, or invited experts, as well as staffing to support exploration activities to further develop proof of concept and opportunities/challenges for scaling. > > Phase 3 Project Implementation: > Grants in Phase 3 will support more substantial collaborative activities of the CoP. Proposals will be due late April 2024 to support project implementation over the 24-25 academic year. Further details on grant amounts will be released in January 2024 (award amounts are expected to be more than $25,000). > > I’m happy to discuss this further with you, and I also welcome your feedback as to our overall approach, as we’re refining the call for participation right now and I’m eager to put together a program that will be useful for faculty. > > If you have any suggestions for participants or leaders beyond the Sanghani Center that might help me, I’d sincerely appreciate any advice you can share. > > Thanks, > Dale > -- > Dale D. Pike | (he/him) | Associate Vice Provost for Technology-enhanced Learning > > Virginia Tech | Technology-enhanced Learning & Online Strategies > 801 University City Blvd. #2148 > Blacksburg, VA 24060 > Phone: 540-231-7108 <tel:540-231-7108> | dalepike(a)vt.edu <mailto:dalepike@vt.edu>

1 0

ML Day: Amazon - VT Initiative for Efficient and Robust Machine Learning - April 25th
by Shalaby, Wanawsha 03 Apr '23

03 Apr '23

**This event is by invite only, please do not forward** Limited travel funding to attend the event is available. Contact Wanawsha Shalaby (wanah92(a)vt.edu) and Juanita Victoria (juaniv6(a)vt.edu) for more information. [cid:e4ce4474-983a-4ad1-b781-65bc00139c22] RSVP Link: https://www.eventbrite.com/e/ml-day-amazon-virginia-tech-initiative-for-eff… Wanawsha Shalaby Manager of Operations UrbComp Program Coordinator Sanghani Center for AI and Data Analytics - Virginia Tech 900 N. Glebe Road, Arlington, VA 22203 Phone: (571) 858-3361 E-Mail: wanah92(a)vt.edu<mailto:wanah92@vt.edu>

1 1