BU AI Research Updates and AI-Powered Pathology
Details
Let's get together, eat pizza, and hear from academic AI researchers, local companies and experts in the field. This week we will get short updates on cutting-edge computer vision and AI research from Boston University PhD students and hear from Harsha Vardhan Pokkalla (LinkedIn),
VP of ML R&D at PathAI.
PathAI builds best-in-class AI-powered pathology products for biopharma and diagnostic labs, aiming to discover histology biomarkers and optimize diagnostic workflows to improve patient outcomes. Harsha will walk through three recent research directions from PathAI: PLUTO-4, the latest generation of their pathology vision foundation model; Similarity Search for Model Training, using large-scale embedding retrieval to curate data and mine rare morphologies from slide archives; and Multimodal Models for Disease Classification, recent work combining histology image embeddings with text embeddings to improve disease classification accuracy. He'll close with lessons from deploying these models at scale and open problems where the CV community can have a real clinical impact.
Meeting Location: BU CDS-1750 (Center for Computing & Data Sciences) at 665 Commonwealth Ave, Boston, MA 02215.
Agenda:
- 6:00 - 6:30: Networking & Pizza.
- 6:30 - 6:45: Research Lightning Talks
- Tianle Chen (Prof. Deepti Ghadiyaram) -- Cross-Modal Attacks in Frontier Multimodal LLMs
- Max Whitton (Prof. Boqing Gang) -- Making Sense of Touch from the Child’s View for Contrastive Learning
- 6:45 - 7:20: Speaker: Harsha Vardhan Pokkalla (LinkedIn)
- 7:20 - 7:30: 1 min pitches from the audience. Need help? Looking for a job? Let us know.
- 7:30 - 8:00: More networking.
Cross-Modal Attacks in Frontier Multimodal LLMs
Tianle Chen is a PhD student at Boston University advised by Deepti Ghadiyaram, working on multimodal large language models with a focus on robustness, cross-modal reasoning, and safety.
Making Sense of Touch from the Child’s View for Contrastive Learning
Max Whitton is a first year Ph.D. student advised by Professor Boqing Gong at Boston University, where he researches deep learning and vision–language models with a focus on data and pretraining.
Abstract: Is the sense of touch a mechanism for human babies' learning of visual concepts? If so, can we quantify its importance, and to what extent do babies rely on their sense of touch for visual learning? To approach these questions in a principled way, we propose a structured coding system for baby-centric touch events, yielding a dataset of over 200,000 two-second clips of touch events coded according to this system. Using this dataset, we pretrain developmentally grounded models that reveal promising insights into the nature of baby learning from touch.
Details about the venue:
- Registration with your full legal name is required to attend the event.
Parking/Commute Options:
- There are many spots for on-street parking available along Commonwealth Avenue, as well as on nearby side streets. It is free after 6pm: https://en.parkopedia.com/parking/locations/665_commonwealth_avenue_2g2hdrt2yj88wmn1je/?arriving=202604301800&leaving=202604302000
- There are a few garages within walking distance (also listed on the parkopedia link)
- The venue is easily accessible by the MBTA Green Line B branch (Boston University East or Central stops).