Skip to content
VirtualOnlinePast event

Advances in AI at Johns Hopkins University

Event artwork for Advances in AI at Johns Hopkins University

This event has ended

This listing is kept as an archive. Explore upcoming events to find your next one.

See upcoming events
About this event

What to expect

This virtual meetup brings together cutting-edge AI research from Johns Hopkins University, featuring three PhD researchers presenting their latest work in image generation, medical imaging, and trustworthy AI. Taking place on April 23, 2026 at 9:00 AM PST, this online event offers a unique opportunity to learn from emerging leaders at one of the world's premier research institutions.

Featured Presentations

Recent Advancements in Image Generation and Understanding — Aniket Roy, Research Scientist at NEC Labs America and recent Johns Hopkins PhD graduate, explores the evolution of image generation from Gaussian Mixture Models to state-of-the-art diffusion models. His presentation covers three breakthrough works: DiffNat (TMLR'25), which leverages kurtosis concentration properties to produce sharper, more faithful image generations; DuoLoRA (ICCV'25), achieving superior content-style personalization with 19× fewer trainable parameters; and Cap2Aug (WACV'25), a novel approach to semantic augmentation that addresses data scarcity through cross-modal backtranslation.

From Representation Analysis to Data Refinement — Basudha Pal, recent ECE PhD graduate from Johns Hopkins, examines how deep learning models encode information beyond their intended objectives. Using mutual information-based expressivity estimation, she reveals how models implicitly capture demographic and anatomical attributes, and demonstrates data attribution techniques to identify and remove training samples that reduce model robustness. This work is particularly relevant for biometric and medical imaging applications where fairness and reliability are critical.

Scalable & Precise Histopathology — Tawsifur Rahman, PhD candidate in Biomedical Engineering, tackles the computational challenge of whole slide images (WSIs) at gigapixel resolution. He presents two complementary solutions: a Task-Specific Self-Supervised Learning framework that outperforms existing foundation models for cancer classification, and CEMIL, a contextual attention-based approach achieving state-of-the-art accuracy while using only a fraction of WSI patches. With publications in Nature Modern Pathology and over 8,500 citations, Rahman brings proven expertise in computational pathology.

Towards Trustworthy AI Under Real World Data Challenges — [Ay

Running an AI event?

List it here, free.

Meetups, workshops and conferences go in front of thousands of builders after a quick review.

List your event