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

Supporting this event

Sponsors

Hugging Face
About this event

What to expect

Johns Hopkins University researchers present cutting-edge advances in artificial intelligence at this virtual meetup, scheduled for April 23, 2026, at 9 AM PST. This online event brings together leading academics and practitioners to explore breakthrough developments in deep learning, image generation, and computational pathology—topics at the forefront of modern AI research.

Featured Research Presentations

The session opens with Recent Advancements in Image Generation and Understanding by Aniket Roy, Research Scientist at NEC Labs America and recent Johns Hopkins PhD graduate. Roy explores the evolution from Gaussian Mixture Models to diffusion models, showcasing three groundbreaking works: DiffNat (TMLR'25), which improves generation quality through kurtosis concentration properties; DuoLoRA (ICCV'25), achieving content-style personalization with 19× fewer parameters; and Cap2Aug (WACV'25), addressing data scarcity through cross-modal backtranslation for enhanced few-shot classification.

Basudha Pal, PhD graduate from Johns Hopkins ECE, presents From Representation Analysis to Data Refinement, examining how deep models encode information beyond their training objectives. Using mutual information-based expressivity estimation, this research reveals hierarchical attribute encoding patterns and introduces data attribution techniques to identify and remove harmful training samples that reduce model robustness—critical insights for biometric and medical imaging applications.

Tawsifur Rahman, PhD candidate in Biomedical Engineering at Johns Hopkins, tackles Scalable & Precise Histopathology with next-generation deep learning for digital pathology. His work addresses the computational challenge of analyzing gigapixel whole slide images (WSIs) through two innovative frameworks: Task-Specific Self-Supervised Learning (TS-SSL) for domain-optimized representations, and CEMIL, a contextual attention-based MIL framework achieving state-of-the-art accuracy with significantly reduced computational cost. With publications in Nature Modern Pathology, Nature Digital Medicine, and over 8,500 citations, Rahman brings clinical-grade expertise to computational pathology.

The final presentation explores trustworthy AI under real-world data challenges, addressing fundamental assumptions about data qualit

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