6 papers
PUNCH: Physics-informed Uncertainty-aware Network for Coronary Hemodynamics
Sukirt Thakur, Marcus Roper, Yang Zhou +8
More than 10 million coronary angiograms are performed globally each year, providing a gold standard for detecting obstructive coronary artery disease. Yet, no obstructive lesions…
Aligning to What? Limits to RLHF Based Alignment
Logan Barnhart, Reza Akbarian Bafghi, Stephen Becker +1
Reinforcement Learning from Human Feedback (RLHF) is increasingly used to align large language models (LLMs) with human preferences. However, the effectiveness of RLHF in addressin…
Fine Tuning without Catastrophic Forgetting via Selective Low Rank Adaptation
Reza Akbarian Bafghi, Carden Bagwell, Avinash Ravichandran +2
Adapting deep learning models to new domains often requires computationally intensive retraining and risks catastrophic forgetting. While fine-tuning enables domain-specific adapta…
Where Did Your Model Learn That? Label-free Influence for Self-supervised Learning
Nidhin Harilal, Amit Kiran Rege, Reza Akbarian Bafghi +2
Self-supervised learning (SSL) has revolutionized learning from large-scale unlabeled datasets, yet the intrinsic relationship between pretraining data and the learned representati…
MixDiff: Mixing Natural and Synthetic Images for Robust Self-Supervised Representations
Reza Akbarian Bafghi, Nidhin Harilal, Claire Monteleoni +1
This paper introduces MixDiff, a new self-supervised learning (SSL) pre-training framework that combines real and synthetic images. Unlike traditional SSL methods that predominantl…
Parameter Efficient Fine-tuning of Self-supervised ViTs without Catastrophic Forgetting
Reza Akbarian Bafghi, Nidhin Harilal, Claire Monteleoni +1
Artificial neural networks often suffer from catastrophic forgetting, where learning new concepts leads to a complete loss of previously acquired knowledge. We observe that this is…