🏢 Mila, Université De Montréal
Efficient Adversarial Training in LLMs with Continuous Attacks
·2099 words·10 mins·
loading
·
loading
Large Language Models
🏢 Mila, Université De Montréal
Boosting LLM robustness against attacks efficiently: Continuous adversarial training in embedding space outperforms discrete methods, achieving improved robustness with less computation.
Amortizing intractable inference in diffusion models for vision, language, and control
·5979 words·29 mins·
loading
·
loading
AI Generated
Machine Learning
Reinforcement Learning
🏢 Mila, Université De Montréal
Amortized sampling from complex posteriors using diffusion models is achieved via a novel data-free learning objective, Relative Trajectory Balance (RTB). RTB’s asymptotic correctness is proven, offe…