Avoiding unsafe sets when training with Langevin Dynamics
Adam M. Oberman
cs.LG, math.AP, stat.ML
Submitted: 2026-07-08
Comments: v2: expanded and revised; 21 pages, 3 figures
Project page: https://chewisinho.github.io/main.pdf
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- Implicit Gradient Regularization
- Safety from Honesty in a Disinterested AI Predictor
- Analysis of Langevin Monte Carlo from Poincar'e to Log-Sobolev
- On Large Batch Training and Sharp Minima: A Fokker-Planck Perspective
- Convergence of Langevin Monte Carlo in Chi-Squared and Renyi Divergence
- On the diffusion approximation of nonconvex stochastic gradient descent
- Risks from Learned Optimization in Advanced Machine Learning Systems
- Stochastic Modified Equations and Dynamics of Stochastic Gradient Algorithms I: Mathematical Foundations
- Stochastic Gradient Descent as Approximate Bayesian Inference
- Non-convex learning via Stochastic Gradient Langevin Dynamics: a nonasymptotic analysis
- Defining and Characterizing Reward Hacking
- On the Origin of Implicit Regularization in Stochastic Gradient Descent
- Rapid Convergence of the Unadjusted Langevin Algorithm: Isoperimetry Suffices
- A Hitting Time Analysis of Stochastic Gradient Langevin Dynamics
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks