theoretical analyses
Comprehensive studies that explore the principles, limitations, and capabilities of AI algorithms from a mathematical or statistical perspective, forming a theoretical foundation for empirical experiments.
- ALMGuard: Safety Shortcuts and Where to Find Them as Guardrails for Audio–Language Models
- Asymptotic theory of SGD with a general learning-rate
- CHPO: Constrained Hybrid-action Policy Optimization for Reinforcement Learning
- Discrete Diffusion Models: Novel Analysis and New Sampler Guarantees
- Entropy-Calibrated Label Distribution Learning
- FairDD: Fair Dataset Distillation
- ReDit: Reward Dithering for Improved LLM Policy Optimization
- Removing Concepts from Text-to-Image Models with Only Negative Samples
- Shallow Diffuse: Robust and Invisible Watermarking through Low-Dim Subspaces in Diffusion Models
- Understanding Generalization in Physics Informed Models through Affine Variety Dimensions
- Why and How LLMs Hallucinate: Connecting the Dots with Subsequence Associations