Luna Bound Propagator: Efficient Neural Network Verification
Discover Luna, a C++ bound propagator offering fast, versatile formal analysis for neural networks with superior performance over Python-based tools.
HDPO: Optimizing RL with Hybrid Distillation for LLMs
Discover HDPO, a novel hybrid distillation method that boosts reinforcement learning in large language models for better math reasoning and prompt handling...
Robust Reasoning in VLMs: A Neuro-Symbolic Approach
Explore how neuro-symbolic methods improve vision-language models' reasoning under distribution shifts for robust AI performance.
Max Second Derivative of Activations & Adversarial Robustness
Explore how the max second derivative of activation functions impacts neural network adversarial robustness and model resilience.
PoliticsBench: Evaluating Political Bias in Large Language Models
Discover how PoliticsBench measures political bias in top LLMs using multi-turn roleplay to reveal left and right-leaning tendencies in AI responses.