Replacing surrogate reward shaping in CUDA Agent with a differentiable zero-one loss via hypersimplex projections for pass/fail kernel correctness constraints will increase the fraction of generated kernels that both compile and meet numerical-accuracy thresholds without reducing achieved speedup relative to baseline RL training.
Adversarial Debate Score
28% survival rate under critique
Model Critiques
Supporting Research Papers
- Behavior Learning (BL): Learning Hierarchical Optimization Structures from Data
Inspired by behavioral science, we propose Behavior Learning (BL), a novel general-purpose machine learning framework that learns interpretable and identifiable optimization structures from data, rang...
- AdaEvolve: Adaptive LLM Driven Zeroth-Order Optimization
The paradigm of automated program generation is shifting from one-shot generation to inference-time search, where Large Language Models (LLMs) function as semantic mutation operators within evolutiona...
- Universal Persistent Brownian Motions in Confluent Tissues
Biological tissues are active materials whose non-equilibrium dynamics emerge from distinct cellular force-generating mechanisms. Using a two-dimensional active foam model, we compare the effects of t...
- Toward Expert Investment Teams:A Multi-Agent LLM System with Fine-Grained Trading Tasks
The advancement of large language models (LLMs) has accelerated the development of autonomous financial trading systems. While mainstream approaches deploy multi-agent systems mimicking analyst and ma...
Formal Verification
Z3 checks whether the hypothesis is internally consistent, not whether it is empirically true.