Emergent Sparsity in Frozen Random CNN Feature Extractors for Deep Reinforcement Learning

A collection of arXiv machine-learning research abstracts covering sparse representations in frozen CNN reinforcement-learning agents, team-agnostic football prediction, meta-learned reward shaping for RLHF, molecular structure elucidation, supervised fine-tuning transfer and robustness, auditing of purported dynamic inference, weak-to-strong on-policy distillation, LoRA initialization, and adaptive rollout allocation. The document is benign academic content with no reported security vulnerabilities, exploitation guidance, malicious indicators, or affected products.