Dimensions overall score 7.0

GitHub Code Pulse

No public code linked for this paper yet.

• I-PPO consistently outperforms SFT and PPO baselines in experiments.result 85%
• The filtering process in I-PPO acts as an intrinsic early stopping mechanism, accelerating training efficiency.result 80%
• I-PPO effectively reduces unfaithful Chain-of-Thought (CoT) reasoning.result 80%
• Traditional PPO algorithms train on the entire rollout buffer, assuming all generated episodes provide beneficial optimization signals.technical 90%
• Episodes generated during PPO training frequently contain noisy or unfaithful reasoning, which can degrade model performance and slow down training.limitation 85%
• The proposed Influence-Guided PPO (I-PPO) framework integrates data attribution into the RL post-training loop.method 95%
• I-PPO calculates an influence score for each episode using a gradient-based approximation to identify episodes anti-aligned with a validation gradient.method 90%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card