SLAT: Segment-Level Adaptive Trimming for Efficient CoT Reasoning
Quick Answer
The SLAT framework introduces Segment-Level Adaptive Trimming to enhance chain-of-thought reasoning in large models, reducing reasoning length by 50% while maintaining accuracy.
Quick Take
This approach selectively targets redundant segments, addressing inefficiencies that arise from structural redundancy in reasoning chains. Empirical results show SLAT establishes a superior accuracy-efficiency balance on standard benchmarks.
Key Points
- SLAT reduces reasoning length by 50% compared to uncompressed baselines.
- The framework targets high-probability segments with low marginal utility.
- Empirical results indicate a superior accuracy-efficiency Pareto frontier.
- Traditional methods rely on coarse, segment-agnostic length penalties.
- SLAT is grounded in a theoretical characterization of segment suboptimality.
Paper Resources
Article Content
From source RSS / original summaryarXiv:2605. 30832v1 Announce Type: new Abstract: Recent advances in Large Reasoning Models have significantly improved chain-of-thought (CoT) capabilities via reinforcement learning (RL). However, generated reasoning chains frequently suffer from structural redundancy (i. e. , \emph{overthinking}), incurring high computational overhead without improving answer correctness.
Existing mitigation strategies typically rely on token-uniform length penalties, which provide coarse, segment-agnostic pressure toward shorter outputs and can inadvertently suppress useful reasoning alongside redundancy. To address this, we demonstrate that inefficiency concentrates in high-probability segments with low marginal utility.
We derive a theoretical characterization of segment suboptimality under the correctness-length trade-off objective and propose \textsc{SLAT} (Segment-Level Adaptive Trimming), an RL framework that selectively suppresses redundant segments based on this criterion. Empirical results on standard benchmarks indicate that \textsc{SLAT} establishes a superior accuracy-efficiency Pareto frontier, reducing reasoning length by $50\%$ relative to uncompressed baselines while maintaining competitive accuracy.
Overall, our results suggest that theoretically grounded, segment-aware trimming is a promising direction for efficient CoT reasoning in .
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.AI
See more →HOBA: Hierarchical On-Policy Bidding Agents for Adaptive Online Advertising
HOBA (Hierarchical On-policy Bidding Agents) is a novel hierarchical reinforcement learning framework that enhances online advertising bidding systems by improving adaptability and reducing hyperparameter tuning costs. It utilizes a for hyperparameter inference, a SARSA agent for expert model selection, and a dynamic expert pool for bid execution, achieving a +3.6% increase in target cost during large-scale deployment and outperforming state-of-the-art baselines on AuctionNet.