Skip to main content
Search

# Long-Horizon Reasoning

3 related articles

Subscribe to updates
  1. arXiv

    iS-KV compresses KV cache online via block-incremental SVD, retaining 82.6% accuracy at 4.06x compression on DeepSeek-R1-Distill-Llama-8B

    The work proposes iS-KV, an online low-rank KV-cache compression method for long-horizon reasoning that keeps a recent window exact, incrementally folds older states into bounded-rank representations, and synchronizes historical coordinates as the low-rank basis evolves to maintain representation consistency; on DeepSeek-R1-Distill-Llama-8B it reaches 82.6% accuracy at 4.06-fold persistent-KV compression (original model 83.6%), and on Qwen3-8B it reaches 89.2% accuracy at 5.64-fold compression, consistently outperforming token-eviction baselines under matched memory budgets.

    University of MacauChinese Academy of SciencesShenzhen University of Advanced TechnologyThe Hong Kong University of Science and Technology (Guangzhou)
    Yiren Zhao（First author）
    iS-KVLow-rank KV cache compressionKV CacheLong-Horizon Reasoning
  2. arXiv

    ACG lifts long-horizon agent success from 44.5% to 50.2%, with an 11.1-point gain on BrowseComp-Plus

    The work introduces the Adaptive Consistency Graph (ACG), which incrementally organizes execution evidence and its provenance into a persistent graph and builds a temporary requirement-centered view for each decision under a bounded context budget, raising GPT-5.6-luna's equal-weight average success across three benchmarks from 44.5% with ReAct to 50.2% and BrowseComp-Plus from 62.4% to 73.5%, without replacing the base planner or tool executor.

    Beihang University
    Jiecong Wang（First author）
    Adaptive Consistency GraphAgent MemoryContext ManagementLong-Horizon Reasoning
  3. arXiv

    SAGE combines algebraic sparsification and hyperbolic structural guidance to curb long-horizon reasoning biases, beating baselines across 12 benchmarks and 7 model families with up to 8-fold gains on Andrews-Curtis

    The work introduces Symbolic Closure Analysis (SCA) to characterize exploration bias and compounding bias in long-horizon reasoning, and builds SAGE, a framework that injects structural priors during post-training via algebraic sparsification and hyperbolic structural guidance, outperforming SFT, GRPO, EMPO, and GRPO-PRM across 12 benchmarks and 7 model families and achieving up to an 8-fold improvement in Lean-verified proofs on the open Andrews-Curtis task.

    University of Wisconsin–MadisonVirginia TechDartmouth College
    Xinyue Zeng（First author）
    SAGE (Structural Admissibility-Guided Exploration)Symbolic Closure AnalysisLong-Horizon ReasoningHyperbolic Structural GuidanceAndrews-Curtis Problem

You’re all caught up