MARCH extends recurrent memory via content routing and periodic state checkpoints, improving long-range dependency modeling.

Ming Zhang, Kaisen Yang, Shu Yu et al.

MindMemOS offers a portable, self-evolving memory layer, achieving 94.03% accuracy on LOCOMO.

Kaichao Liang, Yuqi Cui, Hao Kong et al.

Lapis employs First-Spike Latency and membrane leakage to realize Laplacian attention, achieving 96.56% accuracy on CIFAR-10 with 14.5× energy reduction.

Kaiwen Tang, Jiaqi Zheng, Zixuan Zhu et al.

DARC uses failure diagnosis to restrict repair interventions, improving agent self-correction with 20-30% success rate gains.

Pan Wang, Yihao Hu, Hang Wang et al.

ContactIPM combines structure-exploiting interior-point method with stagewise elastic relaxation, achieving 2-8x speedup in contact-implicit trajectory optimization.

Yucheng Chen

XBRIDGE combines lexical anchor mapping and latent enrichment bridge for efficient heterogeneous LLM communication, reducing latency by 11×.

Wooseong Yang, Wei-Chieh Huang, Weizhi Zhang et al.

Introduces low-interaction rank functions for unified dual-encoder design, validating spectral decay and normalization effects.

Zijian Zhao, Sen Li

Proposes HPSE, a hybrid rollout self-distillation method, significantly improving unstructured knowledge editing's composability, with +6.8 average points across four LLMs.

Tianci Liu, Zihan Dong, Tianchun Li et al.

RIFT constructs full future K/V cache in one pass using learned anticipation tokens, achieving 98.8% success and reducing latency by up to 89%.

Chushan Zhang, Jinguang Tong, Xuesong Li et al.

Introducing Social Chain of Thought (SCoT), a multi-agent framework that improves differential diagnosis recall by 4-12% through multi-round specialist interactions.

Del Coburn, Scott Sanner, Dan Silver

Proposes SpeedRunner, a programmatic skill learning method that reduces agent costs by analyzing trajectories without replay, achieving over 60% cost savings and performance gains.

Zixi Huang, Xiheng Wang, Andrew Wang et al.

Meta-detection with verifiable temporal grounding and reward redistribution boosts AI-generated video detection F1 to 75.39%, outperforming existing methods.

Bowei Liu, Zheng Lu, Yuhan Bian et al.

Using AI to tighten bounds on the Grothendieck constant, achieving a lower bound of 6π/11≈1.7135, surpassing previous 1.6769.

Alan Li, Rahul Saha, Anton Xue et al.

Proposes a reflection-guided self-distillation framework for test-time GUI grounding, achieving a 7.4% accuracy boost without human annotations.

Shiyu Xuan, Zechao Li

Proposes VIScore, integrating veracity, influence, and sobriety, to evaluate planning-relevant quality in latent world models, with correlation exceeding 0.75.

Haiyu Wu, Randall Balestriero, Morgan Levine

Proposes a quantum softmax attention based on Born-rule measurement, using angle mapping for exact single-head transformer layers with sparse boundary expression.

Eric A. F. Reinhardt, Adam J. Hauser

This survey systematically reviews conditional independence (CI) tests in constraint-based causal discovery, analyzing assumptions, robustness, and scalability in high-dimensional and mixed data.

Pavel Averin, Theodoros Moysiadis, Ioannis Katakis

Introduced the first subquadratic-sample estimator for von Neumann entropy, breaking the quadratic barrier with complexity o(d^2).

Minbo Gao, Qisheng Wang

This paper introduces a cross-lingual policy retention metric based on action traces, revealing that four models retain 71-73% of their strategies across languages after bias correction.

Sourabrata Mukherjee, Kalika Bali, Sunayana Sitaram

Analyzing 21 million papers with NLP, found Chinese researchers favor open-source models like Qwen, reaching 44% usage in 2026, indicating regional shifts in AI ecosystems.

Zackary Okun Dunivin