SPA method uses carefully designed prompts to generate large-scale synthetic data for effective knowledge injection.

Kexian Tang, Jiani Wang, Shaowen Wang et al.

Proposes Spectral-Sphere-Constrained Hyper-Connections (sHC), overcoming Birkhoff polytope limitations to enhance expressivity and stability.

Zhaoyi Liu, Haichuan Zhang, Ang Li

Proposed TARNet for bias-robust uplift modeling; validated via semi-synthetic data under structural biases.

Yuxuan Yang, Dugang Liu, Yiyan Huang

This study employs Lipschitz theory to analyze how demonstrations, CoT, and prompts influence ICL performance.

Xuhan Tong, Yuchen Zeng, Jiawei Zhang

Proposed a new algorithm for online learning with ranking feedback, addressing the absence of traditional numeric feedback.

Mingyang Liu, Yongshan Chen, Zhiyuan Fan et al.

A cost-aware evasion framework reveals robustness gaps in phishing detection; median evasion cost is 2, with over 80% attacks on three low-cost features.

Julian Allagan, Mohamed Elbakary, Zohreh Safari et al.

The paper introduces a maximum-entropy exploration method using future state-action visitation measures, improving feature visitation and convergence speed.

Adrien Bolland, Gaspard Lambrechts, Damien Ernst

ALIGN uses adversarial learning to enhance cross-session generalization in speech neuroprostheses, significantly reducing phoneme and word error rates.

Zhanqi Zhang, Shun Li, Bernardo L. Sabatini et al.

RAMP uses reinforcement learning for adaptive mixed-precision quantization, achieving 6% size and 1-3% quality improvements for on-device LLM inference.

Arpit Singh Gautam, Saurabh Jha

MetaClaw combines skill-driven fast adaptation and policy optimization, boosting accuracy by up to 32% in continuous deployment.

Peng Xia, Jianwen Chen, Xinyu Yang et al.

Efficient reasoning in small LLMs using LoRA adapters and RL, significantly reducing response length.

Yelysei Bondarenko, Thomas Hehn, Rob Hesselink et al.

Proposes five-level AI integration model and an open-source autonomous research framework supporting multi-model, multi-node experiments.

Max Zimmer, Nico Pelleriti, Christophe Roux et al.

HorizonMath evaluates AI progress in mathematical discovery using an automated verification framework, with GPT 5.4 Pro achieving breakthroughs on two problems.

Erik Y. Wang, Sumeet Motwani, James V. Roggeveen et al.

Effective distillation of xLSTM architectures recovers and exceeds teacher model performance.

Lukas Hauzenberger, Niklas Schmidinger, Thomas Schmied et al.

PokeAgent Challenge tests AI decision-making via Pokemon battles and RPG, offering a 20M+ dataset and standardized evaluation framework.

Seth Karten, Jake Grigsby, Tersoo Upaa et al.

Conformal gate routing controls surrogate violations distribution-free; on 35 OpenML tasks, τ=2 achieved 66% coverage and 12% violations.

Iqtedar Uddin, Mazin Khider, André Bauer

CausalEvolve adds causal scratchpads to program evolution, improving four open-ended tasks and reaching 38.89% on AIME.

Yongqiang Chen, Chenxi Liu, Zhenhao Chen et al.

SemRep employs generative code representation with semantic-preserving transformations, improving correctness by 6.9%, performance by 1.1×, and robustness in code transformation tasks.

Weichen Li, Jiamin Song, Bogdan Alexandru Stoica et al.

PhysMoDPO optimizes humanoid motion for physical realism and task performance through preference optimization.

Yangsong Zhang, Anujith Muraleedharan, Rikhat Akizhanov et al.

Using Joint Embedding Predictive Architectures (JEPA) for learning representations in latent space significantly enhances parameter estimation accuracy.

Helen Qu, Rudy Morel, Michael McCabe et al.