SOMA is a hierarchical reinforcement learning framework that enables a quadruped robot to perform whole-body, non-prehensile manipulation of multi-object stacks. It incorporates explicit base-height control to allow full three-dimensional pushing, and separates high-level manipulation decisions from low-level locomotion adjustments. In simulation, SOMA achieves over 90% success rates for single and two-box reposing tasks, and its zero-shot sim-to-real transfer is validated on a Unitree Go1 robot handling various payloads.
By Stergios E. Bachoumas, Panagiotis Artemiadis
PremiseCheck is a framework that improves fact‑checking of peer‑review comments by first having an LLM rewrite each comment into explicit premises without reading the paper, and then using a decision model to verify each premise against the paper. This approach achieves accuracy comparable to GPT‑5.5 across NLP, machine learning, and natural science reviews while cutting costs roughly in half and reducing latency, especially in NLP reviews where it cuts cost to a quarter and latency to less than half. The system is built on the ReviewFact benchmark, which labels factual errors in review comments and measures cost, latency, and accuracy.
By Dingjie Song, Zhiling Yan, Ziqian Zhang, Yuxuan Zhang, Lichao Sun
The paper proposes treating an ensemble of language models as a jury rather than an electorate, where each model’s reliability is inferred from patterns of agreement and disagreement without using labels. Using Expectation Maximisation, the authors estimate each expert’s latent reliability and derive a log‑odds weighted linear scoring rule. On a 1560‑question benchmark, this label‑free approach matches the performance of label‑fitted weights and surpasses the best single model and other label‑free alternatives.
By Alessandro Rozza, Davide Mambelli, Steven Motta, Juri Taresch
The paper introduces SSR (Structural Signal Retriever), a deterministic, training‑free method for selecting relevant Agent Cards in large multi‑agent systems without exposing the full registry to an LLM orchestrator. SSR extracts structural lexical signals from Agent Card fields, weights them with inverse document frequency, and uses an Inverted Signal Index to score candidates efficiently. Experiments on synthetic benchmarks, MS MARCO, and end‑to‑end orchestration show that SSR cuts LLM input by 98.8 % and latency by 70.9 % at 1,000 agents, while preserving retrieval effectiveness and preventing operational failures seen with full‑catalog approaches.
By Giancarlo Maldonado C\'ardenas, Carlos A. Astudillo, Nelson L. S. da Fonseca
The paper introduces DST‑Net, a decomposed spatio‑temporal network that separately models trend and seasonal components of time series data. It employs a Latent Graph Block for the trend branch to capture global spatial dependencies and a Gaussian Multi‑scale Attention mechanism for the seasonal branch to focus on local temporal patterns while reducing attention distraction. Experiments on real‑world benchmarks show that this tailored approach improves forecasting performance.
By Yaohua Zha, Hai Wu, Shu-Tao Xia
CoSTA is a post‑hoc framework that decouples the generation and synthesis token spaces in Spoken Dialogue Models, allowing the use of compressed speech tokens for faster generation while preserving the original tokenizer and decoder. It extends the generation token space with compressed representations through mean‑initialized vocabulary extension, pronunciation adaptation, and dialogue adaptation, then deterministically expands back to the original token space for synthesis. Experiments on bilingual speech‑dialogue benchmarks show about 1.7× faster generation, lower latency, higher throughput, and maintained response quality, with a practical trade‑off between compression, decoding efficiency, and quality.
By Yingzhuo Deng, Chong Li, Bingxin Liu, Yi Shu, Tianyu Peng, Can Xie, Siyao Song, Shuo Ren, Jiajun Zhang
System1-Arena is a benchmark that evaluates which decisions can be handled by System‑1 models—those that output probabilities in a single forward pass—versus System‑2 models that generate reasoning steps. It contains 92,000 questions across 105 tasks from 268 public sources, grouped into text classification, applied decisions, evaluation and control, and agent/action, and also includes 138 closed‑loop environments such as web shopping, coding control, robotics, games, and driving. The study tests 30 models, showing that large System‑1 models can match or exceed fast closed LLMs in non‑agentic accuracy with better‑calibrated probabilities at lower latency, while reasoning LLMs are most accurate but far more expensive, and large open LLMs perform best in closed‑loop settings.
By Tianyu Chen, Xinyang Liu, Ruyi An, Zhendong Wang, Mingyuan Zhou
The paper introduces a benchmark of 83 published works to evaluate whether language models can generate nonlinear narratives—stories where the order of events differs from the order in which they are told. For each work, the authors annotate both chronological and narrative event orders, a causal story graph, character knowledge, and related questions. Experiments show that while models tend to produce more chronological stories than the originals, they only match human performance about half the time when asked to produce a specific nonlinear structure, and readers generally find the default chronological outputs easier to follow.
By Syed Rifat Raiyan, Md. Tariquzzaman, Md Farhan Ishmam, Tenghao Huang
The paper evaluates Muon, a data‑efficient optimizer, against AdamW on a single A100 GPU across 17.7M‑ and 33.6M‑parameter causal language models and a 2.70M‑parameter vision transformer. Using a 23‑candidate development budget and five matched seeds, Muon consistently reduces validation‑loss area under the curve by 3.25–22.44% and lowers test loss in all 20 primary pairs, while AdamW outperforms Muon on language tasks by 3.38–6.96%. Muon’s full optimizer call consumes 56.1–69.6% of training time versus 6.1–9.1% for AdamW, and it shows higher expected regret under restricted candidate pools, indicating trade‑offs between data efficiency and computational cost for small models with reference kernels.
By Mushir Akhtar, Hazem N. Nounou, Majdi Mansouri, Mohamed N. Nounou
The paper investigates the effectiveness of tool graph memory—a component that records past tool transitions to guide future tool selection—in large language model agents performing multi‑turn tasks. Across nine models and two benchmarks, the authors find that tool graph memory rarely yields significant performance gains, often incurs more than twice the number of LLM calls, and its guidance is frequently incorrect. They observe that random or no guidance can achieve similar improvements, suggesting the benefits arise from additional computation rather than the memory itself, and that storing experiences as textual insights outperforms the graph approach with minimal extra calls.
By Uiji Hwang, Richeng Xuan, Taeuk Kim
Agentic-Priors uses large language models to generate informative priors for preference learning agents, addressing the cold start problem by translating a new user's trade-offs into a prior over multinomial logit coefficients. The Bayesian learner refines this prior as choices are observed, improving cold start accuracy by 5–40 percentage points across seven datasets. Compared to state‑of‑the‑art methods, Agentic-Priors outperforms AutoElicit on three datasets, surpasses in‑context learning by up to 18 points, and is valued at roughly 6–40 surveyed users in log‑loss terms, making it especially useful for early deployment users.
By Bruno Arcanjo, Zhaoxing Li, Sebastian Stein
LIBERO-PeRM is a benchmark for studying personalized robotic manipulation, focusing on how policies learn and follow user preferences in zero‑shot settings, from demonstrations, and across varying tasks and layouts. It introduces a taxonomy of language‑expressible preferences—motion, skill, and user‑profile choices—within the LIBERO simulation framework, generating 312 tasks with paired demonstrations and preference‑satisfaction metrics. Experiments reveal that current vision‑language‑action policies struggle with zero‑shot personalization, while supervised fine‑tuning improves learnability, yet challenges remain in generalization, decomposition, and conflict handling.
By Zhixu Li, Keqian Tang, Litian Gong, Jingyu Yao, Wenqian Zhang, Tianze Xu, Zehao Wang, Yuping Wang, Junge Zhang, Jiachen Li
GraphCompliance is a structure‑aware framework that models regulatory documents as a Policy Graph and contextual information as a Context Graph of entity‑relation triples. It aligns these graphs via a Compliance Gate to identify relevant regulatory requirements and evaluate them against contextual evidence, thereby preserving structural dependencies and reducing terminology mismatch. Experiments on EU2UK, GDPR‑13, and CONTRACT datasets show higher Matthews correlation coefficients than competing methods, and ablation studies confirm the importance of each component.
By Jiseong Chung, Ronny Ko, Won-Yong Shin, Wonchul Yoo, Tae-Wan Kim
The paper introduces Pairwise Difference Learning (PDL) for classification, extending a meta‑learning technique originally designed for regression. PDL learns a function that predicts the difference between outcomes of two instances; predictions for a new instance are obtained by averaging differences with all training examples. Experiments show that the resulting PDL classifier surpasses state‑of‑the‑art methods, and the authors release a user‑friendly Python implementation.
By Mohamed Karim Belaid, Maximilian Rabus, Eyke H\"ullermeier
The paper introduces FACTGUARD, a multi-task workflow that generates evidence‑grounded answerable and unanswerable questions to improve long‑context reasoning in large language models. It presents FACTGUARD‑BENCH, a benchmark of 25,220 examples with context lengths up to 128K, focusing on Lack of Evidence and Misleading Evidence scenarios. Experiments on eleven LLMs show a reliability gap, with the best model achieving only 71.62% accuracy; training on FACTGUARD‑BENCH raises accuracy to 81.17% and enhances reasoned rejection of unanswerable queries.
By Qian-Wen Zhang, Biao Liu, Fang Li, Jie Wang, Lingfeng Qiao, Yifei Yu, Di Yin, Xing Sun
The paper introduces AXIOM, a perturbation-based framework for creating a balanced benchmark of code evaluation problems. AXIOM refines program quality by applying controlled semantic perturbations and then calibrates scores through selective manual annotation, resulting in the multilingual AXIOMBENCH dataset with 1,962 problems. This approach addresses limitations of prior benchmarks that rely on coarse binary labels or subjective human scoring.
By Ruiqi Wang, Xinchen Wang, Cuiyun Gao, Chun Yong Chong, Xin Xia, Qing Liao
The article examines whether large language models (LLMs) can effectively use representations they learn in-context to perform downstream tasks. Experiments with open‑weight LLMs on next‑token prediction and a new adaptive world modeling task reveal that these models struggle to deploy novel semantics defined in-context, even when such semantics are encoded in their latent representations. The study also shows that state‑of‑the‑art closed‑source reasoning models similarly fail to reliably leverage in‑context patterns, highlighting a gap in flexible representation usage.
By Michael A. Lepori, Tal Linzen, Ann Yuan, Katja Filippova
The paper introduces Span-Centric Learning (SCL), a method that decouples coding knowledge acquisition from its application in full clinical documents. SCL first trains on short text spans to learn coding knowledge, then fine‑tunes on a small set of annotated documents, enabling an 8B LLM to outperform state‑of‑the‑art methods with only 248 full documents. The approach also yields explicit, human‑aligned supporting evidence for each predicted ICD code and can learn code knowledge from spans even when the codes are absent from the full documents.
By Xu Zhang, Wenxin Ma, Chenxu Wu, Rongsheng Wang, Zhiyang He, Xiaodong Tao, Kun Zhang, S. Kevin Zhou
The paper examines privacy risks in large language models (LLMs) through a unified threat model and systematic ablation study of four attacks: Membership Inference, Attribute Inference, Data Extraction, and Backdoor Attacks. It finds that membership inference is highly sensitive to dataset and attack strategy, attribute inference depends on model and training characteristics, data extraction grows with model size and repeated training data, and backdoor attacks consistently succeed. The authors provide practical recommendations for privacy‑aware LLM deployment, such as prioritizing membership leakage checks in retrieval‑augmented systems, reducing repeated personal data in training, monitoring attribute inference, and protecting fine‑tuning pipelines from backdoors.
By Karima Makhlouf, Lamiaa Basyoni, Syed Khaderi, Gabriel Marquez, Peter Sotomango, Mahmoud Awawdah, Sami Zhioua
The paper investigates whether the reported out‑of‑domain performance gains of Reinforcement Learning with Verifiable Rewards (RLVR) are due to improved answer quality or simply to changes in answer formatting. By comparing scores from standard extractors with a custom extractor that distinguishes answers inside a oxed{} notation, the authors find that a large portion (about 82%) of the apparent gains on 7B SimpleRL models vanish, indicating that the gains are largely due to formatting rather than better answers. In contrast, a 32B model trained with a different recipe shows minimal format‑related gains, suggesting that the effect depends on the training setup.
By Zongye Lyu