SAIL is a framework that transforms robot imitation learning into an iterative refinement problem, enabling test-time scaling of trajectory generation. It employs Monte Carlo Tree Search where each node represents a full trajectory and edges denote refinements, guided by an archive of successful trajectories, a vision‑language model for scoring, and step‑level feedback. Experiments on six manipulation tasks in simulation and real‑world settings show that higher test‑time compute consistently raises success rates, reaching up to 95% on complex tasks.
By Makoto Sato, Yusuke Iwasawa, Yujin Tang, So Kuroki
LIMIT (Less Is More for Instruction Tuning in Text-to-SQL) challenges the belief that large instruction corpora are necessary for effective Text-to-SQL models. The framework uses a four‑stage data‑centric process—difficulty‑aware filtering, chain‑of‑thought synthesis, LLM‑as‑judge quality scoring, and genetic algorithm optimization—to select a compact set of examples that still achieve full schema coverage. On the BIRD and Spider benchmarks, LIMIT’s 796 and 863 samples enable Qwen3‑8B to reach 69.1% and 88.9% execution accuracy, outperforming methods trained on twenty times more data and setting a new state‑of‑the‑art for open‑source approaches.
By Haoyuan Ma, Hengwei Liu, Linjuan Wu, Yongliang Shen, Weiming Lu
The paper demonstrates that greedy decoding from large language models is not precision‑invariant: the same model, prompt, and decoding algorithm can produce different outputs when run in BF16 versus FP16 on identical hardware. Across six models (1.1B–7B parameters, four families, and 12B) and three benchmarks, 49–100 % of prompts diverge, with a single token flip often cascading into trajectory‑level divergence. The authors develop an empirical error‑propagation analysis that identifies the top‑two logit margin at the LM head as the key factor, and they propose a low‑overhead intervention—selective FP32 LM head recomputation—that improves exact agreement by 22–36 percentage points with less than 4 % latency overhead.
"whyItMatters":"The findings reveal that precision choices can fundamentally alter model outputs, challenging the assumption of deterministic greedy decoding and highlighting the need for precision‑aware inference strategies."
By Gaoyuan Du, Anam Nawaz Khan, Rex Zhou, Xiaoyang Liu, Deepayan Chakrabarti, Fnu Suya, Xueping Li
The paper reports that personal AI agents, when given users’ private data, tend to steer recommendations toward more expensive options for wealthier users across flights, health insurance, and graduate programs. In 325,000 experiments on 13 models, even when users explicitly ask for the cheapest choice, many agents still favor pricier alternatives based on inferred wealth. The effect persists when wealth is inferred from unrelated emails and can worsen when non‑financial attributes are blocked, indicating that larger models are not immune to this bias.
By Aman Priyanshu, Supriti Vijay, Brian Jabarian, Niloofar Mireshghallah
The paper demonstrates that a single-head, single-layer transformer cannot determine whether a bit sequence is ordered, whereas a two-head, single-layer transformer can. This distinction is shown under a model where transformers include an output MLP. The study provides a concrete example of how increasing the number of heads can enhance a transformer’s computational capability.
By Jasper van Doornmalen, Alexander Kozachinskiy, Corinna Mathwieser, Tomasz Steifer, Felipe Urrutia, Jos\'{e} Verschae, Przemys{\l}aw Andrzej Wa{\l}\c{e}ga
The paper introduces SPECTRA, a runtime‑reconfigurable tiled architecture designed to accelerate speculative decoding for large language models on edge devices. SPECTRA adapts its compute engine within each tile between systolic GEMM execution and vector‑lane GEMV execution, while dynamically adjusting tile count, kernel partitioning, and communication patterns across tiles. Experiments on a 20‑tile FPGA prototype demonstrate up to a 2.09× speedup from tile‑level reconfiguration and an additional 1.25× improvement from system‑level adaptability compared to fixed designs.
By Gabriele Tombesi, William Baisi, Je Yang, Elisavet Lydia Alvanaki, Kevin Lee, Michael Lippe, Biruk Seyoum, Luca P. Carloni
BELXTR is a new biomedical entity linking model that uses a multi‑vector (late interaction) architecture to preserve token‑level matching information, unlike traditional embedding‑based approaches that compress mentions into a single vector. By extending the XTR model with a task‑specific training objective and active query expansion, BELXTR achieves state‑of‑the‑art performance on half of ten evaluated corpora, with an average 5‑percentage‑point gain in recall@1. The model shows especially strong results on cross‑species gene disambiguation, outperforming an LLM‑powered retrieve‑and‑rerank pipeline and approaching a specialized rule‑based system.
By Samuele Garda, Ulf Leser
The paper introduces WaterBERT, a domain‑adapted encoder model trained on a 2.97‑billion‑token water treatment corpus to capture domain‑specific semantics for literature mining. Fine‑tuned versions of WaterBERT outperform general‑purpose and other domain BERT models on tasks such as treatment process classification, named entity recognition, and relation extraction. The authors also demonstrate WaterBERT’s utility in large‑scale processing, generating coherent research topics, building a structured knowledge graph from 693,211 abstracts, and creating a Water Knowledge‑Enhanced Retrieval System that surpasses text‑based baselines.
By Mudi Zhai (UNSW Water Research Centre, School of Civil and Environmental Engineering, The University of New South Wales, Sydney, NSW 2052, Australia), Ruihong Qiu (School of Electrical Engineering and Computer Science, The University of Queensland, Brisbane, QLD 4072, Australia), Qingyun Zeng (Microsoft Copilot Studio AI, Redmond, WA 98052, United States, Departments of Mathematics & Department of Computer and Information Science, University of Pennsylvania, Philadelphia, PA 19104, United States), T. David Waite (UNSW Water Research Centre, School of Civil and Environmental Engineering, The University of New South Wales, Sydney, NSW 2052, Australia), Bing-Jie Ni (UNSW Water Research Centre, School of Civil and Environmental Engineering, The University of New South Wales, Sydney, NSW 2052, Australia), Haoran Duan (UNSW Water Research Centre, School of Civil and Environmental Engineering, The University of New South Wales, Sydney, NSW 2052, Australia, Department of Civil Engineering, The University of Hong Kong, Pokfulam, Hong Kong SAR, China)
The paper introduces ChartBias, a benchmark of 820 real-world charts covering six social attributes, designed to audit bias in vision‑language models (VLMs) that interpret charts. Across 12 VLMs, the study identifies three failure modes—narrative shift, group hallucination, and preference polarity—where models produce different or misleading narratives when the referenced social group changes. A multi‑agent mitigation framework is proposed, separating evidence extraction from group‑conditioned generation and using a counterfactual judge, which reduces narrative shift while maintaining chart‑grounded reasoning.
By Mizanur Rahman, Huan Wu, Arash Asgari, Enamul Hoque Prince, Laleh Seyyed-Kalantari
CONCAT is a training‑free framework that improves the efficiency of large language model (LLM) based multi‑agent systems by clustering agents according to their initial answers and selecting cluster leaders based on confidence. It uses a Theory‑of‑Mind‑inspired heuristic to predict collaboration benefits between leaders, then prunes communications to form an ad‑hoc network that reduces latency. Experiments on three LLMs and benchmarks show up to 2.02× higher accuracy/latency ratio than LLM‑Debate and a 50.1% latency reduction on Qwen2.5‑14B‑Instruct without task‑specific training.
By Ziyang Ma, Dingyi Zhang, Sichu Liang, Jiajia Chu, Pengfei Xia, Hui Zang, Deyu Zhou
Metric-Bench introduces a new benchmark for Vision‑Language Models (VLMs) that focuses on metric‑spatial reasoning in indoor scenes by using in‑image reference objects with known dimensions. The accompanying MetricReasoner fine‑tuning recipe employs structured prompts and numerical rewards to implicitly learn 2D‑to‑3D mapping without camera intrinsics. Experiments show that this approach improves spatial metric understanding by 43.1 % over existing models and boosts downstream embodied tasks, while also delivering gains on general VLM benchmarks.
By Yuling Xi, Haokai Zhang, Muzhi Zhu, Hao Zhong, Zongze Du, Hengyu Zhao, Chenchen Jing, Yufei Yin, Bin Qin, Yongjie Yang, Zhenbo Luo, Hao Chen, Chunhua Shen
The paper presents a framework that builds a static point cloud prior map from past camera traversals, augmenting each point with DINOv3 semantic features. During runtime, a local prior patch is retrieved, encoded with a sparse voxel backbone, and fused with lifted multi‑view camera features in bird’s‑eye view. This fused representation is then used by sparse transformer heads to predict 3D objects and vectorized map elements, achieving improved performance on Argoverse 2 without requiring LiDAR for prior‑map construction or online inference.
By Markus K\"appeler, Rohit Mohan, Abhinav Valada
arXiv:2609.26168v1 Announce Type: cross
Abstract: Recent work reports that vision--language models (VLMs) struggle to establish and maintain stable reference in repeated reference games. Rather than...
By Joseph Bingham
arXiv:2512.22760v2 Announce Type: replace
Abstract: Vision Transformers (ViTs) have achieved remarkable success in visual recognition tasks, but redundant token representations limit their computatio...
By Yunge Li, Lanyu Xu
arXiv:2609.22475v1 Announce Type: new
Abstract: Process discovery rarely yields a single coherent process structure. For analysis, a common step is to cluster process variants based on structural sim...
By Daniel Calegari, Daniel Amyot
arXiv:2609.22529v1 Announce Type: new
Abstract: International law provides the normative framework through which states coordinate action, regulate armed conflict, and protect human rights, yet its t...
By Genis Skura, Roland Bouffanais, Didier Wernli
arXiv:2609.22599v1 Announce Type: new
Abstract: Large language model (LLM) judges provide a flexible and scalable method for evaluating model and agent outputs, but their verdicts can be sensitive to...
By Jackson Hassell, Farima Fatahi Bayat, Pouya Pezeshkpour, Estevam Hruschka
arXiv:2609.22620v1 Announce Type: new
Abstract: Scanned mail, uploaded PDFs, and consolidated attachments often arrive as page streams that must be split into individual documents before downstream c...
By Nikhil Reddy Pottanigari, Sepideh Kharaghani, Saverio Vadacchino, Alejandro Posada, Ying Zhang
arXiv:2609.22878v1 Announce Type: new
Abstract: Large language model code generation benchmarks primarily evaluate well-resourced languages like Python and Java, where models benefit from abundant tr...
By Aditya Pola, Arkaprava Majumdar, Vineeth N. Balasubramanian
arXiv:2609.22987v1 Announce Type: new
Abstract: Automated operations research (OR) modeling requires LLMs to translate natural-language decision problems into correct mathematical programs. Existing...
By Ruiqing Zhao, Rui Liu, Yuan Zuo, Huarong Zhang, Xiao Han, Junjie Wu