ViSTA is a lightweight adapter that adds irregular numerical measurements to a pretrained vision‑language model’s chart representations, enabling accurate clinical time‑series prediction without altering the base model’s parameters. On the MIMIC‑IV dataset, ViSTA outperforms other adaptations across four metrics for acute kidney injury and mortality prediction, achieving an AUC of 0.7376 with only 0.516 million trainable parameters. It also delivers strong temporal question‑answering performance, reaching 69.27% accuracy with significantly fewer trainable parameters than low‑rank adaptation methods.
By Junyi Gao, Yu Shi, Pingzhao Hu, Ewen M Harrison
arXiv:2609. 31450v1 Announce Type: cross Abstract: Medical vision-language model (VLM) post-training is commonly evaluated through answer accuracy.
By Wang Jingxin
The paper "Can You Check That? The Checkability Boundary for Local LLM Network Automation" proposes a method called Touchstone that uses local small language models (SLMs) to generate network‑automation candidates and applies task‑specific intrinsic checks to reject incorrect outputs before escalating to a larger frontier LLM. By defining a task as checkable when a cheap, deterministic test can reject outputs violating a necessary correctness condition, the authors demonstrate that Touchstone achieves high end‑to‑end accuracy (98.6% on conflict detection and 93.8% on intent translation) while escalating only a small fraction of inputs. The study shows that local inference is preferable when precise, low‑cost checks are available, whereas tasks lacking such checks should rely on the frontier model.
By Maleeha Masood, Momina Nofal
The paper examines how evaluation design can cause significant fluctuations in the benchmark results of reasoning models, particularly the Deepseek‑R1‑Distill series. It shows that subtle changes in evaluation conditions lead to large variations in reported performance, a phenomenon also seen in other open‑source models fine‑tuned from Deepseek‑R1‑Distill and in the QwQ‑32B model. The authors call for a more rigorous evaluation paradigm and provide empirical assessments of the Deepseek‑R1‑Distill models.
By Yongfu Zhu, Lin Sun, Jinzhu Wu, Weihong Lin, Xiaoqi Jian, Guangxiang Zhao, Change Jia, Linglin Zhang, Sai-er Hu, Yuhan Wu, Xiangzheng Zhang
The paper introduces AttriBench, a benchmark dataset that balances author fame and demographics to study quote attribution in large language models (LLMs). Using AttriBench, the authors evaluate 11 popular LLMs and find that accurate attribution remains difficult, with significant disparities across race, gender, and intersectional groups. They also identify a new failure mode—suppression—where models omit attribution entirely, which is unevenly distributed across demographics and not reflected by standard accuracy metrics.
By Eliza Berman, Bella Chang, Daniel B. Neill, Emily Black
The paper introduces TokenAdapt, a model‑agnostic tokenizer transplantation method that uses a hybrid heuristic to initialize new token embeddings, and a novel pre‑tokenization learning approach for multi‑word Supertokens to improve compression. TokenAdapt combines local subword decomposition and global semantic similarity to preserve semantics while reducing retraining needs. Empirical results show that TokenAdapt outperforms existing baselines such as Transtokenizer and ReTok, achieving lower perplexity ratios and significant compression gains.
By Shaurya Sharthak, Vinayak Pahalwan, Adithya Kamath, Adarsh Shirawalmath
SLMFix is a code‑generation pipeline that uses a small language model fine‑tuned with reinforcement learning to correct syntactic errors in programs produced by large language models for domain‑specific languages. The approach relies on interpreter feedback to guide the error‑fixing process. Experiments show that SLMFix improves validator pass rates by 40% on low‑resource programming languages and removes over 50% of syntactic errors on high‑resource DSLs, outperforming supervised fine‑tuning even for 7B models.
By David Jiahao Fu, Aryan Gupta, Aaron Councilman, Yu-Xiong Wang, Vikram Adve
The Digital Consciousness Model (DCM) is introduced as a systematic, probabilistic framework for evaluating evidence of consciousness in AI systems, drawing on multiple leading theories of consciousness. Initial results indicate that, while the evidence suggests 2024 large language models (LLMs) are not conscious, this conclusion is not decisive and is weaker than evidence against consciousness in simpler AI systems.
By Derek Shiller, Laura Duffy, Arvo Mu\~noz Mor\'an, Adri\`a Moret, Chris Percy, Hayley Clatterbuck
The paper introduces CaC, a coarse‑to‑fine anomaly reward model that uses Vision‑Language Models to first scan globally for anomalous time windows, then ground anomalies spatially, and finally reason with structured spatiotemporal Chain‑of‑Thought. It builds the first large‑scale generated video anomaly dataset with detailed annotations and trains the model through a three‑stage progressive paradigm, including reinforcement learning with Group Relative Policy Optimization. Experiments show CaC improves fine‑grained anomaly detection by 25.7% and reduces generated‑video anomalies by 11.7% while enhancing overall video quality.
By Jiyuan Wang, Huan Ouyang, Jiuzhou Lin, Chunyu Lin, Dewen Fan, Boheng Zhang, Haonan Fan, Honglie Wang, Yiyang Fan, Zhenlong Yuan, Zijun Li, Yongrui Heng, Guosheng Lin, Fan Yang
The paper introduces PromptCCZSL, a framework that enables vision‑language models to continually learn new attributes, objects, and their unique compositions while avoiding forgetting. It uses a frozen VLM backbone with prompt‑based techniques, recency‑weighted multi‑teacher distillation, and several loss functions (CAL, OPL, IDL) to maintain prior knowledge and promote diverse, distinct embeddings. Experiments on UT‑Zappos and C‑GQA show significant performance gains over existing VLM‑based and non‑VLM baselines, establishing a new benchmark for continual compositional zero‑shot learning.
By Sauda Maryam, Sara Nadeem, Faisal Qureshi, Mohsen Ali
The paper evaluates a production text‑to‑SQL pipeline that uses an LLM as a judge, finding that the deployed gpt‑4o‑mini judge agrees with human annotators only weakly (Cohen’s kappa 0.04 on a disagreement‑enriched set and 0.42 on a random spot‑check). The authors identify a specific failure mode, GRADE‑HALLUCINATION, responsible for most over‑flags, and demonstrate that a self‑hosted Qwen3.6‑27B model achieves substantially higher agreement (kappa 0.72) at a much lower cost. They also show that ensembling judges does not improve performance, and that their audit method flags a significant portion of out‑of‑domain SQLs as potential issues.
By Haowei Liu, Hsin-Tai Wu, Yi Fang
The paper introduces a method called randomized intermediate guidance for training tandem speech-to-speech models, where a large language model (LLM) acts as a backend providing candidate responses while the user is speaking. Instead of simulating the backend’s guidance, the approach derives guidance directly from the conversation corpus, using target responses for informative guidance and randomly sampled responses to simulate irrelevant updates. Experiments on synthetic dialogues and 3.8k hours of real conversations show that this technique yields response quality comparable to LLM-generated baselines while improving natural turn‑taking and audio‑judge naturalness.
By Manato Yaguchi, Yotaro Kubo, Hikaru Asano, So Kuroki
ToolSearcher is a reinforcement learning framework designed to improve large‑scale tool selection for large language models. It introduces category‑constrained discrimination, event‑level search modeling, and trajectory‑aligned credit allocation to better distinguish similar tools, optimize multi‑turn search, and provide fine‑grained rewards. Experiments on large‑scale benchmarks show that ToolSearcher outperforms strong baselines in iterative search and complex tool composition scenarios.
By Zhenlong Dai, Xujie Song, Zitong Wang, Tong Niu, Jian liu, Weiqiang Wang, Xiu Tang, Sai Wu, Chang Yao, Jingyuan Chen
MDSkin-Net is a multi‑task skin lesion analysis framework that integrates Pattern Analysis priors into a hybrid CNN‑Transformer architecture. It introduces a Pattern Analysis‑Guided Attention Module (PAGAM) with improved Efficient Channel Attention, Multi‑Scale Spatial Attention, and Biased Asymmetry Attention, along with a multi‑scale spatial alignment regularization that uses segmentation masks as soft supervision. Trained only on the ISIC 2017 training split, the model achieves high segmentation and classification performance on multiple datasets, demonstrating strong zero‑shot generalization across different cohorts.
By Yijian Li, Saad Bedros, Paul Bigliardi, Mei Bigliardi Qi, Vassilios Morellas, Nikolaos Papanikolopoulos
STORM-Bench is a new benchmark for online video question answering that evaluates models’ ability to track state transitions and selectively abstain when visual evidence is insufficient. It contains 5,736 questions across 630 short, change‑dense episodes in five egocentric domains and two simulation subsets, with questions stratified by change intensity and answerability. The benchmark introduces STORM‑BR, a harmonic metric that reveals abstention failures and overconfidence on uncertain queries, showing that traditional accuracy masks gaps in epistemic reliability and state tracking.
By Siru Zhong, Shenghan Tan, Rihong Yan, Xiaohui Lv, Yuzheng Zhuang, Shuai Tao, Wulong Liu, Haohuan Fu, Yuxuan Liang
The paper introduces ReasonCVS, a structured reasoning framework for assessing the Critical View of Safety in laparoscopic cholecystectomy. It uses a Vision‑Language Model to build an Anatomical Scene Graph Abstraction and a Large Language Model–based Rationale‑Aware Reasoning Agent to verify sub‑criteria, producing a final verdict with traceable clinical rationale. Experiments on the Endoscapes‑CVS201 benchmark show ReasonCVS outperforms existing methods with a 68.1% mAP while offering interpretable, criterion‑level explanations.
By Qing Xu, Yuxiang Luo, Zhen Chen
InternW0-Δ is a unified World Action Model that integrates pretrained visual dynamics, scene semantics, 4D geometry, and motion priors within a Mixture-of-Transformers framework to generate robot actions. It leverages a frozen VLM for semantic guidance, a 4D foundation model for geometric priors, and introduces Causal Imprint to learn future-relevant scene changes without future-video rollout. The model is pretrained on a newly curated 20K‑hour heterogeneous corpus of robot and human demonstrations, achieving superior performance on simulation benchmarks and real‑robot platforms.
By Xingyu Miao, Zizun Li, Baole Fang, Kaiwen Song, Tenghui Wang, Hanxue Zhang, Yating Wang, Xudong Li, Yuping He, Xueyuan Wei, Chao Gao, Xijie Yang, Yingxiang Xu, Kerui Ren, Wenqi Guo, Jianjun Zhou, Xinzhe Wang, Weiguang Zhao, Ni Yang, Zetao Cai, Yufei Xue, Hengjie Li, Zeyu He, Yuanzhen Zhou, Rong Fu, Jianyang Zhang, Siwei Cui, Fuxian Huang, Yunsong Zhou, Xing Gao, Yifei Yao, Qiaojun Yu, Kailin Li, Ming Zhou, Mu Huang, Xinyue Li, Wenze Cui, Bingqi Jiang, Xueyue Zhu, Junting Dong, Haoyu Guo, Tao Lu, Mulin Yu, Bowen Zhou, Bin Zhao, Tianfan Xue, Weinan Zhang, Chunhua Shen
FLIP is a final‑layer inference‑time probe designed to test whether a logit‑facing intervention site in an open‑weight vision‑language model (VLM) supports structured, task‑linked computation rather than generic perturbation. The probe applies elementwise flooring to the final normalized hidden state before logit computation, leaving other model components unchanged. By sweeping intervention strength on a controlled detection/counting task, FLIP identifies three regimes—negligible change, a bounded interior regime with improved detection recall and reduced counting error, and over‑suppression—while a four‑criterion protocol ensures the observed effects are mechanistically interpretable.
By Drandreb Earl O. Juanico, Rowel O. Atienza
ActKV is a new KV cache compression framework designed for agentic large language model (LLM) inference. It prioritizes cache entries that contribute to action generation, using action-oriented eviction, confidence-driven budget allocation, and page-aware compression to reduce memory usage while preserving accuracy. In long-trace tasks, ActKV retains 98.53% of FullKV’s accuracy using only 25.98% of its peak memory and boosts token and task throughput by 3.97× and 3.58×, respectively.
By Zihan Wang, Cheng Tang, Lei Gong, Chao Wang, Wenqi Lou, Teng Wang, Xuehai Zhou
SkillFlow is an open, multi-stage retrieval system that helps AI agents selectively load relevant skills from a large library of community-contributed SKILL.md definitions. The pipeline uses dense retrieval, two rounds of cross-encoder reranking, and LLM-based selection to balance recall and precision. Evaluations on SkillsBench and Terminal-Bench show that SkillFlow improves performance when high-quality skills are available, but retrieval alone does not help if the corpus lacks executable skills for the target domain.
By Fangzhou Li, Pagkratios Tagkopoulos, Ilias Tagkopoulos