πŸ“„ Notable* Recent AI/ML arXiv Papers

Last updated just now...

πŸ“„ Ambient LLMs: Data-Efficient Supervised Fine-tuning via Temperature Matching
πŸ—“οΈ Published: 11/17/2026
πŸ”— http://arxiv.org/abs/2611.23229v1
πŸ‘₯ Authors: Giannis Daras, Marios Papachristou, Adam Klivans, Antonio Torralba (possible past Massachusetts Institute Of Technology affiliation), Constantinos Daskalakis (possible past University Of California, Berkeley affiliation)
Abstract

Collecting large-scale, diverse, and high-quality post-training datasets for autoregressive language models is challenging. Consequently, Supervised Fine-Tuning (SFT) is typically applied to datasets significantly smaller than those used for pre-training, which leads to overfitting and the catastrophic forgetting of the tails of the data distribution. To prevent this collapse, we introduce Ambient SFT, a principled framework for leveraging imperfect predictions to augment the fine-tuning set. In...

πŸ“„ Reading Scene States: Uncovering Safety Failures in Vision-Language Models
πŸ—“οΈ Published: 11/17/2026
πŸ”— http://arxiv.org/abs/2611.23206v1
πŸ‘₯ Authors: Tongxi Wu, Yuyang Cheng, Jiacheng Xian, Menglin Yang, Kai Chen (possible past Shanghai Jiao Tong University affiliation), Jian Zhang (possible past Tencent (China) affiliation), Yang Gao (possible past Tencent (China) affiliation)
Abstract

Vision-language models (VLMs) can provide harmful assistance by interpreting visual events, even without an explicit request for the underlying harmful objective. We identify this \textbf{scene-state safety vulnerability} and introduce \textbf{SceneSafetyBench}, comprising 375 image--question pairs spanning Initial, Medium, and Ultimate scene states and five visual presentations. Across eight VLMs, scene-state inputs elicit harmful assistance with attack success rates (ASRs) of 45.6--72.3\%. Com...

πŸ“„ PeriaTrace: Tooth-Level Grounding Diagnosis and Patient-Facing Explanations from Periapical Radiographs
πŸ—“οΈ Published: 11/17/2026
πŸ”— http://arxiv.org/abs/2611.23012v1
πŸ‘₯ Authors: Zhenhuan Zhou, Yuchen Zhang (possible past University Of California, Berkeley affiliation), Jingbo Zhu, Shiqing Ma, Yanlin Wu, Tao Li (possible past Baidu (China) affiliation)
Abstract

Periapical radiographs (PRs) are central to endodontic diagnosis, but interpreting them and preparing reports are time-consuming. Technical language can limit patients' understanding, while existing automated methods often lack tooth-level links between findings and image evidence. We present PeriaTrace, a three-stage framework integrating tooth-level diagnosis, visually grounded report generation, and patient-friendly explanations and care recommendations. PeriaTrace was developed and evaluated...

πŸ“„ When Do Typed Decision Models Work in Transportation? Evidence from Travel Choice and Traffic Signal Control
πŸ—“οΈ Published: 11/17/2026
πŸ”— http://arxiv.org/abs/2611.22926v1
πŸ‘₯ Authors: Hongyang Zhang, Boyuan Deng, Siyuan Feng (possible past Carnegie Mellon University affiliation), Kai Chen (possible past Shanghai Jiao Tong University affiliation), Junbo Zhang
Abstract

Typed decision models combine natural-language understanding with a structured interface that returns a selected action and an option-level probability distribution without generating lengthy textual responses. Their low inference cost and response latency make them attractive for transportation problems with predefined alternatives. This paper investigates their effectiveness, robustness, and efficiency in two representative transportation tasks: individual travel-mode choice prediction and seq...

πŸ“„ Verifying Agents in Rubric-Graded Environments
πŸ—“οΈ Published: 11/17/2026
πŸ”— http://arxiv.org/abs/2611.22841v1
πŸ‘₯ Authors: Markus DΓΌcker, Vaibhav Kumar, Yi Liu (possible past Google (United States) affiliation), Ronak Chaudhary, Andreas Plesner, Francisco GuzmΓ‘n (possible past Meta (United States) affiliation), Anish Athalye
Abstract

As AI agents take on increasingly open-ended and complex tasks, verifying their outputs has proven to be difficult. Rubric-graded agent environments, where a verifier judges multiple natural-language criteria against the deliverables and environment state changes produced by the agent, have emerged as a popular paradigm. We conduct the first systematic study of verifiers for rubric-graded agent environments. First, we create BankerVerifierBench (BVB), a meta-evaluation dataset of 2,464 human-jud...

πŸ“„ Right Diagnosis, Unsafe Route: Auditing Clinical AI Decision Trajectories and Corrective Oversight for Validity and Potential Harm
πŸ—“οΈ Published: 11/17/2026
πŸ”— http://arxiv.org/abs/2611.22612v1
πŸ‘₯ Authors: Guanqiao Chen, Wei Jiang (possible past Apple (United States) affiliation), Shu Zhang (possible past Google (United States) affiliation), Qian Lin, Guanxu Yu, Siran Li, Chunhui Wang, Tianyu Shi
Abstract

A clinical AI system can reach a correct final diagnosis through intermediate decisions that could have harmed the patient, and can repair an error while introducing new risks. Most evaluations score final diagnoses; even stepwise benchmarks, which release a case's evidence over time, score whether systems reach the right diagnosis or gather the right information, not whether the decisions a patient passes through are safe. We make patient safety the target of stepwise evaluation. A trajectory a...

πŸ“„ Motion-Aware Reward Modeling for Instruction-Guided Video Editing
πŸ—“οΈ Published: 11/17/2026
πŸ”— http://arxiv.org/abs/2611.22495v1
πŸ‘₯ Authors: Li Li (possible past Google (United States) affiliation), Xiangbo Gao, Bo Ni, Samyadeep Basu, Viet Dac Lai, Jing Shi, Seunghyun Yoon, Gang Wu, Yu Wang (possible past Tsinghua University affiliation), Thien Huu Nguyen, Tyler Derr, Franck Dernoncourt, Zhengzhong Tu (possible past Google (United States) affiliation), Yue Zhao
Abstract

Reward models for instruction-guided video editing can favor temporal consistency at the expense of motion. We study this failure, termed \emph{motion collapse}, using controlled motion-suppression variants of real edits. On VE-Bench, the evaluated composite reward prefers a motion-suppressed candidate in $97\%$ of cases. Learned video rewards and a VLM judge also struggle with mild suppression, while latent optimization and reward-guided fine-tuning exploit this weakness to increase reward as m...

πŸ“„ SciWorkBench: A Benchmark for Real Scientific Workflows
πŸ—“οΈ Published: 11/17/2026
πŸ”— http://arxiv.org/abs/2611.22490v1
πŸ‘₯ Authors: Yuchen Zhang (possible past University Of California, Berkeley affiliation), Taolin Han, Jinghang Wang, Jinxin Wang, Yun Wu, Wai Yuet Chiu, Xiaogang Li, Yuting Fu, Kewei Sun, Jiaxin Li, Qile Jin, Jiajia Li, Ruodan Chen, Qingteng Chen, Jialu Nie, Junchao Li, Yifei Zhang, Chen Zhao (possible past Stanford University affiliation), Hu Wei, Lin Qu, Yanni Yin, Bing Zhao
Abstract

Language models are increasingly being used as agents capable of planning and executing projects, as well as refining their actions through feedback. This shift raises a question: can current agents complete projects like real scientists? To examine this question, we introduce SciWorkBench, a benchmark for evaluating agents' abilities on real science tasks across biology, chemistry, materials science, and physics. SciWorkBench is built from peer-reviewed literature or expert-contributed research...

πŸ“„ From Ideal to Real: Integrating Language Models into Smart Home for Human-Centric IoT Safety and Security Fuzzing
πŸ—“οΈ Published: 11/17/2026
πŸ”— http://arxiv.org/abs/2611.22457v1
πŸ‘₯ Authors: Lin Li, Xinbo Ban, Chao Chen (possible past Tencent (China) affiliation), Lei Pan, Ming Ding (possible past Tsinghua University affiliation), Kok-Leong Ong, Jun Zhang (possible past Tencent (China) affiliation)
Abstract

Integrating IoT devices into household systems offers enhanced control over living environments. However, as these devices become pervasive in critical sectors such as healthcare, transportation, and home automation, they expose inter-rule vulnerabilities that may lead to privacy breaches, data theft, and physical harm. Traditional fuzz testing methods, which often operate in simulated environments assuming ideal human decisions, fail to account for the complexities of human interactions that ca...

πŸ“„ Multistage Contextual Optimization for Vehicle Repositioning
πŸ—“οΈ Published: 11/17/2026
πŸ”— http://arxiv.org/abs/2611.22330v1
πŸ‘₯ Authors: Jinyu Zhang, Di Huang (possible past Google (United States) affiliation), Zhiyuan Liu (possible past Tsinghua University affiliation), Nikolas Geroliminis
Abstract

Ride-hailing platforms reposition idle vehicles to improve spatial supply-demand balances. This problem is prediction-driven and multistage: at each stage, the platform forecasts future demand, executes current decisions, and uses the resulting state to initialize the next stage. We formulate a multistage linear program (LP) with a macroscopic fundamental diagram (MFD) linking accumulation to trip completion. We then develop a multistage contextual optimization approach that uses decision perfor...

πŸ“„ EviDARL: Evidence-Centric Dual-Agent Reasoning via Alternating Reinforcement Learning for Multi-Table Question Answering
πŸ—“οΈ Published: 11/17/2026
πŸ”— http://arxiv.org/abs/2611.22310v1
πŸ‘₯ Authors: Dunhan Li, Leilei Gan, Lirong Gao, Wenyan Fan, Yongwei Wang, Sheng Guo (possible past Google (United States) affiliation), Fei Wu (possible past Google (United States) affiliation)
Abstract

Despite advances in LLM-based table reasoning, most studies focus on single-table settings. Multi-table QA requires multi-hop reasoning over complex table relationships and evidence dispersed across long contexts, often without verifiable intermediate traces. To address these challenges, we introduce EviDA, an evidence-centric dual-agent framework comprising an Evidence Construction Agent and a Reasoning and Answering Agent. The Evidence Construction Agent generates and executes table-manipulati...

πŸ“„ When the World Is the Test Suite: Agentic Intelligence for the Physical World
πŸ—“οΈ Published: 11/17/2026
πŸ”— http://arxiv.org/abs/2611.22291v1
πŸ‘₯ Authors: Kaixin Li, Meng Luo, Yanlin Li, Yuchen Tian, Xiang Liu, Xiangru Jian, Yan Yang (possible past Google (United States) affiliation), Zhiheng Xi, Yang Zhang (possible past Tsinghua University affiliation), Ziyang Luo, Mingfei Cheng, Hongzhan Lin
Abstract

Coding agents built to repair software now write, test, and repair programs for robots, laboratories, and industrial controllers. Unlike in software engineering, no test suite comes with the code. Whether a program works must be established by running it on a robot, in a physics simulator, or through the policies it helps to train, and such evidence is costly to obtain and can itself be wrong. As these agents take on more physical tasks, how their programs are judged matters as much as how they ...

πŸ“„ SeasonRainBench: Global Seasonal Precipitation
πŸ—“οΈ Published: 11/17/2026
πŸ”— http://arxiv.org/abs/2611.22281v1
πŸ‘₯ Authors: Qiuyan Sun, Mengxuan Chen, Haoyuan Liang, Guowen Li, Zhen Huang, Jianhao Lin, Juepeng Zheng (possible past Tsinghua University affiliation), Haohuan Fu (possible past Tsinghua University affiliation)
Abstract

Accurate seasonal precipitation forecasts help societies prepare for drought, flooding, and crop failure months in advance. While data-driven methods have made rapid progress in weather and subseasonal forecasting, operational seasonal precipitation prediction still relies on numerical ensemble systems. These systems are computationally expensive and exhibit biases and drift that increase with lead time, motivating more efficient data-driven alternatives. Yet this transition faces three linked c...

πŸ“„ RAG in a Trenchcoat: When Minimal Memory Is Enough for Agentic Systems, and When It Isn't
πŸ—“οΈ Published: 11/17/2026
πŸ”— http://arxiv.org/abs/2611.22179v1
πŸ‘₯ Authors: Jingyu Liu, Zongze Li, Zach Xu, Zhanhui Zhou, Tahseen Rabbani, Dawn Song (possible past University Of California, Berkeley affiliation), Ce Zhang (possible past Eth Zurich affiliation)
Abstract

Agentic memory research has accumulated extraction, supersession, reflection, and graph synthesis on top of retrieval-augmented generation. However, we found that almost none of those primitives earn their cost in our tested memory benchmarks: we show this with BRIDGE, a deliberately minimal RAG memory with chained observation notes, hybrid BM25 with cosine retrieval, and chronological rendering, which reaches parity within bootstrap CI with the prior SOTA on LongMemEval-S and BEAM, and beats th...

πŸ“„ TeleChat to Xing4: Advancing Frontier Agentic Models on Domestic Chips and Frameworks
πŸ—“οΈ Published: 11/17/2026
πŸ”— http://arxiv.org/abs/2611.22115v1
πŸ‘₯ Authors: Xingchen Agi Lab, :, Baotian Hu, Bingkai Yang, Boyang Wang, Changzai Pan, Chao Lyu, Chao Wang (possible past Google (United States) affiliation), Chen Li (possible past Tencent (China) affiliation), Chenshuo Pan, Chenyu Lin, Chenyuan Zhang, Chongming Liu, Dawei Fan, Di Wu, Dong Guo, Dong Li, Dongfang Li, Enlin Liu, Fanye Xiao, Fengkai Li, Fubei Yao, Gang Lin, Han Gao (possible past Tencent (China) affiliation), Hanming Wu, Haowei He, Huapeng Liu, Hua Wang, Huinan Xu, Jiahong Xiong, Jiaxin Peng, Jianxiang Ma, Jianxing Zhuge, Jie Zhang, Ji Xiong, Jinglin Chen, Jingqi Li, Jingxuan Li, Jiwang Ye, Junchen Li, Junbiao Li, Junyang Wu, Kaidong Yu, Kaipeng Jia, Kehai Chen, Kuai Li, Laobin Liu, Lei Li (possible past Carnegie Mellon University affiliation), Libo Qin, Lingling Shi, Luwen Pu, Menghao Xu, Mengjiao Bao, Mengqi Chen, Min Zhang (possible past Tsinghua University affiliation), Mingrui Wang, Naying Wang, Peicong Han, Peiming Guo, Pengfei Li, Qiang Feng, Qi Wang (possible past Tsinghua University affiliation), Qingjun Zhu, Qingli Li, Qitai Ding, Qiyi Xie, Quanchao Zhao, Rongbao Han, Rui Xue, Ruiyu Fang, Ruiting Song, Shijie Li, Shiquan Wang, Shuangyong Song, Sishi Xiong, Teng Su, Wankang Bao, Weijie Zhang, Wei Mao, Weiwei Fu, Weiyang Huang, Wenmin Deng, Wenqing Hu, Wujie Zhong, Xi Yu, Xiaohui Hu, Xiaomin Fang (possible past Baidu (China) affiliation), Xiaomeng Huang, Xiaoqiong Shi, Xiaotong Shang, Xiaoyu Li (possible past Tencent (China) affiliation), Xinglei Xu, Xin Jiang, Xin Wang (possible past University Of Edinburgh affiliation), Xin Zhang (possible past Google (United States) affiliation), Xinzhang Liu, Xue Liu, Xuefeng Bai, Xuewei He, Yahan Li, Yanhan Zhang, Yanlei Xue, Yanzhang Wu, Yang Yu, Yi Liu (possible past Google (United States) affiliation), Yin Zhang, Yingyan Li, Yitong Yao, Yixuan Li (possible past Meta (United States) affiliation), Yongqiang Xia, Yongxiang Li, Yinzhou Zhu, Yu Zhao (possible past Tencent (China) affiliation), Yuhao Jiang, Yunpeng Zhang, Yuting Lin, Yuxuan Wu, Yuyao Huang, Yuyi Wang, Zebo Peng, Zhenwu Pei, Zhenxiong Li, Zhao Wang, Zhaoxi Zhang, Zhihao Yang, Zhiqiang Dong, Zhiwei Yan, Zhilong Lu, Zhonghao Guo, Zhongheng Wu, Zhongjiang He, Zhongqiu Li, Zhu Yuan, Zhuoru Jiang, Zhuoyi Wei, Zijia Guo, Zilu Huang, Zhipeng Guo, Zirui Li, Zihan Wang (possible past Tsinghua University affiliation)
Abstract

We present Xing 4, a series of large language models that extends the TeleChat lineage through coordinated advances in model architecture and training infrastructure on Ascend. This report describes the design principles and systems engineering behind the series, spanning architecture selection, large-scale pretraining, and reasoning and agentic post-training. We study how sparse activation, attention design, model depth and width, and speculative decoding balance model capacity with computation...

πŸ“„ InfuseLoop: Cross-Trajectory Latent Injection for Scalable Parallel Looped Transformers
πŸ—“οΈ Published: 11/17/2026
πŸ”— http://arxiv.org/abs/2611.22426v1
πŸ‘₯ Authors: Zhaofeng Sun, Lanxiang Hu, Yu-Yang Qian, Jiaqi Leng, Yuanbo Yang, Chi Zhang (possible past Peking University affiliation), Youyou Lu, Peng Zhao, Hao Zhang (possible past Tencent (China) affiliation)
Abstract

Parallel Looped Transformers (PLTs) use otherwise idle decoding compute by overlapping recurrent iterations. However, vanilla PLTs scale poorly with depth: they exhibit training-loss spikes, and their quality gains shrink as more loops are added. The benefit of additional loops depends on the information available to each iteration, so we examine whether the existing computational dependencies limit effective depth scaling. We find that only the preceding in-trajectory state is passed across loo...

πŸ“„ Towards Optimism-Pessimism Trade-off in Model-based Offline-to-Online Reinforcement Learning
πŸ—“οΈ Published: 11/17/2026
πŸ”— http://arxiv.org/abs/2611.22172v1
πŸ‘₯ Authors: Guochen Zhou, Yijun Yang, Qiqi Duan, Qing Su, Weiming Ou, Li Shen (possible past Tencent (China) affiliation), Shihao Ji, Yuhui Shi, Zhisong Zhang (possible past Shanghai Jiao Tong University affiliation)
Abstract

Model-based offline-to-online reinforcement learning (RL) enables sample-efficient adaptation by leveraging offline pre-training for online fine-tuning. However, the distribution shifts between offline and online stages often hinder fine-tuning performance. Many existing methods encode the optimism-pessimism trade-off through a single-objective formulation. Selecting its trade-off coefficient through online fine-tuning can require costly bi-level optimization. We identify the balance established...

πŸ“„ KVUse: Reusable-KV User Memory for Efficient LLM Recommenders
πŸ—“οΈ Published: 11/16/2026
πŸ”— http://arxiv.org/abs/2611.21920v1
πŸ‘₯ Authors: Xin Guo, Qiang Zhang (possible past Tsinghua University affiliation), Xiang Zhou (possible past Google (United States) affiliation), Arushi Rai, Siyu Lin, Yifan Wu (possible past Carnegie Mellon University affiliation), Xiang Shen, Lei Huang, Qunshu Zhang, Xiangjun Fan, Lizhu Zhang
Abstract

Memory-augmented recommenders built on large language models (LLMs) maintain a natural-language profile of each user. These profiles are updated by a user profile update path that revises the profile as new interactions arrive and are consumed by downstream task paths like ranking. Because these user profiles are stored as text in the user memory, every ranking LLM call that consumes them must splice them in the prompt and calculate the KV cache of them during the prefill phase. Given these rank...

πŸ“„ tmol: a GPU-accelerated implementation of the Rosetta energy function for machine learning
πŸ—“οΈ Published: 11/16/2026
πŸ”— http://arxiv.org/abs/2611.21857v1
πŸ‘₯ Authors: Andrew Leaver-Fay, Kieran Didi, Jeff Flatten, Alex Ford, Guangfeng Zhou, Nathaniel Corley, Hahnbeom Park (possible past University Of Washington affiliation), Maxwell W. Kazman, Rafael I. Brent, Joseph Kleinhenz, Henry Solberg, Andrew M. Watkins, Brian Kuhlman, David Baker (possible past University Of Washington affiliation), Frank Dimaio (possible past University Of Washington affiliation)
Abstract

Physics-inspired forcefields complement generative machine learning models, showing better generalization beyond the training distribution. However, these forcefields are typically slow to evaluate, and their derivatives are difficult to integrate into ML frameworks. We introduce tmol, a GPU-native PyTorch library implementing the state-of-the-art RosettaGenFF forcefield. tmol supports efficient batched differentiable scoring and whole-structure refinement with up to 362-fold throughput increase...

πŸ“„ VisHead-Spec: Head-Routed Visual Attention for Multimodal Speculative Decoding
πŸ—“οΈ Published: 11/16/2026
πŸ”— http://arxiv.org/abs/2611.21839v1
πŸ‘₯ Authors: Xin Guo, Arushi Rai, Xiang Zhou (possible past Google (United States) affiliation), Qiang Zhang (possible past Tsinghua University affiliation), Siyu Lin, Xiangjun Fan, Lizhu Zhang
Abstract

Speculative decoding accelerates vision-language models(VLMs) while preserving the target model's outputs, but its speedup diminishes under high concurrency serving as the system becomes more compute bounded as the batch size increases. This problem is pronounced for multimodal speculative decoding as the long range of the image tokens in the prompt can significantly increase the computational cost during draft token prediction. The block-diffusion drafters (e.g. DFlash, DSpark) can effectively ...

πŸ“„ Regularized Score Centering
πŸ—“οΈ Published: 11/16/2026
πŸ”— http://arxiv.org/abs/2611.21791v1
πŸ‘₯ Authors: Hao Bai, Chenlu Ye, Xuanchang Zhang, Aviral Kumar (possible past University Of California, Berkeley affiliation), Tong Zhang (possible past Tencent (China) affiliation)
Abstract

Asynchronous reinforcement learning increases training throughput, but data from stale policies and numerical mismatch between training and inference can destabilize learning. To stabilize off-policy learning, score centering (SC) removes the effect of reward baselines on the expected update. Our gradient diagnostics show less extreme gradients with SC than with importance sampling (IS), but SC can still fail when the learner moves far from the behavior policy that generated the data. To limit t...

πŸ“„ Latent Lattice Language Models
πŸ—“οΈ Published: 11/16/2026
πŸ”— http://arxiv.org/abs/2611.21608v1
πŸ‘₯ Authors: Shivam Duggal, Zongze Wu, William T. Freeman (possible past Massachusetts Institute Of Technology affiliation), Antonio Torralba (possible past Massachusetts Institute Of Technology affiliation), Richard Zhang, Eli Shechtman
Abstract

Language models typically increase per-token computation through deeper or looped layers and latent reasoning steps, while richer residual connections improve information flow across layers. We instead increase computational capacity: the number of active computational states a model maintains. We introduce the Lattice of Latents, a set of per-layer horizontal latent streams that run recurrently along time, alongside the vertical residual stream that flows through depth. Like hyper-connections, ...

*Notable papers are those with at least two authors from a "big" AI/ML lab.