πŸ“„ Notable* Recent AI/ML arXiv Papers

Last updated just now...

πŸ“„ Gamer: A Generalist Vision-Language-Action Foundation Model for Long-Horizon Real-Time Game Agent
πŸ—“οΈ Published: 11/13/2026
πŸ”— http://arxiv.org/abs/2611.18397v1
πŸ‘₯ Authors: Run Luo, Kuan Zhang, Qi Deng, Yukun Chen, Zhihao Yang, Yufei Zhang, Renrong Hu, Yi Zhang (possible past Google (United States) affiliation), Guiyu Ma, Tianyu Fan, Longze Chen, Hao Sun, Jiaming Li, Jiaming Yang, Xiaohao Liu, Tao Tan, Minzheng Wang, Hangtian Jia, Shuai Yuan, Lvdi Wang, Xiao Liu (possible past Baidu (China) affiliation), Chao Ma (possible past Shanghai Jiao Tong University affiliation), Yuting Zou, Yinhe Zheng, Xiaobo Xia, Dawei Jiang, Shaoqing Ren
Abstract

We introduce Gamer, a generalist agent framework built for long-horizon tasks in complex interactive environments. Its end-to-end architecture combines perception, memory management, and multi-step reasoning. A shared training and deployment pipeline supports interaction with both conventional 2D graphical interfaces and complex 3D simulated worlds. We first continue pre-training on a dataset constructed from tens of thousands of hours of device-level keyboard and mouse interaction. These trajec...

πŸ“„ Envisioning Agentic Robotics: A New Era of Robotic Systems
πŸ—“οΈ Published: 11/13/2026
πŸ”— http://arxiv.org/abs/2611.18359v1
πŸ‘₯ Authors: Grid Robotics, :, Qingfeng Lan, Duo Li, Siyao Li, Yiheng Lyu, Shijian Lu, Zhenghao Qi, Shizun Wang, Xinchao Wang, Ming Yang (possible past Meta (United States) affiliation), Zhiqin Yang, Zuhao Yang, Tao Yu (possible past University Of Washington affiliation)
Abstract

The release of GPT-6 Astra marks the emergence of a new paradigm in robotics: agentic robotics. Within weeks, frontier models with agentic harnesses matched or exceeded task-specific policies on many embodied benchmarks. Instead of training a dedicated policy for each family of tasks, an agentic robotic system applies a foundation model to open-world embodied tasks. It operates in a closed loop: observing the environment, interpreting user intent, reasoning about the task, invoking skills and to...

πŸ“„ RACE: Ranking-Uncertainty Aware Capacity-Adaptive Customization for Efficient Sequential Recommendation
πŸ—“οΈ Published: 11/13/2026
πŸ”— http://arxiv.org/abs/2611.18244v1
πŸ‘₯ Authors: Tianqi Liu, Kaipu Sun, Kairui Fu, Wenkai Wang, Sihao Liu, Hui Zang, Ran Chen, Tao Wang (possible past Stanford University affiliation), Shengyu Zhang (possible past Tencent (China) affiliation)
Abstract

Efficient sequential recommendation has emerged as a promising paradigm for providing real-time personalized service. To deploy models under resource-constrained environment, model compression becomes a necessity, yet existing methods typically apply uniform compression ratios, overlooking the heterogeneity in user intent complexity. This "one-size-fits-all" approach leads to a fundamental Capacity-Complexity Mismatch, where computational resources are either wasted on simple instances or insuff...

πŸ“„ TelePhysica: Agentic Reconstruction of Physically Interactive Scenes
πŸ—“οΈ Published: 11/13/2026
πŸ”— http://arxiv.org/abs/2611.18078v1
πŸ‘₯ Authors: Ruiqi Shu, Xufeng Zhan, Yuanzhi Liang (possible past Baidu (China) affiliation), Haibin Huang, Chi Zhang (possible past Peking University affiliation), Xuelong Li (possible past Tencent (China) affiliation)
Abstract

Direct image-to-video diffusion models entangle pixel synthesis with implicit latent motion, precluding persistent 3D mechanical equilibrium, mid-rollout intervention, and counterfactual force control. Conversely, single-pass programmatic 3D reconstruction from an uncalibrated photograph leaves monocular depth and support clearance underconstrained, causing reconstructed scenes to interpenetrate or collapse under gravity. We present TelePhysica, a closed-loop framework that reconstructs an execu...

πŸ“„ From Images to Grounded Edits: A Real-World Benchmark for Chinese Grammatical Error Correction in Handwritten Essays
πŸ—“οΈ Published: 11/13/2026
πŸ”— http://arxiv.org/abs/2611.17966v1
πŸ‘₯ Authors: Weiru Zhang, Yuzhong Hong, Junwei Bao, Lemao Liu (possible past Tencent (China) affiliation), Dejing Dou (possible past Baidu (China) affiliation)
Abstract

Chinese Grammatical Error Correction (CGEC) is typically formulated as a text-to-text problem. While suitable for digital texts, this formulation falls short in real-world K-12 education, where AI-assisted corrections are expected directly on essay images, which are primarily handwritten and exhibit many irregularities. To bridge this gap, we introduce Grounded Chinese Grammatical Error Correction (G-CGEC), an image-to-grounded-edit task requiring systems to predict both corrections and their po...

πŸ“„ WebArena-Pro: A Heterogeneous, Multimodal, Reproducible Benchmark for Web Agents
πŸ—“οΈ Published: 11/13/2026
πŸ”— http://arxiv.org/abs/2611.17955v1
πŸ‘₯ Authors: Imene Kerboua, Fatemeh Pesaran Zadeh, Xing Han LΓΉ, Weijian Qi, Alexander Miller (possible past Meta (United States) affiliation), Junyi Song, Yunjia Tian, Dongjin Kang, Seyeon Choi, Marzia Nouri, Ewen Gueguen, Matteo Boglioni, Fengyuan Liu, Zeyi Liao, Mengqi Yuan, Yue Li, Alexandre Lacoste (possible past Google (United States) affiliation), Alexandre Drouin, Spandana Gella (possible past University Of Edinburgh affiliation), Huan Sun, Gunhee Kim, Siva Reddy (possible past University Of Edinburgh affiliation)
Abstract

Web agents powered by large language and vision-language models are increasingly applied to realistic browser work that spans heterogeneous applications, multimodal content, and stateful workflows. However, existing reproducible web-agent benchmarks cover only a small number of web applications drawn from a few software categories, and restrict modality to text and vision. Live benchmarks broaden site coverage but sacrifice reproducibility, since pages and data drift between runs. Moreover, exis...

πŸ“„ RecMem: Recurrent Memory Compression for Long-Sequence Generative Recommendation
πŸ—“οΈ Published: 11/13/2026
πŸ”— http://arxiv.org/abs/2611.17921v1
πŸ‘₯ Authors: Haoyi Hu, Gaoyang Guo, Yunjia Xi, Wen Chen, Weiwen Liu, Jian Wu (possible past Tencent (China) affiliation), Yuning Jiang, Weinan Zhang (possible past Shanghai Jiao Tong University affiliation)
Abstract

Generative recommendation benefits from long interaction histories, but the cost of processing them directly grows with history length. We propose RecMem, a recurrent memory compression framework guided by three desiderata: space invariance, semantic fidelity, and lifelong evolvability. RecMem compresses historical segments into a fixed set of memory slots using a bifurcated encoder-decoder with shared lower layers and slot-wise forget-gated updates. Recommendation supervision trains the memory ...

πŸ“„ When Cooperation Must Change: Semantic Strategy Unlearning in Multi-Agent Reinforcement Learning
πŸ—“οΈ Published: 11/13/2026
πŸ”— http://arxiv.org/abs/2611.17917v1
πŸ‘₯ Authors: Wenhan Wu, Linghan Chen, Yudong Gao, Fan Yang (possible past Tencent (China) affiliation), Honglong Chen, Kai Yin, Jiawei Jiang (possible past Tencent (China) affiliation), Chuang Hu
Abstract

Cooperative multi-agent reinforcement learning (MARL) supports coordinated decision-making in distributed Web services, multi-robot control, and collaborative games through learned coordination conventions. When deployment requirements change, a previously useful convention may become obsolete or unsafe. For example, revised cache-sharing permissions in Web content delivery may require caches to abandon a learned division of storage and serving roles while maintaining efficient delivery. Teams m...

πŸ“„ SWIFT: Sample-Wise Instance Fine-Tuning of Time-Series Foundation Models for Forecasting
πŸ—“οΈ Published: 11/13/2026
πŸ”— http://arxiv.org/abs/2611.17915v1
πŸ‘₯ Authors: Yang Chen (possible past Tencent (China) affiliation), Feifan Tian, Cong Wang, Cheng Feng, Wenbin Li, Yang Gao (possible past Tencent (China) affiliation)
Abstract

While current few-shot adaptation strategies for time series forecasting foundation models are predominantly dataset-centric, the necessity of obtaining an additional dataset for the finetuning process poses a significant hurdle for adoption. To address this constraint, we propose Sample-Wise Instance Fine-Tuning (SWIFT), a novel forecasting adaptation framework that achieves fine-grained specialization by leveraging structural priors from the target instance's own history. This shift allows the...

πŸ“„ Can Agents Keep Up with Time? An Empirical Analysis of Temporal Adaptation After a Conversational Hiatus
πŸ—“οΈ Published: 11/13/2026
πŸ”— http://arxiv.org/abs/2611.17806v1
πŸ‘₯ Authors: Hongguan Su, Ming Wang, Shun Zhang, Yifan Wang (possible past Stanford University affiliation), Rong Yin, Zhiming Ma, Mengzheng Yang, Peng Chen (possible past Tencent (China) affiliation)
Abstract

Long-horizon agents increasingly serve as persistent personal assistants across modern web platforms, managing schedules, applications, and multi-step plans over weeks or months. Users often return to an ongoing conversation after a long conversational hiatus, while earlier bookings and preparation notes stay on record. During this hiatus, scheduled dates may pass even though the retained records remain unchanged and event outcomes go unrecorded. Existing evaluations of long-horizon agents focus...

πŸ“„ Vision-as-a-Tool: Visual Harness for Multimodal Agentic Coding
πŸ—“οΈ Published: 11/13/2026
πŸ”— http://arxiv.org/abs/2611.17797v1
πŸ‘₯ Authors: Haolin Yang, Wenshan Wu, Jin Jiang, Yu Cheng (possible past National University Of Singapore affiliation), Yan Xia, Ting Song, Yangyu Huang, Zhiliang Peng, Wenhui Wang, Hangbo Bao, Yuzhe Zhang, Wen-Da Wei, Tengchao Lv, Li Dong, Lei Cui (possible past Tsinghua University affiliation), Furu Wei
Abstract

Multimodal agentic coding requires robust visual reasoning to analyze observations and evaluate generated artifacts. Current frameworks largely rely on a Thinking-with-Images paradigm, where agents depend on their intrinsic vision to execute visually grounded tasks. However, this architectural coupling introduces a fundamental bottleneck. Agent performance remains strictly bounded by internal perceptual limitations. Furthermore, text-only agents lack native perception. Without specialized tools,...

πŸ“„ Towards Seconds-level Video Representation: Quadratically Gated Feature Grid with Adaptive, Prediction-free Sampling
πŸ—“οΈ Published: 11/13/2026
πŸ”— http://arxiv.org/abs/2611.17786v1
πŸ‘₯ Authors: Tie-Yun Li, Hao-Ran Wang, Chao Wang (possible past Google (United States) affiliation), Zhiyong Yin, Yiming Xu, Zhenwei Cai, Li Li (possible past Google (United States) affiliation), Rongzhi Lin, Feng-Lei Fan
Abstract

Implicit neural video representations are notoriously slow to build, which does not cross the basic practical threshold for most industrial applications. Here, we present Q-GAS, a coordinate-queryable video representation that couples multi-resolution explicit 3D feature grids with a compact SwiGLU-style quadratic-gating (QG) decoder and an error-aware, prediction-free sampler. The explicit grids avoid hash-collision coefficient sharing at the cost of modestly higher storage yet greatly boost th...

πŸ“„ A Spectral Framework for Universal Proxy Regularization in Model Merging
πŸ—“οΈ Published: 11/13/2026
πŸ”— http://arxiv.org/abs/2611.17728v1
πŸ‘₯ Authors: Shangge Liu, Danyue Qian, Qi Fan, Yinghuan Shi, Wenbin Li, Lei Wang (possible past Baidu (China) affiliation), Yang Gao (possible past Tencent (China) affiliation), Dacheng Tao
Abstract

Model merging constructs multi-task models without expensive joint training but often suffers from parameter interference. Some solutions mitigate this by constraining finetuning using data from the other tasks to be merged, violating the core independence and privacy assumption of model merging. Recent work offers preliminary empirical evidence that a universal, task-agnostic proxy distribution can substitute for such data, but leaves open when and why it suffices. To this end, we propose a \te...

πŸ“„ An AI-enabled high-resolution chlorophyll-a forecasting model that outperforms the operational global biogeochemical system
πŸ—“οΈ Published: 11/13/2026
πŸ”— http://arxiv.org/abs/2611.17704v1
πŸ‘₯ Authors: Yong Jiang (possible past Tsinghua University affiliation), Zheng Jiang, Gaowei Zhang, Zengzhou Hao, Suixiang Shi, Chi Zhang (possible past Peking University affiliation), Wei Wang (possible past University Of Oxford affiliation)
Abstract

Surface chlorophyll-a concentration is a key indicator of marine phytoplankton variability, and short-term forecasts of its distribution provide advance information for marine ecosystem monitoring and management. However, interactions between physical transport and biological processes complicate chlorophyll-a forecasting, and simulating these coupled processes with conventional numerical models imposes substantial computational demands. Here we introduce ChlCast, an AI model that generates dail...

πŸ“„ AscendGYM: A Cost-Aware Kernel Optimization Benchmark for Real-World Workloads on Ascend NPUs
πŸ—“οΈ Published: 11/13/2026
πŸ”— http://arxiv.org/abs/2611.17621v1
πŸ‘₯ Authors: Ai Training Platform Team, Bingwei Zhang, Jing Li (possible past Tencent (China) affiliation), Amir Abdullin, Sihang Chen, Xin Chen (possible past Tencent (China) affiliation), Liangjun Feng, Yingmeng Gao, Tao Guo, Junjie Huang, Dongfang Li, Zhanglian Lin, Jiajun Liu, Zekun Liu, Zhengxuan Lu, Xiaodong Luo, Jian Meng, Ivan Nosov, Linyuan Qiu, Weijian Sun, Wanrui Yan, Binfan Zheng, Zhicheng Zhou, Baotian Hu, Min Zhang (possible past Tsinghua University affiliation), Haizhou Li, Zhiquan Luo
Abstract

Kernel optimization is essential for efficient large language model (LLM) training and inference on Ascend NPUs. Existing Ascend kernel benchmarks primarily focus on kernel generation, with less emphasis on the ability of agents to optimize existing implementations under input configurations drawn from real workloads. To address this requirement, we introduce AscendGYM, a benchmark whose initial suite comprises 48 kernels across three difficulty levels. Each kernel is paired with test cases whos...

πŸ“„ Rethinking Visual Reasoning in Text-to-Image Reward Modeling
πŸ—“οΈ Published: 11/12/2026
πŸ”— http://arxiv.org/abs/2611.17585v1
πŸ‘₯ Authors: Shiye Su, Xiaohan Wang (possible past Baidu (China) affiliation), Ludwig Schmidt (possible past University Of Washington affiliation), Serena Yeung-Levy
Abstract

Reward models are a core component of modern text-to-image systems, serving as the objective for inference-time selection and RL-based post-training. A key prerequisite is visual understanding: to judge prompt adherence and ground aesthetic preferences, a reward model must bind attributes, resolve relations, and count reliably. We find that widely-used reward models perform poorly on targeted visual reasoning evaluations. Even when starting from strong vision-language backbones, preference fine-...

πŸ“„ Advancing Quantization for R&R Models with Microscaling Formats
πŸ—“οΈ Published: 11/12/2026
πŸ”— http://arxiv.org/abs/2611.17501v1
πŸ‘₯ Authors: Yavuz Yetim, Bor-Yiing Su (possible past Google (United States) affiliation), Zhe Wang (possible past Deepmind (United Kingdom) affiliation), Arthur Liu, Pranav Sharma, Ying Wang (possible past Tsinghua University affiliation), Randy Shuai, Luke Pfister, Ruichao Xiao, Bill Zhao, Zihao Liu, Liyuan Li, Jiangyuan Li, Gaoxiang Liu, Nathan Yan, Geonhwa Jeong, Jatin Chhugani, Elias Ellison, Uladzimir Pashkevich
Abstract

Recommendation and Ranking (R&R) models, particularly in digital advertising, require massive inference throughput but exhibit extreme sensitivity to precision loss. Unlike Large Language Models (LLMs) that frequently tolerate quantization-induced accuracy drops of 1% or more, production-grade R&R models enforce a stringent launch bar: model quality must remain within a 0.01% Normalized Entropy (NE) margin relative to the baseline, which commonly uses FP16 or FP32. This tight constraint has hist...

πŸ“„ CoReMem: Cost-Internalizing Memory for Successive Coding Tasks with Prospective Valuation
πŸ—“οΈ Published: 11/12/2026
πŸ”— http://arxiv.org/abs/2611.17316v1
πŸ‘₯ Authors: Peiwen Li, Harshit Verma, Sophia Zhou, Haiwen Wang, Ngoc Bui, Heng Chang (possible past Tsinghua University affiliation), Fei Fei, David Van Dijk, Rex Ying (possible past Stanford University affiliation)
Abstract

Coding agents tackle successive tasks on an evolving repository, carrying forward edits and writing memory before the next task is known. Selecting memory for unknown next tasks requires balancing future usefulness against the costs of writing and reading it. The actor must also decide when acting on available memory saves effort as the task unfolds. We introduce Coding-Repository Memory (CoReMem), a cost-internalizing framework that guides memory design and actor adaptation through measured mem...

πŸ“„ Consistent Video-to-Video Translation via Explicit Correspondences
πŸ—“οΈ Published: 11/12/2026
πŸ”— http://arxiv.org/abs/2611.17298v1
πŸ‘₯ Authors: Gaurav Parmar, Zhengqi Li (possible past Google (United States) affiliation), Richard Zhang, Jun-Yan Zhu (possible past University Of California, Berkeley affiliation), Srinivasa Narasimhan, Eli Shechtman, Yotam Nitzan
Abstract

Interactive video-to-video applications require real-time generation while maintaining long-range temporal consistency. However, recent methods achieve speed by restricting attention to a fixed-length causal window. This limits long-range consistency whenever relevant history falls outside the context window. To address this limitation, we introduce vid2vid-long, a correspondence-based approach grounded in a simple observation. Video-to-video translation provides an explicit cross-frame referenc...

πŸ“„ LiveGeo: Benchmarking Geographic Agents in Continuously Evolving Worlds
πŸ—“οΈ Published: 11/12/2026
πŸ”— http://arxiv.org/abs/2611.17267v1
πŸ‘₯ Authors: Pengcheng Zhou (possible past Carnegie Mellon University affiliation), Xuanyu Liu (possible past Baidu (China) affiliation), Yanchen Yin, Tianjie Ju, Bobo Li, Shengqiong Wu, Hao Fei, Mong-Li Lee, Wynne Hsu (possible past National University Of Singapore affiliation)
Abstract

We introduce GeoWorld, a controllable and continuously evolving geographic environment platform for LLM agents, and LiveGeo, a benchmark for evaluating reliable decision-making in real-world geographic environments using GeoWorld. Unlike existing evaluations that largely reason and plan over fixed geographic states, GeoWorld is grounded in real-world geographic structure and allows external world states to evolve along their own temporal processes while preserving verifiable state histories. Liv...

πŸ“„ SALA: Semantically Aligned Latent Actions Enable Dual-Axis Scaling of Embodied Foundation Models
πŸ—“οΈ Published: 11/12/2026
πŸ”— http://arxiv.org/abs/2611.17110v1
πŸ‘₯ Authors: Jiachun Jin, Zetong Zhou, Haonan Wang, Xiao Yang (possible past Tencent (China) affiliation), Jun Zhu (possible past Tsinghua University affiliation), Zhijie Deng
Abstract

Embodied foundation models hinge on diverse robot interaction data, which are costly to collect. Egocentric human videos offer a complementary source of experience, and latent action models (LAMs) provide a popular means of leveraging these videos for policy supervision. Compared with world-action models (WAMs), LAMs can offer greater training efficiency and potentially lower inference costs by avoiding explicit pixel prediction during policy learning and execution. However, most existing LAMs l...

πŸ“„ AURA 2.0: From User-Driven Responses to Model-Initiated Conversations in Streaming Video
πŸ—“οΈ Published: 11/12/2026
πŸ”— http://arxiv.org/abs/2611.17048v1
πŸ‘₯ Authors: Shuhan Li, Huankang Guan, Xintong Guo, Yang Bo, Yuyang He, Dunyuan Xu, Jingyan Chen, Fang Liu (possible past Massachusetts Institute Of Technology affiliation), Tsz Kin Wan, Jinpeng Chen, Zhenqi He, Yuxuan Hu, Wenbo Li, Han Gao (possible past Tencent (China) affiliation), Qijun Li, Long Chen (possible past Tencent (China) affiliation), Hongsheng Li, Rui Liu
Abstract

Most AI assistants respond only when prompted, remaining silent even when events in the video stream offer meaningful opportunities for conversation. To move beyond purely user-driven interaction, we present AURA 2.0, a streaming video large language model that introduces three core upgrades to AURA. The key advance is Model-Initiated Conversation (MIC), which enables the model to identify topics worth discussing in the evolving visual stream and initiate conversations at appropriate moments wit...

πŸ“„ From Delegation to Coordination: Learning Efficient Multi-Agent Deep Research
πŸ—“οΈ Published: 11/12/2026
πŸ”— http://arxiv.org/abs/2611.17041v1
πŸ‘₯ Authors: Yuyang Wu, Yang Liu (possible past Tsinghua University affiliation), Zequn Xie, Junjie Wang, Dan Yang, Zehao Tan, Ding-Chu Zhang, Junming Yang, Yue Shen, Jian Wang (possible past Baidu (China) affiliation), Jinjie Gu
Abstract

Multi-agent deep-research systems use delegation to parallelize search and keep long tool traces out of the main agent's context. Yet delegation helps only when task allocation and subtask execution are coordinated: the main agent must issue useful subtasks, while subagents must return reliable evidence. Existing training either optimizes only the main agent or broadcasts the same team advantage to all agents, which can reward misleading subagent reports and penalize useful ones. We introduce LE...

πŸ“„ The Value of Thinking: Learning When to Reason in Hybrid Decision Models
πŸ—“οΈ Published: 11/12/2026
πŸ”— http://arxiv.org/abs/2611.16961v1
πŸ‘₯ Authors: Zhengxu Yu, Ke Xu, Meng Fang (possible past Tencent (China) affiliation), Haitham Ammar, Weilin Luo, Jun Wang (possible past Tencent (China) affiliation)
Abstract

Existing adaptive-reasoning methods shorten the explicit reasoning a model writes, or skip it by switching modes; either way the answer is still written out token by token. \emph{Rather than deciding how much to think}, we explore a different paradigm, \emph{zero-or-$T$ thinking}: a single model either answers from the implicit thinking inside one forward pass, generating no token, or performs full autoregressive reasoning at a cost of $T$ tokens when that is worth paying. We realise it in a \em...

πŸ“„ Training Reasoning Models in Simulation to Natively Use Real Robots
πŸ—“οΈ Published: 11/13/2026
πŸ”— http://arxiv.org/abs/2611.18562v1
πŸ‘₯ Authors: Jiaming Tang, Zhuoyang Zhang, Junxian Guo, Fangzhou Zhao, Yilong Zhao, Zekai Li, Xingyang Li, Yuyang Zhao, Song Bian, Zhijian Liu (possible past Massachusetts Institute Of Technology affiliation), Song Han (possible past Stanford University affiliation)
Abstract

Robot-use agents (RUAs) have emerged as a promising approach to general-purpose robot control. However, strong performance has so far been achieved only by frontier proprietary models such as GPT-6 Astra and Claude Fable 5.2, whereas open-weight models remain substantially less capable. We show that an open-weight 27B reasoning model can be trained to use real robots natively through reinforcement learning, entirely in simulation and without demonstrations. Like a computer-use agent (CUA), the m...

πŸ“„ StreamEmbed: Real-Time Indexing from Pixels for Screenshot Retrieval
πŸ—“οΈ Published: 11/13/2026
πŸ”— http://arxiv.org/abs/2611.18516v1
πŸ‘₯ Authors: Dayoon Ko, Yichuan Wang, Soyeong Jeong, Joseph E. Gonzalez (possible past University Of California, Berkeley affiliation), Gunhee Kim, Sewon Min (possible past University Of Washington affiliation)
Abstract

Screen history can serve as searchable visual memory for people and agents, but unlike an offline corpus, it must be encoded and indexed as each screenshot arrives. OCR-based indexing is fast but loses visual information; visual encoding takes over two seconds. Yet screens change incrementally: most pixels repeat within and across screenshots. This raises a question: do visual encoders need to recompute their representations from scratch for every new screenshot? We present StreamEmbed, a system...

πŸ“„ Delayed Effects of Temporary Weight Decay in Muon Pretraining
πŸ—“οΈ Published: 11/13/2026
πŸ”— http://arxiv.org/abs/2611.17876v1
πŸ‘₯ Authors: Md Rifat Arefin, Ravid Shwartz-Ziv, Ernie Chang, Chinnadhurai Sankar (possible past Google (United States) affiliation), Rylan Conway, Aristide Baratin, Irina Rish (possible past Deepmind (United Kingdom) affiliation), Adithya Sagar, Patrick Huber
Abstract

A temporary optimization intervention can increase validation loss when it ends yet reduce it at the final training budget. We demonstrate this with temporary weight decay in Muon-trained decoder-only Transformers with 162M, 405M, 1.4B, and 3.4B parameters. For models up to 1.4B parameters, moving a fixed-duration decay window from early to late training reverses its final effect. At 162M, changing window duration alters final loss even at approximately matched total shrinkage. Most strikingly, ...

πŸ“„ Disentangling Forgetting from Utility: Lightweight LLM Unlearning via Orthogonal Null-Space Projection
πŸ—“οΈ Published: 11/13/2026
πŸ”— http://arxiv.org/abs/2611.17864v1
πŸ‘₯ Authors: Xiuyuan Wang, Weiming Liu, Zongxin Yang (possible past Baidu (China) affiliation), Fan Wang (possible past Baidu (China) affiliation), Liemiao Jin, Rubin Qin, Wei Chen, Bin Shi, Chaochao Chen, Xiaolin Zheng
Abstract

Large Language Models (LLMs) pose significant security risks by retaining private or hazardous information, creating an urgent need for unlearning techniques that excise specific knowledge without compromising general utility. However, prevailing fine-tuning-based unlearning methods struggle to achieve this balance, often suffering from computational inefficiency and unintended utility degradation due to the entangled nature of knowledge within model parameters. To disentangle forgetting from ut...

πŸ“„ Learning Anisotropic Prediction Targets for Flow Models
πŸ—“οΈ Published: 11/13/2026
πŸ”— http://arxiv.org/abs/2611.17767v1
πŸ‘₯ Authors: Rayhan Zirvi, Hansheng Chen, Gordon Wetzstein (possible past Stanford University affiliation), Leonidas Guibas (possible past Stanford University affiliation)
Abstract

Flow-based generative models are trained on a hand-chosen prediction target, such as the velocity, the noise, or the clean data. This choice strongly affects training and sample quality. We introduce Anisotropic Flow Models (AnisoFlow), which replaces a fixed prediction target with a fully learnable one. AnisoFlow learns to scale the noise component of the target anisotropically, assigning a separate nonnegative weight to each direction in a jointly learned basis. A closed-form map recovers the ...

πŸ“„ Improved Regret via Variance-Aware Policy Optimization in Adversarial MDPs
πŸ—“οΈ Published: 11/12/2026
πŸ”— http://arxiv.org/abs/2611.17469v1
πŸ‘₯ Authors: Ofir Schlisselberg, Alon Cohen (possible past Google (United States) affiliation), Yishay Mansour (possible past Google (United States) affiliation)
Abstract

We study episodic adversarial MDPs with unknown dynamics and full-information rewards and achieve regret with leading term $\widetilde O(\sqrt{H^3SAK})$. Here $H$ is the horizon, $S$ the total number of states across decision layers, $A$ the number of actions, and $K$ the number of episodes. This improves the two incomparable prior leading terms by factors $\sqrt{S/H}$ and $\sqrt{H^3}$, leaving a factor $\sqrt H$ to the lower-bound scale. We give a reduction from unknown to known dynamics for \e...

πŸ“„ Unbundling Multimodal Vision Encoding
πŸ—“οΈ Published: 11/12/2026
πŸ”— http://arxiv.org/abs/2611.17334v1
πŸ‘₯ Authors: Ellis Brown, Chris Hoang, Shengyi Qian, Shang-Wen Li (possible past Amazon (United States) affiliation), Mike Lewis (possible past Meta (United States) affiliation), Marjan Ghazvininejad (possible past Meta (United States) affiliation), Luke Zettlemoyer (possible past University Of Washington affiliation), Lucas Beyer (possible past Google (United States) affiliation), Mengye Ren (possible past University Of Toronto affiliation), Jakob Verbeek, Saining Xie (possible past Meta (United States) affiliation), Hu Xu
Abstract

How multimodal language models should encode visual input remains up for debate. Most models use a pretrained vision encoder, while some recent recipes train the encoder from scratch with the language model or remove it entirely. Vision encoders bundle together three design choices: a visual prior from upstream pretraining, dedicated vision capacity, and pre-fusion placement of that capacity (i.e., before the multimodal backbone). Because published models also differ in data, architecture, and c...

πŸ“„ ArchSpace: Recording, Reviewing, and Referencing LLM Architecture Research
πŸ—“οΈ Published: 11/12/2026
πŸ”— http://arxiv.org/abs/2611.17332v2
πŸ‘₯ Authors: Jie Wang (possible past Tsinghua University affiliation), Tong Zhu (possible past Nvidia (United States) affiliation), Jiahao Kuang, Zhe Xu, Mowen Ruan, Yunkai Chai, Yuliang Liu, Dehong Zheng, Rubin Wei, Yu Cheng (possible past National University Of Singapore affiliation), Bowen Zhou, Tao Ji, Zhouhan Lin, Qi Zhang (possible past Tencent (China) affiliation)
Abstract

Model architecture shapes the scaling behavior, training efficiency, and inference cost of LLMs, making it a central focus of both industry and academia. In industry, teams such as those behind DeepSeek, Qwen, and Kimi routinely introduce architectural innovations, yet public reports rarely document which ideas succeeded, which were abandoned, and why. In academia, limited compute and inconsistent baselines often restrict evaluations to small-scale, single-point comparisons, making it difficult ...

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