OrchSLM is a routing framework that unifies non‑interactive orchestration methods for small language models (SLMs). It allows heterogeneous SLMs to independently generate candidate solutions while a router manages their cached outputs without further model interaction. By systematically probing OrchSLM, the study shows how orchestration behavior depends on task structure, model‑pool composition, and multi‑agent consensus.
By Chengxi Zhang, Yu Yao
The paper introduces PriCoder, a method for teaching large language models (LLMs) to effectively use private library APIs for code generation. PriCoder synthesizes training data by constructing a graph and applying two operators—Progressive Graph Evolution to increase diversity and Multidimensional Graph Pruning to enhance quality. Experiments on three mainstream LLMs demonstrate that PriCoder boosts private‑library code generation by over 20% in pass@1, while leaving general code generation largely unchanged.
By Yitong Zhang, Chengze Li, Ruize Chen, Guowei Yang, Xiaoran Jia, Yijie Ren, Jia Li
The paper introduces a communication‑efficient method for adapting large language models on decentralized GPU meshes. It proposes an asynchronous two‑circuit system that uses fast compressed training with activation masking for pipeline‑parallel transfer and compressed data‑parallel synchronization, while a slower anchor circuit performs occasional unmasked passes. A spectral correction optimizer then denoises the masked gradients using these anchor priors, enabling high compression rates and achieving up to 40× throughput gains over internet‑grade connections while matching dense uncompressed performance.
By Sameera Ramasinghe, Shamane Siriwardhana, Thalaiyasingam Ajanthan, Hadi Mohaghegh Dolatabadi, Chamin P Hewa Koneputugodage, Gil Avraham, Violetta Shevchenko, James Snewin, Karol Pajak, Harry Xi, Alexander Long
MoVT is a new framework for text‑to‑motion generation that uses a cross‑modal augmented motion tokenizer to project 3D motion tokens into 2D, enriching the motion codebook with real‑world video patterns. The enriched tokens are mapped back to 3D, creating aligned 3D and 2D codebooks that better capture intricate motions. These codebooks feed a generative masked transformer, which predicts masked motion tokens in a modality‑agnostic way, allowing text‑index pairs from the 2D codebook and annotated videos to further improve generation quality. Empirical tests show MoVT outperforms previous state‑of‑the‑art methods on several key metrics.
By Beibei Jing, Tianle Guo, Youjia Zhang, Zikai Song, Yawei Luo, Junqing Yu, Tao Guan, Wei Yang
The paper introduces G-ray, a ray-level relative position encoding for multi-view vision Transformers that remains consistent across different camera projections. By parameterizing rotary phases with camera-local ray angles, G-ray achieves projection-invariant positional consistency and can be integrated with existing encodings without extra learned parameters. Experiments on 3D reconstruction and novel-view synthesis benchmarks show that G-ray improves performance, notably reducing mean pointmap relative error by 45.8% over MapAnything.
By Shuo Zhang, Xin Su, Wei Wang, Jun Liu, Xinrui Zeng, Yongsen Chen, Chenjie Wang, Guibo Zhu, Jinqiao Wang, Bin Luo, Liangpei Zhang
The paper introduces Atria Dawn Preview, a foundation agentic language model aimed at scientific research and engineering workflows. Trained through a Verifiable Experience Pipeline, it performs competitively across 16 real‑world benchmarks, achieving the highest scores on five. The authors also present a detailed case study of human–AI collaboration, showing that while AI proposes methods and revisions, humans retain final decision‑making and guide the research direction.
By Honglin Guo, Tao Gui, Yicheng Chen, Guanting Dong, Qiming Ge, Yuyang Hu, Zixian Huang, Jiajie Jin, Alexander Lam, Yining Li, Jiahang Lin, Yanjiang Liu, Xinyu Lu, Haijun Lv, Junlin Shang, Qisheng Su, Guoqiang Wang, Rui Wang, Zhecan Wang, Hao Xiang, Xinchen Xie, Shuhao Xing, Xiaoyu Xing, Wanghan Xu, Xinyu Yang, Yajie Yang, Chengfeng Zhao, Haoran Zhao, Ruojun Zhou, Yunhua Zhou, Yicheng Zou, Kun Cai, Qiye Cai, Xinmeng Che, Haodong Chen, Jiabei Chen, Jiahao Chen, Jiayi Chen, Yujia Chen, Lizhi Cui, Youheng Dai, Xin Deng, Yi Dong, Shihan Dou, Chenya Gu, Xu Guo, Ding Han, Feiyang Hao, Haotan He, Jie Hou, Binze Hu, Zijian Hu, Junhao Huang, Huicheng Jiang, Jiazhen Jiang, Shufan Jiang, Jiahao Kuang, Bowen Lai, Bo Li, Jiaqiang Li, Peng Li, Qilong Li, Zhuoqun Li, Jiaxiang Liu, Shuainan Liu, Tong Liu, Yi Liu, Zhonghang Lu, Jianwen Luo, Yanyi Luo, Huijie Lv, Ningsheng Ma, Zerun Ma, Houcheng Min, Chengjun Pan, Qiyuan Peng, Xiaoxuan Peng, Jianmin Qian, Jiantao Qiu, Wanying Ren, Huayu Sha, Jifei Shan, Zixin Shang, Bing Shao, Zhuohui Sheng, Jiayang Shi, Yang Shu, Aierpanjiang Simayi, Sirui Song, Yuxiao Song, Zhe Sun, Zhichao Sun, Wenzhe Tan, Wenhui Tian, Zhongbo Tian, Hanchen Wang, Pengbo Wang, Rui Wang, Yiding Wang, Yuhui Wang, Zhiheng Xi, Caijun Xu, Chao Xu, Yongfeng Xu, Xiaolei Yang, Zhixiong Yang, Qian Yao, Shihong Yi, Yuankai Ying, Jia Yu, Dingbo Yuan, Hao Yuan, Junjie Yuan, Bo Zhang, Caixian Zhang, Qiuyinzhe Zhang, Jiyuan Zhao, Penghao Zhao, Ying Zhao, Pujun Zheng, Xiaoxue Zhong, Xiaohao Zhou, Xinyu Zhou, Dongsheng Zhu, Guanru Zhu, Yulun Zhu, Yaojie Lu, Tao Ji, Hongyu Lin, Yutao Zhu, Pengfei Cao, Guoxiu He, Xianpei Han, Ben He, Zhicheng Dou, Kang Liu, Qi Zhang, Le Sun, Jun Zhao, Ji-Rong Wen, Xuanjing Huang, Yu-Gang Jiang, Bowen Zhou
The paper presents a transfer‑learning approach that adapts a multimodal spatiotemporal vision transformer, originally trained on Demographic and Health Survey data, to estimate socioeconomic conditions in forced‑displacement settings. Using satellite‑derived geospatial covariates, the adapted model explains up to 66% of variation in socioeconomic outcomes in camp‑intersecting grids and 41% in non‑camp areas, achieving mean absolute errors of 4.37 and 5.41 index points respectively. This framework supplements periodic household surveys by providing regularly updated, spatially granular socioeconomic estimates that bridge data gaps between survey rounds.
By Steven Ndung'u, Adel Daoud, Ismael Yacoubou Djima, Hai-Anh H. Dang, Patrick Michael Brock
arXiv:2609.13202v1 Announce Type: new
Abstract: Feature engineering has long been a cornerstone of tabular machine learning. Tabular foundation models (TFMs) are pretrained on a wide range of tabular...
By Yifan WU, Pinjun Dong, Jiran Tao, Binyan Jiang
arXiv:2609.14092v1 Announce Type: cross
Abstract: Sparse DNN accelerators specialize in ANN or SNN execution, leaving energy or latency on the table when workload characteristics vary within a layer....
By Varun Manjunath, Pranav Ramesh, Gopalakrishnan Srinivasan
arXiv:2609. 15498v1 Announce Type: new Abstract: Recurrent Transformers reusing their weights rather than stacking $L$ distinct layers are becoming widely adopted due to their parameter efficiency [1,2,3].
By Pawel Olszowiec, Michal Byra, Grzegorz Gruszczynski, Grzegorz Stefanski, Alberto Presta
arXiv:2609.13445v1 Announce Type: new
Abstract: Speech-to-speech LLMs like Moshi, and its derivative PersonaPlex, can listen and speak concurrently through full-duplex generation. However, they can b...
By Kento Nishi
EvoOntology introduces a self‑evolving ontology layer for data agents, encapsulating the ontology as an MCP server with schema, content, and tool layers. It enables agents to query and interact with the ontology at runtime, using a builder agent for autonomous construction and a self‑evolution loop that refines the ontology through attribution‑guided edits validated by backbone‑conditional evaluation. Experiments on three data‑agent benchmarks with four LLM backbones show that EvoOntology consistently outperforms strong baselines and existing semantic‑layer approaches, effectively bridging the agent‑data gap for heterogeneous data.
By Meiduo Chong, Shaolei Zhang, Ju Fan, Xiaoyong Du
The paper critiques the standard fixed-rule approach for deriving labels from human feedback in human-robot collaboration, showing that human-provided implication labels often differ and improve reward learning. It introduces IMPLIED, a method that starts with fixed-rule implications but learns to infer and revise accepted/rejected action labels over time, outperforming both the fixed rule and LLM baselines on recorded trajectories and a physical pizza‑making study. As a result, IMPLIED reduces preference‑estimation error and yields robot actions that better align with combined reward objectives.
By Qiping Zhang, Kate Candon, Debasmita Ghose, Marynel V\'azquez
The paper investigates how continuous latent states in large language models can store multiple reasoning steps through superposition. It challenges the intuition that retaining only the current reasoning frontier is optimal, showing that cumulative superposition of the full reasoning history can actually require fewer representational dimensions. The authors demonstrate that this approach preserves more valid evidence, improves downstream outcome discrimination, and delays unreliability, while also establishing that uniform cumulative weighting of memories is minimax‑optimal for future reasoning.
By Hongyu Gu, Chang Liu, Jingwen Fu
VisInteract introduces a new paradigm for Text-to-Visualization that treats imperfect user queries as a dynamic, interaction-driven problem, requiring the system to recover true intent through multi-turn dialogue. The authors present VisInteract-Bench, the first benchmark for interactive Text-to-Vis, featuring controlled imperfection injection, a realistic user agent, and dual-perspective automated evaluation. They also propose Vis-MCTS, an enhanced Monte Carlo Tree Search algorithm that incorporates progressive widening, cross-rollout information sharing, and dimension-aware reward decomposition, achieving significant performance gains over existing baselines.
By Wenxin Xu, Jinwei Lu, Hwanhee Kim, Chen Jason Zhang, Xiao-Yong Wei, Haoyang Li, Yuanfeng Song
Stellar Colosseum is a model‑agnostic harness designed to improve long‑horizon research in mathematics and theoretical computer science by allocating inference across multiple agents. It explores alternative strategies before constructing proofs, uses a readiness gate to decide when a route is mature enough to decompose, represents proof plans as interdependent subproblems, and routes verifier findings back to the relevant part of the argument. The workflow generates candidates in parallel, attacks them with targeted falsification, and combines candidates and critiques into a single research artifact through overlapping random‑sample tree aggregation, and has been integrated into Google Antigravity's Teamwork framework as the Long Proof pattern. Demonstrations show that, when paired with Gemini 3.1 Pro, Stellar Colosseum achieves 71.0% accuracy on the TCS‑Bench theorem‑proving benchmark and solves 218 of 222 Codeforces problems.
By Honghao Lin, David P. Woodruff, Yuan Deng, Jieming Mao, Song Zuo, Vahab Mirrokni
The paper proposes a modular tokenizer framework for multilingual large language models, allowing the creation of language‑specific subtokenizers that match monolingual compression quality. It introduces a pretraining strategy that samples these subtokenizers to limit predictions to relevant vocabularies, enabling efficient training and inference. This approach reduces memory usage and speeds up inference without compromising performance.
By Franck Signe, Hippolyte Pilchen, Fran\c{c}ois Yvon, \'Edouard Grave
The study examines how duplicate content is used to detect coordination in social media information operations. It distinguishes between generic, low‑information duplicates and non‑generic, more specific duplicates, labeling 187,000 tweets with an LLM‑assisted protocol and supervised classifiers. Results show that generic duplicates are rare with lexical matching but comprise nearly 39% of campaigns identified by embedding methods, and filtering out generic duplicates yields smaller, denser coordination graphs, indicating a more focused structure.
By Ashfaq Ali Shafin, Khandaker Mamun Ahmed
The paper introduces self‑orchestrating language models that annotate semantic dependence—identifying which tokens rely on others—to guide efficient inference. By leveraging these annotations, the authors design runtimes that parallelize autoregressive decoding, evict intermediate context, or determine denoising orders, achieving Pareto‑optimal quality‑efficiency trade‑offs. Three systems—PASTA, TIP, and Planned Diffusion—demonstrate these techniques for parallel decoding, memory‑efficient reasoning, and efficient discrete diffusion, respectively.
By Tian Jin
EEG-Xplain introduces a unified attribution framework to interpret EEG foundation models such as BIOT, LaBraM, and EEGMamba. The framework combines gradient, perturbation, and activation-based methods to analyze model behavior across spatial, temporal, and frequency dimensions, identifying critical channels, decision-relevant signal segments, and contributions of canonical EEG rhythms. It evaluates explanation reliability with population-level metrics and uses large language models to convert structured attributions into natural-language reports, demonstrating consistency with known neurophysiological markers on benchmark datasets.
By Hansong Ma, Junxiao Wang