The paper proposes an Infinite-Parameter LLM architecture that generates and adapts its weights from live interaction data using a compact hypernetwork and Bayesian updating, allowing the model to learn from real-time user input rather than relying solely on static pretraining. This approach keeps the stored footprint fixed while effectively enabling an infinite set of weights, potentially improving compute efficiency, freeing context windows, and providing persistent, generalizable knowledge across turns. The authors also outline an evaluation protocol to compare this method against traditional in-context learning and retrieval techniques.
By Jinli Hu, Ross M. Clarke, Yichuan Zhang, Jos\'e Miguel Hern\'andez-Lobato
The paper investigates why large vision‑language models sometimes misclassify harmful memes, attributing failures to either missing internal evidence or poor routing of evidence to the output. Using sparse autoencoders, role‑conditioned probes, and causal interventions on Gemma‑3 and Qwen3.5, the authors show that sparse readouts consistently outperform native predictions across six harmful content benchmarks, revealing a readout gap that is largely due to routing rather than representation. The study also demonstrates that calibration‑only routing recovers most of the performance gap and that the issue persists across languages and is not solely driven by OCR signals.
By Girish A. Koushik, Diptesh Kanojia, Helen Treharne
arXiv:2609.18304v1 Announce Type: new
Abstract: Large language model (LLM) agents increasingly tackle long-horizon tasks through multi-step environment interaction, yet a single erroneous action can...
By Yi Yu, Liuyi Yao, Yaliang Li, Enshu Wang, Libing Wu
arXiv:2604.14121v3 Announce Type: replace
Abstract: Large language models (LLMs) have become increasingly used for various tasks, often coupled with Chain-of-Thought (CoT) prompting to boost accuracy...
By Zipeng Ling, Shuliang Liu, Seonil Son, Shenghong Fu, Yuehao Tang, Yao Wan, Xuming Hu
TabICLv2 is a new state‑of‑the‑art tabular foundation model that outperforms existing methods on regression and classification tasks. It relies on a synthetic data generation engine for diverse pretraining, architectural innovations such as a scalable softmax attention, and optimized training protocols that replace AdamW with the Muon optimizer. On the TabArena and TALENT benchmarks, TabICLv2 surpasses the current best model, RealTabPFN‑2.5, without any tuning, while also being faster and capable of handling million‑scale datasets with limited GPU memory.
By Jingang Qu, David Holzm\"uller, Ga\"el Varoquaux, Marine Le Morvan
The paper addresses the memory bottleneck in reinforcement learning for large language models caused by the large Key-Value (KV) cache during rollout phases. It highlights that while KV cache compression can reduce memory usage, it introduces a significant off‑policy bias that standard statistical corrections cannot adequately mitigate. The authors argue that even tiny compression errors are amplified by RL’s instability, leading to inefficient learning.
By Rui Zhu, Weiheng Bai, Qiushi Wu, Yang Ren, Haixu Tang, Yuchu Liu
The paper introduces the Tri‑Metric Router, a deterministic, training‑free policy that chooses among Raw, Neural, and Lexical pipelines for retrieval‑augmented generation on commodity GPUs. It uses three CPU‑side signals—spatial complexity, syntactic density, and type‑token ratio—to balance VRAM headroom and latency, calibrated on LongBench qasper. The method eliminates out‑of‑memory failures and improves alignment and F1 scores compared to always‑on lexical compression without extra VRAM or training costs.
By Saipraveen Vabbilisetty, Ajay Kumar Boddepalli, Deep Narayan Mishra, Shashank Kapadia, Haoan Wang, Anupriya Sharma
FoundAna is a GNN‑assisted foundation model designed for graph anomaly detection across diverse datasets. It combines a GNN component with a transformer encoder enhanced by four positional encodings to capture both local and global structure, using reconstruction errors as anomaly scores. Experiments on nine benchmark datasets from financial, social, and citation networks show that FoundAna consistently outperforms state‑of‑the‑art baselines.
By Suprim Nakarmi, Chahana Dahal, Yue Zhao, Junggab Son, Zuobin Xiong
The paper introduces Temperon, a training strategy that uses plain SGD for the first 43% of the epoch budget and then hands off to a SAM‑wrapped Muon refiner for the remaining training. On datasets such as CIFAR‑10/100, SVHN, and Tiny ImageNet, Temperon achieves the same or better accuracy as full‑time SAM while reaching key performance targets faster and at lower cost. Ablation studies show that the Muon refiner contributes the majority of the performance gain, while the initial SGD explorer and its restarts add negligible benefit.
By Stamatis Mastromichalakis
ASPIRE introduces a non‑synchronized batched self‑speculative decoding framework for long‑context LLM inference, addressing the memory bottleneck of attention by drafting tokens with sparse attention and verifying them with full attention. It combines a unified mixed forward pass, a lightweight online speculation scheduler that lets each request independently decide when to verify, and an intra‑draft refresh layer that updates the sparse context at every draft step. Experiments on three models and five benchmarks show 1.70–4.58× speedup over autoregressive baselines and a 27% average improvement over the strongest prior self‑speculative methods.
By Amir Ziashahabi, Hossein Entezari Zarch, Lei Gao, Murali Annavaram, Salman Avestimehr
arXiv:2609.17981v1 Announce Type: cross
Abstract: Speech Large Language Models (Speech-LLMs), typically built from a pre-trained speech encoder, a modality projector, and an LLM fine-tuned with Low-R...
By Mohan Shi, Zilai Wang, Natarajan Balaji Shankar, Kaiyuan Zhang, Eray Eren, Abeer Alwan
arXiv:2609.19142v1 Announce Type: new
Abstract: World models endow perceptual systems with the ability to predict how scenes evolve under interaction. They are most beneficial when trained on diverse...
By Bardienus P. Duisterhof, Kaifeng Zhang, Adam Hung, Bowen Wen, Stan Birchfield, Yunzhu Li, Deva Ramanan, Jeffrey Ichnowski
The paper studies Graph-Guided Token Merging (G2TM), a module that reduces token count in Vision Transformers. It evaluates G2TM across multiple segmentation frameworks and decoder types, finding that its performance gains are tied to the encoder rather than the decoder. The authors report consistent reductions in GFLOPs (22‑47%) and throughput improvements (up to 74%) on ADE20K, with optimal hyperparameters depending mainly on backbone pre‑training and target dataset.
By Victor Bercy, Martyna Poreba, Michal Szczepanski, Samia Bouchafa
This paper proposes a Retrieval-Augmented Generation (RAG) framework for scientific image quality assessment, designed to simultaneously address both the understanding track (SIQA-U) and the scoring t...
Complexity measured from generated code is failure-dependent: a difficult prompt can yield a short failing program and be assigned low output complexity. We introduce a six-dimension prompt-side struc...
This empirical study is an independent reproduction of the dissociation Zhao reported in 2026. The shape of a large language model's chain-of-thought entropy trajectory predicts whether the final answ...
Claude Cowork and the standard chat interface are merging into a single Claude experience, allowing users to hand over tasks or ask quick questions that the system will continue to handle even after the laptop is closed. The rollout will begin with Pro and Max plan users across web, desktop, and mobile, and will extend to new users on these plans over the coming weeks. This integration suggests Claude is evolving into a general agent, simplifying the distinction between Cowork and regular Claude usage.
Intelligent systems that act in the world require image understanding that is both comprehensive and spatially grounded. Current vision-language models (VLMs) can generate fluent and detailed image ca...
World models endow perceptual systems with the ability to predict how scenes evolve under interaction. They are most beneficial when trained on diverse volumes of data, to instill a rich prior into do...
Emergent coordinated behaviors of AI agents are starting to present critical safety risks. A key phenomenon driving these behaviors is the rapid formation and spread of beliefs about the world, and me...