The paper evaluates Vision Language Models (VLMs) on Visual Question Answering tasks where questions violate Grice's maxims. By generating question modifiers that add non-essential, ambiguous, or false information, the authors show that VLMs such as ChatGPT, Claude, Gemini, and Llava exhibit reduced performance. They also compare human pragmatic reasoning to VLM reasoning, noting differences in how each handles human‑induced versus AI‑generated violations, and find that humans spend less time resolving VLM‑induced violations while VLMs are less accurate in those cases.
By Monika Shah, Sudarshan Balaji, Somdeb Sarkhel, Sanorita Dey, Deepak Venugopal
The paper investigates how easily large language models (LLMs) can retrieve knowledge for a given query, introducing the concept of knowledge accessibility. It discovers that the accessibility of a query is reflected in its geometric position in the model’s representation space: queries closer to a central point are more accessible, while those farther away are less so. This geometric insight identifies a knowledge boundary, shows that accessibility ordering is consistent across datasets, and informs which interventions—such as query rewriting, chain-of-thought reasoning, or retrieval—are most effective depending on a query’s position relative to the center.
By Lihu Chen
OmniConfess is a training‑free method designed to reduce hallucinations in omni‑modal large language models (OmniLLMs) that handle text, images, audio, and video. The approach fixes a candidate response and re‑scores it at token resolution while selectively intervening on evidence from each modality, producing a token‑by‑channel confession that shows which evidence supports each part of the response. Using this confession, OmniConfess preserves grounded content and corrects commitments that rely on irrelevant or contradictory evidence. The authors evaluated the method on OmniHalluBench, a 3,540‑example benchmark drawn from six datasets across multiple modalities and tasks, and found that OmniConfess mitigates hallucinations across diverse settings.
By Huiqiang Rong, Haoran Luo, Hui Feng, Zhonghong Ou, Kaiwen Xue, Guoxin Zhang, Yifan Zhu
The paper introduces a stability theory for Joint-Embedding Predictive Architectures (JEPAs), showing that training dynamics involve a driving force and a decay effect that determine representation collapse. By linearising the gradient flow, the authors derive a per‑mode stability ratio that separates data‑side and predictor‑side contributions, predicting a phase boundary confirmed across 800 configurations. Using this insight, they propose ResidualPred, a transformer predictor that biases attention toward the identity at initialization, improving representation rank and downstream accuracy on tabular and image benchmarks.
By Jos\'e Lucas De Melo Costa, Seong Woo Ahn, Fabrice Popineau, Arpad Rimmel, Bich-Li\^en Doan
The paper introduces a generative‑informed neuro‑symbolic framework that combines generative syntactic theory with AraBERT to resolve structural ambiguity in Modern Standard Arabic noun phrases. By treating ambiguity as a candidate‑based decision task, the model explicitly constructs and evaluates linguistically motivated alternatives, achieving high accuracy (96.88%) and strong F1 scores on an unseen evaluation set. Analysis shows uneven performance across attachment types, with near‑perfect recall for high/VP attachment but lower recall for low/NP/embedded attachment, highlighting challenges in recovering embedded interpretations.
By Mohammed Damom, Muneef Y. Alshawsh, Ashraf A. Naji, Mustafa Ali Alhamzi, Fawwaz An-Nashef, Jameel Ahmed Elayah, Mohammed Q. Shormani, Noman AL-Sayadi
The paper investigates how to reduce the per-user adaptation burden in personalized large language models by separating reusable personalization capacity from user-specific adjustments. Through empirical studies, it shows that shared low‑rank factors can capture much of the cross‑user structure, while a tiny user code suffices for individual correction. The proposed LINEUP framework achieves state‑of‑the‑art performance on six personalized tasks while using only eight scalars per user compared to millions of private‑LoRA parameters.
By Songyuan Sui, Srikanth Malla, Chiho Choi, Joon Hee Choi
The paper introduces Spatial Memory Intelligence (SMI), a framework that enhances long‑video world models by systematically managing spatial memory using an understanding model. SMI employs four coordinated operations—spatial clustering, within‑cluster sparsification, action‑aware retrieval, and reliability‑aware filtering—to handle increasingly complex and lengthy memory sequences. Experiments across various baselines and benchmarks show that SMI improves memory sparsity, generation stability, and spatial consistency, demonstrating its effectiveness and generalizability.
By Ying Yang, Guiyu Zhang, Lianghua Huang, Chang Nie, Chenyang Si, Haofan Wang, Shaoshuai Shi, Li Jiang
The paper investigates how much information can be extracted from different internal representations of vision‑language models, focusing on two bottlenecks: low‑dimensional projections of the residual stream (via tuned lenses) and the final top‑k logits. It systematically compares the amount of retained information at these levels and finds that even the easily accessible top‑logit bottleneck can leak task‑irrelevant details from image queries, sometimes matching the leakage seen from full residual projections.
By Masha Fedzechkina, Eleonora Gualdoni, Rita Ramos, Sinead Williamson
The paper introduces GraphMemory, a lightweight graph-based memory system designed to improve token efficiency in test-time continual learning for large language models. By retrieving only relevant subgraphs for each query, GraphMemory keeps the amount of retrieved memory constant as more examples are processed, avoiding the token cost and performance degradation of traditional shared-context approaches. Experiments demonstrate that GraphMemory achieves competitive downstream performance while using roughly 81‑85% fewer memory‑construction tokens than baseline methods.
By Yehya Farhat, Michael Desmond, Anastasios Kyrillidis
PAPER2LLM++ is a framework that enables large language models to continually evolve by learning from new research papers. It treats each paper as a source of evidence and supervision, extracting findings, testing them against the current model, and converting them into learning signals when necessary. A try‑evaluate‑commit loop ensures that updates improve targeted behaviors without forgetting prior gains or harming general capabilities, allowing models to progressively incorporate new findings while retaining earlier improvements.
By Hongji Pu, Yilun Zhao, Wenpeng Yin
arXiv:2610.02800v1 Announce Type: new
Abstract: Speculative decoding accelerates autoregressive generation by using a lightweight draft to propose multiple tokens for parallel verification. However,...
By Chence Yang, Ningxi Cheng, Arash Akbari, Qitao Tan, Qingchan Zhu, Ci Zhang, Changdi Yang, Yanzhi Wang, Wei Niu, Jinhui Wang, Jin Lu, Geng Yuan
arXiv:2610.03128v1 Announce Type: new
Abstract: Quantitative trading strategy design aims to discover trading programs from historical data that remain effective in future markets, which can be viewe...
By Chaoqun Yang, Qian Wang, Fengbin Zhu, Xinyu Lin, Bingsheng He, Roger Zimmermann, Tat-Seng Chua
The paper investigates how modern selective state space models (SSMs), like Mamba2, behave in distributed learning settings. It derives architecture‑aware gradient and smoothness bounds for single‑ and multi‑layer selective SSMs, and provides convergence bounds for FedAvg and FedProx that incorporate recurrent stability, input‑dependent discretization, and state‑projection norms. Numerical experiments validate the theoretical bounds and compare nine federated learning algorithms on Mamba2 language modeling across six text domains, demonstrating that SSM‑specific bounds help interpret practical federated learning behavior.
By Adam Piaseczny, Md Kamran Chowdhury Shisher, Shiqiang Wang, Christopher G. Brinton
The paper introduces PROVE-REC, a two‑pass framework that generates verifiable preference proofs for large language model (LLM) recommendation systems. In Pass A, the model condenses user interaction histories into a compact proof of positive and avoidance claims linked to specific evidence entries. Pass B then uses only this proof and its evidence to predict the next item, ensuring the recommendation follows the reasoning path. Verification steps compare masked evidence and removed claims to confirm grounding and influence, while a ranking‑preservation objective retains useful historical information. Experiments on diverse real‑world datasets show PROVE‑REC outperforms strong baselines by up to 7.45%, producing claims that are both better grounded and more influential to recommendation quality.
By Yu Hou, Nathaniel Kang, Pengkai Wang, Hua Li
The paper introduces an AI-driven framework for thermal‑aware capacity planning in data centers, enabling temperature predictions in milliseconds by learning from key parameters such as rack power, server placement, and HVAC settings. The model is validated against high‑fidelity CFD simulations and achieves a 10,000‑fold speedup while maintaining high accuracy for unseen designs. This rapid prediction capability allows designers and operators to instantly optimize workload distribution and cooling efficiency.
By Yixing Li, Mark Fenton, Matthew Kaufeler, Ka Ming Leung, Xin Ai, Zhiyu Zeng
RASPER is a reward‑aligned summarizer that tailors the extraction of information from unstructured discharge notes to improve downstream clinical predictions. It uses a tunable LLM summarizer trained with reinforcement learning, where the reward comes from the loss of a downstream predictor, and incorporates patient‑specific context via a longitudinal encoder that soft‑prompts the summarizer with structured codes. The approach consistently outperforms strong baselines on readmission prediction and medication recommendation tasks in the MIMIC‑III and MIMIC‑IV datasets.
By Arya Hadizadeh Moghaddam, Mohsen Nayebi Kerdabadi, Chen Chen, Dongjie Wang, Zijun Yao
DyadMem introduces a new benchmark for evaluating how long‑term agents remember user‑specific relational information, called User‑conditioned Relational Agent Memory (URAM). The dataset contains 3,065 episodes, 50,961 sessions, and 61,210 QA pairs, with detailed annotations for memory capture, update, and recall across multi‑session trajectories. Experiments on 20 models show strong performance in a gold‑memory setting but a significant drop in full‑pipeline QA, highlighting gaps in current LLM memory capabilities.
By Yifei Tao, Xinyu Zhong, Henry Hengyuan Zhao, Fanyi Wang, Tengda Guo, Wentao Qiu, Ying Wang, Liujian Tang
The paper introduces ComInsight, a method for generating analytical reports from relational tables by composing verifiable atomic insights. It defines atomic insights as minimal executable analytical units, enumerates them from the database schema, and organizes them into a multi-relational insight graph. Composition operators then fuse these atomic nodes into higher-order conclusions, each accompanied by executable SQL and fine-grained provenance, achieving superior factual correctness, novelty, and structural completeness on three benchmarks.
By Teng Lin, Xinyu Liu, Nan Tang
The paper examines how different author representations affect zero‑shot authorship attribution (AA). It compares a label‑only prompting baseline with three strategies: representative writing samples, LLM‑generated descriptions, and style embeddings (LISA). The study finds that incorporating author‑specific representations improves performance, with a two‑stage LISA framework achieving the best results, while LLM‑generated descriptions offer a more compact but slightly less accurate alternative.
By Nudrat Habib, Tosin Adewumi, Sana Sabah Al-Azzawi, Marcus Liwicki, Elisa Barney
The Writerslogic team participated in the CLEF 2026 SimpleText shared task, tackling both text simplification (Task 1) and complexity spotting (Task 2). For simplification, they built a multi‑candidate pipeline with GPT‑4o‑mini, selecting the best candidate via a reference‑free heuristic, and their Claude Sonnet 4 submission achieved a SARI of 47.43 and BLEU of 14.21, ranking third overall on the Task 1 leaderboard. For complexity spotting, they fine‑tuned a DeBERTa‑v3‑large NLI model on 350 K labeled pairs, achieving a macro F1 of 0.8081 (0.8085 in an ensemble) on binary over‑generation identification and 0.804 accuracy on multi‑class error classification, placing them second among unique teams.
By David L. Condrey