The paper introduces RoSA, a saliency-aware fine‑tuning framework for long‑text image‑text alignment that addresses the failure mode of saliency collapse. RoSA decomposes image‑text pairs into salient and contextual views using a modality‑independent strategy that combines visual saliency estimation with LLM‑based text decomposition, and employs auxiliary objectives to align these views. The method improves long‑text retrieval performance and robustness to saliency collapse, and its auxiliary objectives can be added to existing tuning methods for consistent gains.
By Qiuyu Kong, Zanxi Ruan, Marco Cristani, Yiming Wang
The paper proposes an agentic framework for multimodal emotion reasoning that uses frozen models to avoid costly instruction tuning. It employs modality‑specific perception agents, an integration agent, and an MLLM critic to generate emotion predictions with evidence citations, followed by an evidence‑grounded verification stage that identifies errors and applies targeted prompt patches without updating model weights. On the MER2025 dataset, this approach improves F‑score across several large language‑vision models, outperforming static pipelines and matching or exceeding reflection baselines.
By Elyas Rashno, Farhana Zulkernine
The paper introduces an AI surrogate modelling framework that accelerates wildfire hazard mapping across Canada. By training a multi‑task deep learning model on national‑scale data, the surrogate reproduces key outputs of the computationally intensive BurnP3+ simulator—burn probability, mean fire intensity, and mean rate of spread—in seconds instead of days or weeks. The framework includes standardized training, fine‑tuning, and evaluation protocols, domain‑specific metrics, and counterfactual evaluation to ensure robust out‑of‑distribution performance, achieving up to 806‑fold CPU speed‑up and 13% memory usage compared to the original simulator.
By Hager Radi Abdelwahed, Toluwaleke Olutayo, Chirag Goel, Jana Osea, Aldo Zaimi, Tom Swystun, Sandy Erni, Francisco Emiliano Lopez Saavedra, Yan Boulanger, Ga\'etan Marceau Caron
arXiv:2610.31911v1 Announce Type: new
Abstract: As large language models (LLMs) process increasingly long sequences, attention becomes a major computational bottleneck. Sparse attention mostly target...
By Keqi Deng, Ruchao Fan, Jinyu Li
arXiv:2610.32318v1 Announce Type: new
Abstract: System One models, or decision models, answer a question about an input with a calibrated probability instead of generated text. Within weeks of the fi...
By Michael Dinzinger, Kanishka Ghosh Dastidar, Michael Granitzer
arXiv:2610.32982v1 Announce Type: new
Abstract: This article argues that instruction-fine-tuned language models should be understood as implementing a genre, what is proposed as the instructional-con...
By James E. Dobson
arXiv:2610.32132v1 Announce Type: cross
Abstract: Proactive streaming video understanding requires assistants to track unfolding events and respond when a user's requested condition is met. To make t...
By Xueyun Tian, Bingbing Xu, Jiaming Liu, Jinpeng Yu, Yang Shi, Yu Zhao, Ruihua Huang, Yuanzhuo Wang, Huawei Shen
arXiv:2607.15209v4 Announce Type: replace
Abstract: Multilingual pre-trained language models such as XLM-R perform well for major languages but struggle with low-resource Ge'ez-script languages, larg...
By Hailay Kidu Teklehaymanot, Debela Desalegn Yadeta, Wolfgang Nejdl
arXiv:2610.31765v1 Announce Type: new
Abstract: Generating multiple videos of a dynamic scene from a single reference video requires synthesizing both observed and unobserved content consistently acr...
By Farhad G. Zanjani, Animesh Karnewar, Fatih Porikli, Amirhossein Habibian
arXiv:2610.31898v1 Announce Type: new
Abstract: Traffic safety videos provide rich evidence for understanding interactions between pedestrians and vehicles, but synthetic-to-real evaluation makes thi...
By Blessing Agyei Kyem, Richard Dzinyela, Joshua Kofi Asamoah, Eugene Denteh, Andrews Danyo, Solomon Owusu-Ansah, Dickson Gyimah Asiedu, Armstrong Aboah
arXiv:2610.31951v1 Announce Type: new
Abstract: Surgical point tracking requires persistent correspondence through tissue deformation and occlusion, yet dense trajectory annotations are costly to obt...
By Bing Fan, Anwesa Choudhuri, Zhongpai Gao, Meng Zheng, Benjamin Planche, Terrence Chen, Heng Fan, Ziyan Wu
arXiv:2510.01719v5 Announce Type: replace
Abstract: Evaluation of multimodal reasoning models is typically reduced to a single accuracy score, implicitly treating reasoning as a unitary capability. W...
By Jiwan Chung, Neel Joshi, Pratyusha Sharma, Youngjae Yu, Vibhav Vineet
arXiv:2610.32573v1 Announce Type: new
Abstract: Camera-trap species classification is dominated by large models from well-resourced organisations, which often lack the species-level distinctions regi...
By Paul Fergus
arXiv:2606.15417v2 Announce Type: replace
Abstract: Action reasoning in egocentric video requires capturing fine-grained transitions of hand-object interactions, a task where general-purpose Vision-L...
By Bessie Dominguez-Dager, Francisco Gomez-Donoso, Miguel Cazorla, Marc Pollefeys, Daniel Barath, Zuria Bauer
arXiv:2610.15571v2 Announce Type: replace
Abstract: In this work, we propose a transfer learning algorithm for covariate shift, integrating sample reweighting with representation learning. A key feat...
By Yuling Jiao, Lican Kang, Yang Xiang, Jerry Zhijian Yang, Fusheng Zhou
ProfileSim is a profile‑driven user simulation framework designed for financial marketing outbound calls. It jointly models user demographics, Big Five personality traits, and business intents reverse‑engineered from real call logs, allowing fine‑grained control over simulated behavior. The system is fine‑tuned on real voice dialogue data and further refined with Group Relative Policy Optimization using hierarchical rewards, achieving high colloquial naturalness, strong user profile adherence, and low agent scores on long‑tail intents while producing shorter user utterances.
By Juan Li, Fangshu Chang, Xinyi Jiang, Ling Huang, Yunzhang Chen
The paper investigates whether supervised contrastive learning objectives produce a regular simplex geometry in representation space when trained on long‑tailed data. It shows that existing objectives fail to achieve this geometry because they depend on class sample counts. The authors propose a fully balanced contrastive objective that provably forms a regular simplex, leading to a more balanced representation space and improved performance on long‑tailed recognition and transfer learning tasks.
By Joonhyeok Jeong, Jy-yong Sohn, Kibok Lee
arXiv:2610. 32142v1 Announce Type: cross Abstract: Low-Rank Adaptation (LoRA) is often highly sensitive to learning-rate (LR) choices.
By Disen Liao, Yaoliang Yu
The paper introduces TopKLoRA, an interpretable parameter‑efficient fine‑tuning method that activates only the top‑k adapter latents per token. Using this sparse LoRA, the authors fine‑tune large language models on a poisoned dataset, identify and ablate the minimal circuit responsible for a backdoor, and show that the resulting circuits are an order of magnitude smaller than dense LoRA baselines. They also demonstrate how to construct model organisms with localized circuits via gradient‑routing, supporting the development of interpretable‑by‑design architectures.
By Marek Masiak, Andrzej Szablewski, Nicola Cancedda, Constantin Venhoff
The paper introduces Self‑Direction, a method that lets agents learn from mistakes by reflecting on failed trajectories, distilling lessons into lightweight low‑rank adapters, and storing them for future use. These adapters add only about 0.02% of the base model’s parameters, keeping the original weights frozen and avoiding costly fine‑tuning. Experiments on Qwen‑based agents across research, tool‑use, and terminal tasks show 8.9–17.0 percentage‑point gains in success rates, reduced repeated errors, and transfer of lessons to new tasks.
By Yuval Kansal, Karen Ullrich, Polina Kirichenko, Ilia Kulikov, Niraj K. Jha, Mohammad Pezeshki, Mark Ibrahim