Tootfinder

CLARity: Reasoning Consistency Alone Can Teach Reinforced Experts
Jiuheng Lin, Cong Jiang, Zirui Wu, Jiarui Sun, Yansong Feng
https://arxiv.org/abs/2510.09278 https://

CLARity: Reasoning Consistency Alone Can Teach Reinforced Experts
Training expert LLMs in domains with scarce data is difficult, often relying on multiple-choice questions (MCQs). However, standard outcome-based reinforcement learning (RL) on MCQs is risky. While it may improve accuracy, we observe it often degrades reasoning quality such as logical consistency. Existing solutions to supervise reasoning, such as large-scale Process Reward Models (PRMs), are prohibitively expensive. To address this, we propose CLARity, a cost-effective RL framework that enhanc…

Literate Tracing
Matthew Sotoudeh
https://arxiv.org/abs/2510.09073 https://arxiv.org/pdf/2510.09073

Literate Tracing
As computer systems grow ever larger and more complex, a crucial task in software development is for one person (the system expert) to communicate to another (the system novice) how a certain program works. This paper reports on the author's experiences with a paradigm for program documentation that we call literate tracing. A literate trace explains a software system using annotated, concrete execution traces of the system. Literate traces complement both in-code comments (which often lack glo…

Experimental overview: quarkonia as probes for deconfinement
Michael Winn
https://arxiv.org/abs/2510.09075 https://arxiv.org/pdf/2510.09075

Experimental overview: quarkonia as probes for deconfinement
The current status of the experimental investigation of quarkonia in heavy-ion collisions as probes for deconfinement is reviewed. The emphasis is put on the qualitative understanding of charmonia (ccbar bound states) and bottomonia (bbbar) based on the results in the past few years. In particular, the observation of non-primordial J/psi (ccbar) production as signature of deconfinement based on different observables and the sequential suppression patterns of the Upsilon (bbbar) states are discu…

ProDiGy: Proximity- and Dissimilarity-Based Byzantine-Robust Federated Learning
Sena Ergisi, Luis Ma{\ss}ny, Rawad Bitar
https://arxiv.org/abs/2509.09534 https://

ProDiGy: Proximity- and Dissimilarity-Based Byzantine-Robust Federated Learning
Federated Learning (FL) emerged as a widely studied paradigm for distributed learning. Despite its many advantages, FL remains vulnerable to adversarial attacks, especially under data heterogeneity. We propose a new Byzantine-robust FL algorithm called ProDiGy. The key novelty lies in evaluating the client gradients using a joint dual scoring system based on the gradients' proximity and dissimilarity. We demonstrate through extensive numerical experiments that ProDiGy outperforms existing defen…

Visual Analysis of Time-Dependent Observables in Cell Signaling Simulations
Lena Cibulski, Fiete Haack, Adelinde Uhrmacher, Stefan Bruckner
https://arxiv.org/abs/2509.08589 http…

Visual Analysis of Time-Dependent Observables in Cell Signaling Simulations
The ability of a cell to communicate with its environment is essential for key cellular functions like replication, metabolism, or cell fate decisions. The involved molecular mechanisms are highly dynamic and difficult to capture experimentally. Simulation studies offer a valuable means for exploring and predicting how cell signaling processes unfold. We present a design study on the visual analysis of such studies to support 1) modelers in calibrating model parameters such that the simulated s…

Dyna-Mind: Learning to Simulate from Experience for Better AI Agents
Xiao Yu, Baolin Peng, Michel Galley, Hao Cheng, Qianhui Wu, Janardhan Kulkarni, Suman Nath, Zhou Yu, Jianfeng Gao
https://arxiv.org/abs/2510.09577

Dyna-Mind: Learning to Simulate from Experience for Better AI Agents
Reasoning models have recently shown remarkable progress in domains such as math and coding. However, their expert-level abilities in math and coding contrast sharply with their performance in long-horizon, interactive tasks such as web navigation and computer/phone-use. Inspired by literature on human cognition, we argue that current AI agents need ''vicarious trial and error'' - the capacity to mentally simulate alternative futures before acting - in order to enhance their understanding and p…

ReTraceQA: Evaluating Reasoning Traces of Small Language Models in Commonsense Question Answering
Francesco Maria Molfese, Luca Moroni, Ciro Porcaro, Simone Conia, Roberto Navigli
https://arxiv.org/abs/2510.09351

ReTraceQA: Evaluating Reasoning Traces of Small Language Models in Commonsense Question Answering
While Small Language Models (SLMs) have demonstrated promising performance on an increasingly wide array of commonsense reasoning benchmarks, current evaluation practices rely almost exclusively on the accuracy of their final answers, neglecting the validity of the reasoning processes that lead to those answers. To address this issue, we introduce ReTraceQA, a novel benchmark that introduces process-level evaluation for commonsense reasoning tasks. Our expert-annotated dataset reveals that in a…

Can We Reliably Rank Model Performance across Domains without Labeled Data?
Veronica Rammouz, Aaron Gonzalez, Carlos Cruzportillo, Adrian Tan, Nicole Beebe, Anthony Rios
https://arxiv.org/abs/2510.09519

Can We Reliably Rank Model Performance across Domains without Labeled Data?
Estimating model performance without labels is an important goal for understanding how NLP models generalize. While prior work has proposed measures based on dataset similarity or predicted correctness, it remains unclear when these estimates produce reliable performance rankings across domains. In this paper, we analyze the factors that affect ranking reliability using a two-step evaluation setup with four base classifiers and several large language models as error predictors. Experiments on t…

KORMo: Korean Open Reasoning Model for Everyone
Minjun Kim, Hyeonseok Lim, Hangyeol Yoo, Inho Won, Seungwoo Song, Minkyung Cho, Junhun Yuk, Changsu Choi, Dongjae Shin, Huige Lee, Hoyun Song, Alice Oh, Kyungtae Lim
https://arxiv.org/abs/2510.09426

KORMo: Korean Open Reasoning Model for Everyone
This work presents the first large-scale investigation into constructing a fully open bilingual large language model (LLM) for a non-English language, specifically Korean, trained predominantly on synthetic data. We introduce KORMo-10B, a 10.8B-parameter model trained from scratch on a Korean-English corpus in which 68.74% of the Korean portion is synthetic. Through systematic experimentation, we demonstrate that synthetic data, when carefully curated with balanced linguistic coverage and diver…

Fluent but Unfeeling: The Emotional Blind Spots of Language Models
Bangzhao Shu, Isha Joshi, Melissa Karnaze, Anh C. Pham, Ishita Kakkar, Sindhu Kothe, Arpine Hovasapian, Mai ElSherief
https://arxiv.org/abs/2509.09593

Fluent but Unfeeling: The Emotional Blind Spots of Language Models
The versatility of Large Language Models (LLMs) in natural language understanding has made them increasingly popular in mental health research. While many studies explore LLMs' capabilities in emotion recognition, a critical gap remains in evaluating whether LLMs align with human emotions at a fine-grained level. Existing research typically focuses on classifying emotions into predefined, limited categories, overlooking more nuanced expressions. To address this gap, we introduce EXPRESS, a benc…