Publications

Detoxifying LLMs via Representation Erasure-Based Preference Optimization

Eleni Triantafillou

Daniel M. Roy

Large language models (LLMs) trained on webscale data can produce toxic outputs, raising concerns for safe deployment. Prior defenses, based… (voir plus) on applications of DPO, NPO, and similar algorithms, reduce the likelihood of harmful continuations, but not robustly so: they are vulnerable to adversarial prompting and easily undone by fine-tuning-based relearning attacks. Indeed, research has shown that these edits to the model are superficial: linear probing reveals that harmful "directions" remain present in representations. To address this, we propose Representation Erasure-based Preference Optimization (REPO), reformulating detoxification as a token-level preference problem. Using a novel objective with preference data, we force the representations of toxic continuations to converge toward their benign counterparts. Our mechanistic analysis reveals that this granular approach is critical: unlike baselines, REPO induces deep, localized edits to toxicity-encoding neurons while preserving general model utility. Exhaustive evaluations show that REPO achieves state-of-the-art robustness, stopping sophisticated threats-including relearning attacks and enhanced GCG jailbreaks-where existing representation- and output-based methods fail.

2026-02-23

arXiv (inconnu)

doi.org

arxiv.org

WildSVG: Towards Reliable SVG Generation Under Real-Word Conditions

Marco Terral

Haotian Zhang

Tianyang Zhang

Meng Lin

Xiaoqing Xie

Haoran Dai

Darsh Kaushik

Pai Peng

Nicklas Scharpff

David Vázquez

Joan Rodriguez

We introduce the task of SVG extraction, which consists in translating specific visual inputs from an image into scalable vector graphics. E… (voir plus)xisting multimodal models achieve strong results when generating SVGs from clean renderings or textual descriptions, but they fall short in real-world scenarios where natural images introduce noise, clutter, and domain shifts. A central challenge in this direction is the lack of suitable benchmarks. To address this need, we introduce the WildSVG Benchmark, formed by two complementary datasets: Natural WildSVG, built from real images containing company logos paired with their SVG annotations, and Synthetic WildSVG, which blends complex SVG renderings into real scenes to simulate difficult conditions. Together, these resources provide the first foundation for systematic benchmarking SVG extraction. We benchmark state-of-the-art multimodal models and find that current approaches perform well below what is needed for reliable SVG extraction in real scenarios. Nonetheless, iterative refinement methods point to a promising path forward, and model capabilities are steadily improving

2026-02-23

arXiv (prépublication)

doi.org

arxiv.org

Grokking Finite-Dimensional Algebra

Pascal Jr Tikeng Notsawo

Guillaume Dumas

Guillaume Rabusseau

This paper investigates the grokking phenomenon, which refers to the sudden transition from a long memorization to generalization observed d… (voir plus)uring neural networks training, in the context of learning multiplication in finite-dimensional algebras (FDA). While prior work on grokking has focused mainly on group operations, we extend the analysis to more general algebraic structures, including non-associative, non-commutative, and non-unital algebras. We show that learning group operations is a special case of learning FDA, and that learning multiplication in FDA amounts to learning a bilinear product specified by the algebra's structure tensor. For algebras over the reals, we connect the learning problem to matrix factorization with an implicit low-rank bias, and for algebras over finite fields, we show that grokking emerges naturally as models must learn discrete representations of algebraic elements. This leads us to experimentally investigate the following core questions: (i) how do algebraic properties such as commutativity, associativity, and unitality influence both the emergence and timing of grokking, (ii) how structural properties of the structure tensor of the FDA, such as sparsity and rank, influence generalization, and (iii) to what extent generalization correlates with the model learning latent embeddings aligned with the algebra's representation. Our work provides a unified framework for grokking across algebraic structures and new insights into how mathematical structure governs neural network generalization dynamics.

2026-02-22

arXiv (prépublication)

doi.org

arxiv.org

Learning to Solve Complex Problems via Dataset Decomposition

Wanru Zhao

Lucas Caccia

Zhengyan Shi

Minseon Kim

Weijia Xu

Alessandro Sordoni

Curriculum learning is a class of training strategies that organizes the data being exposed to a model by difficulty, gradually from simpler… (voir plus) to more complex examples. This research explores a reverse curriculum generation approach that recursively decomposes complex datasets into simpler, more learnable components. We propose a teacher-student framework where the teacher is equipped with the ability to reason step-by-step, which is used to recursively generate easier versions of examples, enabling the student model to progressively master difficult tasks. We propose a novel scoring system to measure data difficulty based on its structural complexity and conceptual depth, allowing curriculum construction over decomposed data. Experiments on math datasets (MATH and AIME) and code generation datasets demonstrate that models trained with curricula generated by our approach exhibit superior performance compared to standard training on original datasets.

2026-02-22

arXiv (prépublication)

doi.org

arxiv.org

Celo2: Towards Learned Optimization Free Lunch

Abhinav Moudgil

Boris Knyazev

Eugene Belilovsky

Learned optimizers are powerful alternatives to hand-designed update rules like Adam, yet they have seen limited practical adoption since th… (voir plus)ey often fail to meta-generalize beyond their training distribution and incur high meta-training cost. For instance, prior work, VeLO, scaled meta-training to 4,000 TPU months (

2026-02-21

arXiv (prépublication)

doi.org

arxiv.org

Panorama of Soft Tissue Tumours at a Tertiary Care Centre in Bihar: A Retrospective Observational Study

Ankit Anand

Vibhuti Kumar

Objective and Aim: Soft tissue tumors (STTs) represent a heterogeneous group of neoplasms with diverse histogenesis, biological behavior, an… (voir plus)d clinical outcomes. The present study aims to evaluate the spectrum, frequency, demographic distribution, anatomical location, and histopathological patterns of soft tissue tumors diagnosed at a tertiary care center in Bihar, India, with special emphasis on benign–malignant correlation and clinicopathological characteristics. Materials and Methods: This retrospective observational study was conducted in the Department of Pathology at a tertiary care teaching hospital in Bihar over a period of five years (January 2019–December 2023). All histopathologically confirmed cases of soft tissue tumors were included. Tumors were classified according to the WHO Classification of Soft Tissue and Bone Tumors (2020). Statistical analysis was performed using SPSS version 26.0. Descriptive statistics, chi-square test, and logistic regression analysis were applied. Results: A total of 312 cases of soft tissue tumors were analyzed. Benign tumors constituted 76.9%, intermediate tumors 7.4%, and malignant tumors 15.7%. The most common benign tumor was lipoma (38.1%), while undifferentiated pleomorphic sarcoma (21.4%) was the most frequent malignant tumor. Malignant tumors were significantly associated with age >40 years (p 0.001) and deep-seated location (p = 0.002). Conclusion: Soft tissue tumors in Bihar show a predominance of benign lesions with lipoma bei

2026-02-21

International Journal of Current Pharmaceutical Review and Research (publié)

doi.org

VectorGym: A Multitask Benchmark for SVG Code Generation, Sketching, and Editing

Juan Rodriguez

Haotian Zhang

Abhay Puri

Tianyang Zhang

Rishav Pramanik

Meng Lin

Xiaoqing Xie

Marco Terral

Darsh Kaushik

Aly Shariff

Perouz Taslakian

Spandana Gella

Sai Rajeswar

David Vázquez

Christopher Pal

Marco Pedersoli

We introduce VectorGym, a comprehensive benchmark suite for Scalable Vector Graphics (SVG) that spans generation from text and sketches, com… (voir plus)plex editing, and visual understanding. VectorGym addresses the lack of realistic, challenging benchmarks aligned with professional design workflows. Our benchmark comprises four tasks with expert human-authored annotations: the novel Sketch2SVG task (VG-Sketch); a new SVG editing dataset (VG-Edit) featuring complex, multi-step edits with higher-order primitives; Text2SVG generation (VG-Text); and SVG captioning (VG-Cap). Unlike prior benchmarks that rely on synthetic edits, VectorGym provides gold-standard human annotations that require semantic understanding and design intent. We also propose a multi-task reinforcement learning approach that jointly optimizes across all four tasks using rendering-based rewards. Our method, built on GRPO with curriculum learning, trains a Qwen3-VL 8B model that achieves state-of-the-art performance among open-source models, surpassing much larger models including Qwen3-VL 235B and matching GPT-4o. We also introduce a VLM-as-a-Judge metric for SVG generation, validated through human correlation studies. Our evaluation of frontier VLMs reveals significant performance gaps, positioning VectorGym as a rigorous framework for advancing visual code generation. VectorGym is publicly available on huggingface.co/datasets/ServiceNow/VectorGym.

2026-02-21

arXiv (prépublication)

doi.org

arxiv.org

EEG-based quantification of chronic pain in cats: A proof-of-concept study using the Piq algorithm

Aliénor Delsart

Colince Segning

Aude Castel

Colombe Otis

Guillaume Dumas

Maxim Moreau

Bertrand Lussier

Rubens Da Silva

Karen Barros Parron Fernandes

Johanne Martel-Pelletier

Jean-Pierre Pelletier

Eric Troncy

Suzy Ngomo

While chronic pain assessment in household pets remains challenging, the use of non-invasive electroencephalography (EEG) in cats has shown … (voir plus)promise to identify pain more objectively in this species. A novel EEG-based algorithm - Pain identification and quantification (Piq) - was originally developed in humans to quantify pain intensity. In this proof-of-concept study, the objective was to evaluate whether the Piq algorithm could be explored for feasibility to identify and quantify chronic osteoarthritic (OA) pain in cats. Adult neutered cats (n = 5 including n = 2 with osteoarthritis, OA) were assessed for their functional impairment (Montreal instrument for cat arthritis testing for use by veterinarians, MI-CAT(V)) and neuro-sensitization at both peripheral (Paw Withdrawal Threshold, PWT) and spinal (response to mechanical temporal summation, RMTS) levels. Resting-state EEG recordings were acquired from Cz, C3/C4 under conscious and sedated conditions. The first five minutes of EEG data were analyzed using the Piq algorithm, with Piq scores ≥ 10 % used as an exploratory threshold transferred from human studies. Pain-free cats showed gamma frequency band Piq scores  10 % while OA cats exceeded 10 % in both conscious and sedated conditions at Cz. Piq scores were negatively correlated with PWT, sug

2026-02-20

Veterinary Journal (publié)

doi.org

Give Users the Wheel: Towards Promptable Recommendation Paradigm

Fuyuan Lyu

Chenglin Luo

Qiyuan Zhang

Yupeng Hou

Haolun Wu

Xing Tang

Xue Liu

Jin L.C. Guo

xiuqiang He

Conventional sequential recommendation models have achieved remarkable success in mining implicit behavioral patterns. However, these archit… (voir plus)ectures remain structurally blind to explicit user intent: they struggle to adapt when a user's immediate goal (e.g., expressed via a natural language prompt) deviates from their historical habits. While Large Language Models (LLMs) offer the semantic reasoning to interpret such intent, existing integration paradigms force a dilemma: LLM-as-a-recommender paradigm sacrifices the efficiency and collaborative precision of ID-based retrieval, while Reranking methods are inherently bottlenecked by the recall capabilities of the underlying model. In this paper, we propose Decoupled Promptable Sequential Recommendation (DPR), a model-agnostic framework that empowers conventional sequential backbones to natively support Promptable Recommendation, the ability to dynamically steer the retrieval process using natural language without abandoning collaborative signals. DPR modulates the latent user representation directly within the retrieval space. To achieve this, we introduce a Fusion module to align the collaborative and semantic signals, a Mixture-of-Experts (MoE) architecture that disentangles the conflicting gradients from positive and negative steering, and a three-stage training strategy that progressively aligns the semantic space of prompts with the collaborative space. Extensive experiments on real-world datasets demonstrate that DPR significantly outperforms state-of-the-art baselines in prompt-guided tasks while maintaining competitive performance in standard sequential recommendation scenarios.

2026-02-20

arXiv (prépublication)

doi.org

arxiv.org

Sociodynamics of Reinforcement Learning

Yann Bouteiller

Karthik Soma

Giovanni Beltrame

Reinforcement Learning (RL) has emerged as a core algorithmic paradigm explicitly driving innovation in a growing number of industrial appli… (voir plus)cations, including large language models and quantitative finance. Furthermore, computational neuroscience has long found evidence of natural forms of RL in biological brains. Therefore, it is crucial for the study of social dynamics to develop a scientific understanding of how RL shapes population behaviors. We leverage the framework of Evolutionary Game Theory (EGT) to provide building blocks and insights toward this objective. We propose a methodology that enables simulating large populations of RL agents in simple game theoretic interaction models. More specifically, we derive fast and parallelizable implementations of two fundamental revision protocols from multi-agent RL - Policy Gradient (PG) and Opponent-Learning Awareness (LOLA) - tailored for population simulations of random pairwise interactions in stateless normal-form games. Our methodology enables us to simulate large populations of 200,000 independent co-learning agents, yielding compelling insights into how non-stationarity-aware learners affect social dynamics. In particular, we find that LOLA learners promote cooperation in the Stag Hunt model, delay cooperative outcomes in the Hawk-Dove model, and reduce strategy diversity in the Rock-Paper-Scissors model.

2026-02-20

Transactions on Machine Learning Research (accepté)

openreview.net