Open weight models, which are ubiquitous, rarely provide access to their training data or loss function. This makes modifying such models for tasks such as pruning or unlearning constrained by this unavailability an active area of research. Existing techniques typically require gradients or ground-truth labels, rendering them infeasible in settings with limited computational resources. In this work, we investigate the fundamental question of identifying components that are critical to the model's predictive performance, without access to either gradients or the loss function, and with only distributional access such as synthetic data. We theoretically demonstrate that the global reconstruction error is linearly bounded by local reconstruction errors for Lipschitz-continuous networks such as CNNs and well-trained Transformers (which, contrary to existing literature, we find exhibit Lipschitz continuity). This motivates using the locally reconstructive behavior of component subsets to quantify their global importance, via a metric that we term Subset Fidelity. In the uncorrelated features setting, selecting individual components via their Subset Fidelity scores is optimal, which we use to propose ModHiFi, an algorithm for model modification that requires no training data or loss function access. ModHiFi-P, for structured pruning, achieves an 11% speedup over the current state of the art on ImageNet models and competitive performance on language models. ModHiFi-U, for classwise unlearning, achieves complete unlearning on CIFAR-10 without fine-tuning and demonstrates competitive performance on Swin Transformers.

Abstract--Long-tailed multi-label visual recognition poses a significant challenge, as images typically contain multiple labels with highly imbalanced class distributions, leading to biased models that favor head classes while underperforming on tail classes. Recent efforts have leveraged pre-trained vision-language models, such as CLIP, alongside long-tailed learning techniques to exploit rich visual-textual priors for improved performance. However, existing methods often derive semantic inter-class relationships directly from imbalanced datasets, resulting in unreliable correlations for tail classes due to data scarcity . Moreover, CLIP's zero-shot paradigm is optimized for single-label image-text matching, making it suboptimal for multi-label tasks. The framework incorporates a graph convolutional network for label-aware propagation and learnable soft prompts for refined embeddings. Moreover, it improves generalization through test-time ensembling and realigns visual-textual modalities using parameter-efficient fine-tuning to avert overfitting on tail classes without compromising head class performance. Recent progress in long-tailed visual recognition has mainly centered on the single-label multi-class setting. However, real-world images often encompass multiple objects and concepts, giving rise to long-tailed multi-label visual recognition (LTML), where each image is associated with multiple labels exhibiting long-tailed distributions [5], [6], [7], [8], [9], [10]. LTML introduces additional intricacies compared to its single-label multi-class counterpart, including label co-occurrences, which create interdependencies among classes, and intra-class imbalances between positive and negative instances. Existing LTML methods have primarily relied on strategies like class re-sampling, loss re-weighting, and specialized architectures, typically built upon ImageNet-pre-trained convolutional neural networks (CNNs) as backbones [5], [6], [7]. Wei Tang is with the School of Computer Science and Engineering, Southeast University, Nanjing 210096, China, the Key Laboratory of Computer Network and Information Integration (Southeast University), MoE, China, and with the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), Abu Dhabi, UAE (e-mail: tangw@seu.edu.cn).

The rapid proliferation of Large Language Models (LLMs) has revolutionized AI-assisted code generation. This rapid development of LLMs has outpaced our ability to properly benchmark them. Prevailing benchmarks emphasize unit-test pass rates and syntactic correctness. Such metrics understate the difficulty of many real-world problems that require planning, optimization, and strategic interaction. We introduce a multi-agent reasoning-driven benchmark based on a real-world logistics optimization problem (Auction, Pickup, and Delivery Problem) that couples competitive auctions with capacity-constrained routing. The benchmark requires building agents that can (i) bid strategically under uncertainty and (ii) optimize planners that deliver tasks while maximizing profit. We evaluate 40 LLM-coded agents (by a wide range of state-of-the-art LLMs under multiple prompting methodologies, including vibe coding) against 17 human-coded agents developed before the advent of LLMs. Our results over 12 double all-play-all tournaments and $\sim 40$k matches demonstrate (i) a clear superiority of human(graduate students)-coded agents: the top 5 spots are consistently won by human-coded agents, (ii) the majority of LLM-coded agents (33 out of 40) are beaten by very simple baselines, and (iii) given the best human solution as an input and prompted to improve upon, the best performing LLM makes the solution significantly worse instead of improving it. Our results highlight a gap in LLMs' ability to produce code that works competitively in the real-world, and motivate new evaluations that emphasize reasoning-driven code synthesis in real-world scenarios.

Diffusion Models have emerged as a leading class of generative models, yet their iterative sampling process remains computationally expensive. Timestep distillation is a promising technique to accelerate generation, but it often requires extensive training and leads to image quality degradation. Furthermore, fine-tuning these distilled models for specific objectives, such as aesthetic appeal or user preference, using Reinforcement Learning (RL) is notoriously unstable and easily falls into reward hacking. In this work, we introduce Flash-DMD, a novel framework that enables fast convergence with distillation and joint RL-based refinement. Specifically, we first propose an efficient timestep-aware distillation strategy that significantly reduces training cost with enhanced realism, outperforming DMD2 with only $2.1\%$ its training cost. Second, we introduce a joint training scheme where the model is fine-tuned with an RL objective while the timestep distillation training continues simultaneously. We demonstrate that the stable, well-defined loss from the ongoing distillation acts as a powerful regularizer, effectively stabilizing the RL training process and preventing policy collapse. Extensive experiments on score-based and flow matching models show that our proposed Flash-DMD not only converges significantly faster but also achieves state-of-the-art generation quality in the few-step sampling regime, outperforming existing methods in visual quality, human preference, and text-image alignment metrics. Our work presents an effective paradigm for training efficient, high-fidelity, and stable generative models. Codes are coming soon.

Background: As large language models (LLMs) become increasingly integrated into digital health education and assessment workflows, their capabilities in supporting high-stakes, domain-specific certification tasks remain underexplored.In China, the national pharmacist licensure exam serves as a standardized benchmark for evaluating pharmacists' clinical and theoretical competencies. Objective: This study aimed to compare the performance of two LLMs: ChatGPT-4o and DeepSeek-R1 on real questions from the Chinese Pharmacist Licensing Examination (2017-2021), and to discuss the implications of these performance differences for AI-enabled formative evaluation. Methods: A total of 2,306 multiple-choice (text-only) questions were compiled from official exams, training materials, and public databases. Questions containing tables or images were excluded. Each item was input in its original Chinese format, and model responses were evaluated for exact accuracy. Pearson's Chi-squared test was used to compare overall performance, and Fisher's exact test was applied to year-wise multiple-choice accuracy. Results: DeepSeek-R1 outperformed ChatGPT-4o with a significantly higher overall accuracy (90.0% vs. 76.1%, p < 0.001). Unit-level analyses revealed consistent advantages for DeepSeek-R1, particularly in foundational and clinical synthesis modules. While year-by-year multiple-choice performance also favored DeepSeek-R1, this performance gap did not reach statistical significance in any specific unit-year (all p > 0.05). Conclusion: DeepSeek-R1 demonstrated robust alignment with the structural and semantic demands of the pharmacist licensure exam. These findings suggest that domain-specific models warrant further investigation for this context, while also reinforcing the necessity of human oversight in legally and ethically sensitive contexts.

Abstract--Recent advances in large language models have created new opportunities for stylometry, the study of writing styles and authorship. Two challenges, however, remain central: training generative models when no paired data exist, and evaluating stylistic text without relying only on human judgment. In this work, we present a framework for both generating and evaluating sentences in the style of 19th-century novelists. Large language models are fine-tuned with minimal, single-token prompts to produce text in the voices of authors such as Dickens, Austen, Twain, Alcott, and Melville. T o assess these generative models, we employ a transformer-based detector trained on authentic sentences, using it both as a classifier and as a tool for stylistic explanation. We complement this with syntactic comparisons and explainable AI methods, including attention-based and gradient-based analyses, to identify the linguistic cues that drive stylistic imitation. Our findings show that the generated text reflects the authors' distinctive patterns and that AI-based evaluation offers a reliable alternative to human assessment. All artifacts of this work are published online. The ability to recognize and reproduce an author's writing style has long fascinated both literary scholars and computer scientists. Stylometry, the quantitative study of writing style, rests on the idea that every author leaves behind unconscious patterns in vocabulary, syntax, and rhythm [2, 3]. These patterns have been analyzed for centuries in questions of disputed authorship, the study of literary traditions, and more recently in applications such as security and forensics [4].

Large Language Models (LLMs) show strong generalization across diverse tasks, yet the internal decision-making processes behind their predictions remain opaque. In this work, we study the geometry of hidden representations in LLMs through the lens of \textit{intrinsic dimension} (ID), focusing specifically on decision-making dynamics in a multiple-choice question answering (MCQA) setting. We perform a large-scale study, with 28 open-weight transformer models and estimate ID across layers using multiple estimators, while also quantifying per-layer performance on MCQA tasks. Our findings reveal a consistent ID pattern across models: early layers operate on low-dimensional manifolds, middle layers expand this space, and later layers compress it again, converging to decision-relevant representations. Together, these results suggest LLMs implicitly learn to project linguistic inputs onto structured, low-dimensional manifolds aligned with task-specific decisions, providing new geometric insights into how generalization and reasoning emerge in language models.

Satellite constellations in low-Earth orbit are now widespread, enabling positioning, Earth imaging, and communications. In this paper we address the solution of learning problems using these satellite constellations. In particular, we focus on a federated approach, where satellites collect and locally process data, with the ground station aggregating local models. We focus on designing a novel, communication-efficient algorithm that still yields accurate trained models. To this end, we employ several mechanisms to reduce the number of communications with the ground station (local training) and their size (compression). We then propose an error feedback mechanism that enhances accuracy, which yields, as a byproduct, an algorithm-agnostic error feedback scheme that can be more broadly applied. We analyze the convergence of the resulting algorithm, and compare it with the state of the art through simulations in a realistic space scenario, showcasing superior performance.

ABSTRACT Mispronunciation Detection and Diagnosis (MDD) is crucial for language learning and speech therapy. Unlike conventional methods that require scoring models or training phoneme-level models, we propose a novel training-free framework that leverages retrieval techniques with a pre-trained Automatic Speech Recognition model. Our method avoids phoneme-specific modeling or additional task-specific training, while still achieving accurate detection and diagnosis of pronunciation errors. Experiments on the L2-ARCTIC dataset show that our method achieves a superior F1 score of 69.60% while avoiding the complexity of model training. Index T erms-- Mispronunciation detection and diagnosis, retrieval-based methods, training-free framework, automatic pronunciation assessment 1. INTRODUCTION Mispronunciation Detection and Diagnosis is a fundamental task in Computer-Assisted Pronunciation Training (CAPT).

With the advancement of large language models (LLMs), their performance on multiple-choice question (MCQ) tasks has improved significantly. However, existing approaches face key limitations: answer choices are typically presented to LLMs without contextual grounding or explanation. This absence of context can lead to incomplete exploration of all possible answers, ultimately degrading the models' reasoning capabilities. To address these challenges, we introduce BiasPrompting, a novel inference framework that guides LLMs to generate and critically evaluate reasoning across all plausible answer options before reaching a final prediction. It consists of two components: first, a reasoning generation stage, where the model is prompted to produce supportive reasonings for each answer option, and then, a reasoning-guided agreement stage, where the generated reasonings are synthesized to select the most plausible answer. Through comprehensive evaluations, BiasPrompting demonstrates significant improvements in five widely used multiple-choice question answering benchmarks. Our experiments showcase that BiasPrompting enhances the reasoning capabilities of LLMs and provides a strong foundation for tackling complex and challenging questions, particularly in settings where existing methods underperform.