A Minimalist Approach to LLM Reasoning: from Rejection Sampling to Reinforce

Reinforcement learning (RL) has become a prevailing approach for fine-tuning large language models (LLMs) on complex reasoning tasks. Among recent methods, GRPO stands out for its empirical success in training models such as DeepSeek-R1, yet the sources of its effectiveness remain poorly understood. In this work, we revisit GRPO from a reinforce-like algorithm perspective and analyze its core components. Surprisingly, we find that a simple rejection sampling baseline, RAFT, which trains only on positively rewarded samples, yields competitive performance than GRPO and PPO. Our ablation studies reveal that GRPO's main advantage arises from discarding prompts with entirely incorrect responses, rather than from its reward normalization. Motivated by this insight, we propose Reinforce-Rej, a minimal extension of policy gradient that filters both entirely incorrect and entirely correct samples. Reinforce-Rej improves KL efficiency and stability, serving as a lightweight yet effective alternative to more complex RL algorithms. We advocate RAFT as a robust and interpretable baseline, and suggest that future advances should focus on more principled designs for incorporating negative samples, rather than relying on them indiscriminately. Our findings provide guidance for future work in reward-based LLM post-training.

Correlative and Discriminative Label Grouping for Multi-Label Visual Prompt Tuning

Modeling label correlations has always played a pivotal role in multi-label image classification (MLC), attracting significant attention from researchers. However, recent studies have overemphasized co-occurrence relationships among labels, which can lead to overfitting risk on this overemphasis, resulting in suboptimal models. To tackle this problem, we advocate for balancing correlative and discriminative relationships among labels to mitigate the risk of overfitting and enhance model performance. To this end, we propose the Multi-Label Visual Prompt Tuning framework, a novel and parameter-efficient method that groups classes into multiple class subsets according to label co-occurrence and mutual exclusivity relationships, and then models them respectively to balance the two relationships. In this work, since each group contains multiple classes, multiple prompt tokens are adopted within Vision Transformer (ViT) to capture the correlation or discriminative label relationship within each group, and effectively learn correlation or discriminative representations for class subsets. On the other hand, each group contains multiple group-aware visual representations that may correspond to multiple classes, and the mixture of experts (MoE) model can cleverly assign them from the group-aware to the label-aware, adaptively obtaining label-aware representation, which is more conducive to classification. Experiments on multiple benchmark datasets show that our proposed approach achieves competitive results and outperforms SOTA methods on multiple pre-trained models.

CrystalFormer-RL: Reinforcement Fine-Tuning for Materials Design

Reinforcement fine-tuning has instrumental enhanced the instruction-following and reasoning abilities of large language models. In this work, we explore the applications of reinforcement fine-tuning to the autoregressive transformer-based materials generative model CrystalFormer (arXiv:2403.15734) using discriminative machine learning models such as interatomic potentials and property prediction models. By optimizing reward signals-such as energy above the convex hull and material property figures of merit-reinforcement fine-tuning infuses knowledge from discriminative models into generative models. The resulting model, CrystalFormer-RL, shows enhanced stability in generated crystals and successfully discovers crystals with desirable yet conflicting material properties, such as substantial dielectric constant and band gap simultaneously. Notably, we observe that reinforcement fine-tuning enables not only the property-guided novel material design ability of generative pre-trained model but also unlocks property-driven material retrieval from the unsupervised pre-training dataset. Leveraging rewards from discriminative models to fine-tune materials generative models opens an exciting gateway to the synergies of the machine learning ecosystem for materials.

AP-CAP: Advancing High-Quality Data Synthesis for Animal Pose Estimation via a Controllable Image Generation Pipeline

The task of 2D animal pose estimation plays a crucial role in advancing deep learning applications in animal behavior analysis and ecological research. Despite notable progress in some existing approaches, our study reveals that the scarcity of high-quality datasets remains a significant bottleneck, limiting the full potential of current methods. To address this challenge, we propose a novel Controllable Image Generation Pipeline for synthesizing animal pose estimation data, termed AP-CAP. Within this pipeline, we introduce a Multi-Modal Animal Image Generation Model capable of producing images with expected poses. To enhance the quality and diversity of the generated data, we further propose three innovative strategies: (1) Modality-Fusion-Based Animal Image Synthesis Strategy to integrate multi-source appearance representations, (2) Pose-Adjustment-Based Animal Image Synthesis Strategy to dynamically capture diverse pose variations, and (3) Caption-Enhancement-Based Animal Image Synthesis Strategy to enrich visual semantic understanding. Leveraging the proposed model and strategies, we create the MPCH Dataset (Modality-Pose-Caption Hybrid), the first hybrid dataset that innovatively combines synthetic and real data, establishing the largest-scale multi-source heterogeneous benchmark repository for animal pose estimation to date. Extensive experiments demonstrate the superiority of our method in improving both the performance and generalization capability of animal pose estimators.

What, How, Where, and How Well? A Survey on Test-Time Scaling in Large Language Models

As enthusiasm for scaling computation (data and parameters) in the pretraining era gradually diminished, test-time scaling (TTS), also referred to as ``test-time computing'' has emerged as a prominent research focus. Recent studies demonstrate that TTS can further elicit the problem-solving capabilities of large language models (LLMs), enabling significant breakthroughs not only in specialized reasoning tasks, such as mathematics and coding, but also in general tasks like open-ended Q&A. However, despite the explosion of recent efforts in this area, there remains an urgent need for a comprehensive survey offering a systemic understanding. To fill this gap, we propose a unified, multidimensional framework structured along four core dimensions of TTS research: what to scale, how to scale, where to scale, and how well to scale. Building upon this taxonomy, we conduct an extensive review of methods, application scenarios, and assessment aspects, and present an organized decomposition that highlights the unique functional roles of individual techniques within the broader TTS landscape. From this analysis, we distill the major developmental trajectories of TTS to date and offer hands-on guidelines for practical deployment. Furthermore, we identify several open challenges and offer insights into promising future directions, including further scaling, clarifying the functional essence of techniques, generalizing to more tasks, and more attributions.

ElimPCL: Eliminating Noise Accumulation with Progressive Curriculum Labeling for Source-Free Domain Adaptation

Source-Free Domain Adaptation (SFDA) aims to train a target model without source data, and the key is to generate pseudo-labels using a pre-trained source model. However, we observe that the source model often produces highly uncertain pseudo-labels for hard samples, particularly those heavily affected by domain shifts, leading to these noisy pseudo-labels being introduced even before adaptation and further reinforced through parameter updates. Additionally, they continuously influence neighbor samples through propagation in the feature space.To eliminate the issue of noise accumulation, we propose a novel Progressive Curriculum Labeling (ElimPCL) method, which iteratively filters trustworthy pseudo-labeled samples based on prototype consistency to exclude high-noise samples from training. Furthermore, a Dual MixUP technique is designed in the feature space to enhance the separability of hard samples, thereby mitigating the interference of noisy samples on their neighbors.Extensive experiments validate the effectiveness of ElimPCL, achieving up to a 3.4% improvement on challenging tasks compared to state-of-the-art methods.

Robust Dataset Distillation by Matching Adversarial Trajectories

Dataset distillation synthesizes compact datasets that enable models to achieve performance comparable to training on the original large-scale datasets. However, existing distillation methods overlook the robustness of the model, resulting in models that are vulnerable to adversarial attacks when trained on distilled data. To address this limitation, we introduce the task of ``robust dataset distillation", a novel paradigm that embeds adversarial robustness into the synthetic datasets during the distillation process. We propose Matching Adversarial Trajectories (MAT), a method that integrates adversarial training into trajectory-based dataset distillation. MAT incorporates adversarial samples during trajectory generation to obtain robust training trajectories, which are then used to guide the distillation process. As experimentally demonstrated, even through natural training on our distilled dataset, models can achieve enhanced adversarial robustness while maintaining competitive accuracy compared to existing distillation methods. Our work highlights robust dataset distillation as a new and important research direction and provides a strong baseline for future research to bridge the gap between efficient training and adversarial robustness.

Design of an Expression Recognition Solution Employing the Global Channel-Spatial Attention Mechanism

Facial expression recognition is a challenging classification task with broad application prospects in the field of human - computer interaction. This paper aims to introduce the methods of our upcoming 8th Affective Behavior Analysis in the Wild (ABAW) competition to be held at CVPR2025. To address issues such as low recognition accuracy caused by subtle expression changes and multi - scales in facial expression recognition in videos, we propose global channel - spatial attention and median - enhanced spatial - channel attention to strengthen feature processing for speech and images respectively. Secondly, to fully utilize the complementarity between the speech and facial expression modalities, a speech - and - facial - expression key - frame alignment technique is adopted to calculate the weights of speech and facial expressions. These weights are input into the feature fusion layer for multi - scale dilated fusion, which effectively improves the recognition rate of facial expression recognition. In the facial expression recognition task of the 6th ABAW competition, our method achieved excellent results on the official validation set, which fully demonstrates the effectiveness and competitiveness of the proposed method.

FedALT: Federated Fine-Tuning through Adaptive Local Training with Rest-of-the-World LoRA

Fine-tuning large language models (LLMs) in federated settings enables privacy-preserving adaptation but suffers from cross-client interference due to model aggregation. Existing federated LoRA fine-tuning methods, primarily based on FedAvg, struggle with data heterogeneity, leading to harmful cross-client interference and suboptimal personalization. In this work, we propose \textbf{FedALT}, a novel personalized federated LoRA fine-tuning algorithm that fundamentally departs from FedAvg. Instead of using an aggregated model to initialize local training, each client continues training its individual LoRA while incorporating shared knowledge through a separate Rest-of-the-World (RoTW) LoRA component. To effectively balance local adaptation and global information, FedALT introduces an adaptive mixer that dynamically learns input-specific weightings between the individual and RoTW LoRA components using the Mixture-of-Experts (MoE) principle. Through extensive experiments on NLP benchmarks, we demonstrate that FedALT significantly outperforms state-of-the-art personalized federated LoRA fine-tuning methods, achieving superior local adaptation without sacrificing computational efficiency.

Interactive Multimodal Fusion with Temporal Modeling

This paper presents our method for the estimation of valence-arousal (VA) in the 8th Affective Behavior Analysis in-the-Wild (ABAW) competition. Our approach integrates visual and audio information through a multimodal framework. The visual branch uses a pre-trained ResNet model to extract spatial features from facial images. The audio branches employ pre-trained VGG models to extract VGGish and LogMel features from speech signals. These features undergo temporal modeling using Temporal Convolutional Networks (TCNs). We then apply cross-modal attention mechanisms, where visual features interact with audio features through query-key-value attention structures. Finally, the features are concatenated and passed through a regression layer to predict valence and arousal. Our method achieves competitive performance on the Aff-Wild2 dataset, demonstrating effective multimodal fusion for VA estimation in-the-wild.