Scaling RL to Long Videos

We introduce a full-stack framework that scales up reasoning in vision-language models (VLMs) to long videos, leveraging reinforcement learning. We address the unique challenges of long video reasoning by integrating three critical components: (1) a large-scale dataset, LongVideo-Reason, comprising 52K long video QA pairs with high-quality reasoning annotations across diverse domains such as sports, games, and vlogs; (2) a two-stage training pipeline that extends VLMs with chain-of-thought supervised fine-tuning (CoT-SFT) and reinforcement learning (RL); and (3) a training infrastructure for long video RL, named Multi-modal Reinforcement Sequence Parallelism (MR-SP), which incorporates sequence parallelism and a vLLM-based engine tailored for long video, using cached video embeddings for efficient rollout and prefilling. In experiments, LongVILA-R1-7B achieves strong performance on long video QA benchmarks such as VideoMME. It also outperforms Video-R1-7B and even matches Gemini-1.5-Pro across temporal reasoning, goal and purpose reasoning, spatial reasoning, and plot reasoning on our LongVideo-Reason-eval benchmark. Notably, our MR-SP system achieves up to 2.1x speedup on long video RL training. LongVILA-R1 demonstrates consistent performance gains as the number of input video frames scales. LongVILA-R1 marks a firm step towards long video reasoning in VLMs. In addition, we release our training system for public availability that supports RL training on various modalities (video, text, and audio), various models (VILA and Qwen series), and even image and video generation models. On a single A100 node (8 GPUs), it supports RL training on hour-long videos (e.g., 3,600 frames / around 256k tokens).

Generalization Bound of Gradient Flow through Training Trajectory and Data-dependent Kernel

Gradient-based optimization methods have shown remarkable empirical success, yet their theoretical generalization properties remain only partially understood. In this paper, we establish a generalization bound for gradient flow that aligns with the classical Rademacher complexity bounds for kernel methods-specifically those based on the RKHS norm and kernel trace-through a data-dependent kernel called the loss path kernel (LPK). Unlike static kernels such as NTK, the LPK captures the entire training trajectory, adapting to both data and optimization dynamics, leading to tighter and more informative generalization guarantees. Moreover, the bound highlights how the norm of the training loss gradients along the optimization trajectory influences the final generalization performance. The key technical ingredients in our proof combine stability analysis of gradient flow with uniform convergence via Rademacher complexity. Our bound recovers existing kernel regression bounds for overparameterized neural networks and shows the feature learning capability of neural networks compared to kernel methods. Numerical experiments on real-world datasets validate that our bounds correlate well with the true generalization gap.

UAVPairs: A Challenging Benchmark for Match Pair Retrieval of Large-scale UAV Images

The primary contribution of this paper is a challenging benchmark dataset, UAVPairs, and a training pipeline designed for match pair retrieval of large-scale UAV images. First, the UAVPairs dataset, comprising 21,622 high-resolution images across 30 diverse scenes, is constructed; the 3D points and tracks generated by SfM-based 3D reconstruction are employed to define the geometric similarity of image pairs, ensuring genuinely matchable image pairs are used for training. Second, to solve the problem of expensive mining cost for global hard negative mining, a batched nontrivial sample mining strategy is proposed, leveraging the geometric similarity and multi-scene structure of the UAVPairs to generate training samples as to accelerate training. Third, recognizing the limitation of pair-based losses, the ranked list loss is designed to improve the discrimination of image retrieval models, which optimizes the global similarity structure constructed from the positive set and negative set. Finally, the effectiveness of the UAVPairs dataset and training pipeline is validated through comprehensive experiments on three distinct large-scale UAV datasets. The experiment results demonstrate that models trained with the UAVPairs dataset and the ranked list loss achieve significantly improved retrieval accuracy compared to models trained on existing datasets or with conventional losses. Furthermore, these improvements translate to enhanced view graph connectivity and higher quality of reconstructed 3D models. The models trained by the proposed approach perform more robustly compared with hand-crafted global features, particularly in challenging repetitively textured scenes and weakly textured scenes. For match pair retrieval of large-scale UAV images, the trained image retrieval models offer an effective solution. The dataset would be made publicly available at https://github.com/json87/UAVPairs.

On the Role of Label Noise in the Feature Learning Process

Deep learning with noisy labels presents significant challenges. In this work, we theoretically characterize the role of label noise from a feature learning perspective. Specifically, we consider a signal-noise data distribution, where each sample comprises a label-dependent signal and label-independent noise, and rigorously analyze the training dynamics of a two-layer convolutional neural network under this data setup, along with the presence of label noise. Our analysis identifies two key stages. In Stage I, the model perfectly fits all the clean samples (i.e., samples without label noise) while ignoring the noisy ones (i.e., samples with noisy labels). During this stage, the model learns the signal from the clean samples, which generalizes well on unseen data. In Stage II, as the training loss converges, the gradient in the direction of noise surpasses that of the signal, leading to overfitting on noisy samples. Eventually, the model memorizes the noise present in the noisy samples and degrades its generalization ability. Furthermore, our analysis provides a theoretical basis for two widely used techniques for tackling label noise: early stopping and sample selection. Experiments on both synthetic and real-world setups validate our theory.

Fast Quiet-STaR: Thinking Without Thought Tokens

Large Language Models (LLMs) have achieved impressive performance across a range of natural language processing tasks. However, recent advances demonstrate that further gains particularly in complex reasoning tasks require more than merely scaling up model sizes or training data. One promising direction is to enable models to think during the reasoning process. Recently, Quiet STaR significantly improves reasoning by generating token-level thought traces, but incurs substantial inference overhead. In this work, we propose Fast Quiet STaR, a more efficient reasoning framework that preserves the benefits of token-level reasoning while reducing computational cost. Our method introduces a curriculum learning based training strategy that gradually reduces the number of thought tokens, enabling the model to internalize more abstract and concise reasoning processes. We further extend this approach to the standard Next Token Prediction (NTP) setting through reinforcement learning-based fine-tuning, resulting in Fast Quiet-STaR NTP, which eliminates the need for explicit thought token generation during inference. Experiments on four benchmark datasets with Mistral 7B and Qwen2.5 7B demonstrate that Fast Quiet-STaR consistently outperforms Quiet-STaR in terms of average accuracy under the same inference time budget. Notably, Fast Quiet-STaR NTP achieves an average accuracy improvement of 9\% on Mistral 7B and 5.7\% on Qwen2.5 7B, while maintaining the same inference latency. Our code will be available at https://github.com/huangwei200012/Fast-Quiet-STaR.

Scaling Diffusion Transformers Efficiently via $μ$P

Diffusion Transformers have emerged as the foundation for vision generative models, but their scalability is limited by the high cost of hyperparameter (HP) tuning at large scales. Recently, Maximal Update Parametrization ($\mu$P) was proposed for vanilla Transformers, which enables stable HP transfer from small to large language models, and dramatically reduces tuning costs. However, it remains unclear whether $\mu$P of vanilla Transformers extends to diffusion Transformers, which differ architecturally and objectively. In this work, we generalize standard $\mu$P to diffusion Transformers and validate its effectiveness through large-scale experiments. First, we rigorously prove that $\mu$P of mainstream diffusion Transformers, including DiT, U-ViT, PixArt-$\alpha$, and MMDiT, aligns with that of the vanilla Transformer, enabling the direct application of existing $\mu$P methodologies. Leveraging this result, we systematically demonstrate that DiT-$\mu$P enjoys robust HP transferability. Notably, DiT-XL-2-$\mu$P with transferred learning rate achieves 2.9 times faster convergence than the original DiT-XL-2. Finally, we validate the effectiveness of $\mu$P on text-to-image generation by scaling PixArt-$\alpha$ from 0.04B to 0.61B and MMDiT from 0.18B to 18B. In both cases, models under $\mu$P outperform their respective baselines while requiring small tuning cost, only 5.5% of one training run for PixArt-$\alpha$ and 3% of consumption by human experts for MMDiT-18B. These results establish $\mu$P as a principled and efficient framework for scaling diffusion Transformers.

MindOmni: Unleashing Reasoning Generation in Vision Language Models with RGPO

Recent text-to-image systems face limitations in handling multimodal inputs and complex reasoning tasks. We introduce MindOmni, a unified multimodal large language model that addresses these challenges by incorporating reasoning generation through reinforcement learning. MindOmni leverages a three-phase training strategy: i) design of a unified vision language model with a decoder-only diffusion module, ii) supervised fine-tuning with Chain-of-Thought (CoT) instruction data, and iii) our proposed Reasoning Generation Policy Optimization (RGPO) algorithm, utilizing multimodal feedback to effectively guide policy updates. Experimental results demonstrate that MindOmni outperforms existing models, achieving impressive performance on both understanding and generation benchmarks, meanwhile showcasing advanced fine-grained reasoning generation capabilities, especially with mathematical reasoning instruction. All codes will be made public at \href{https://github.com/EasonXiao-888/MindOmni}{https://github.com/EasonXiao-888/MindOmni}.

Bridge the Domains: Large Language Models Enhanced Cross-domain Sequential Recommendation

Cross-domain Sequential Recommendation (CDSR) aims to extract the preference from the user's historical interactions across various domains. Despite some progress in CDSR, two problems set the barrier for further advancements, i.e., overlap dilemma and transition complexity. The former means existing CDSR methods severely rely on users who own interactions on all domains to learn cross-domain item relationships, compromising the practicability. The latter refers to the difficulties in learning the complex transition patterns from the mixed behavior sequences. With powerful representation and reasoning abilities, Large Language Models (LLMs) are promising to address these two problems by bridging the items and capturing the user's preferences from a semantic view. Therefore, we propose an LLMs Enhanced Cross-domain Sequential Recommendation model (LLM4CDSR). To obtain the semantic item relationships, we first propose an LLM-based unified representation module to represent items. Then, a trainable adapter with contrastive regularization is designed to adapt the CDSR task. Besides, a hierarchical LLMs profiling module is designed to summarize user cross-domain preferences. Finally, these two modules are integrated into the proposed tri-thread framework to derive recommendations. We have conducted extensive experiments on three public cross-domain datasets, validating the effectiveness of LLM4CDSR. We have released the code online.

STNet: Prediction of Underwater Sound Speed Profiles with An Advanced Semi-Transformer Neural Network

Real time acquisition of accurate underwater sound velocity profile (SSP) is crucial for tracking the propagation trajectory of underwater acoustic signals, making it play a key role in ocean communication positioning. SSPs can be directly measured by instruments or inverted leveraging sound field data. Although measurement techniques provide a good accuracy, they are constrained by limited spatial coverage and require substantial time investment. The inversion method based on real-time measurement of acoustic field data improves operational efficiency, but loses the accuracy of SSP estimation and suffers from limited spatial applicability due to its stringent requirements for ocean observation infrastructure. To achieve accurate long-term ocean SSP estimation independent of real-time underwater data measurements, we propose a Semi-Transformer neural network (STNet) specifically designed for simulating sound velocity distribution patterns from the perspective of time series prediction. The proposed network architecture incorporates an optimized self-attention mechanism to effectively capture long-range temporal dependencies within historical sound velocity time-series data, facilitating accurate estimation of current SSPs or prediction of future SSPs. Through architectural optimization of the Transformer framework and integration of a time encoding mechanism, STNet could effectively improve computational efficiency. Comparative experimental results reveal that STNet outperforms state-of-the-art models in predictive accuracy and maintain good computational efficiency, demonstrating its potential for enabling accurate long-term full-depth ocean SSP forecasting.

A Multi-task Learning Balanced Attention Convolutional Neural Network Model for Few-shot Underwater Acoustic Target Recognition

Underwater acoustic target recognition (UATR) is of great significance for the protection of marine diversity and national defense security. The development of deep learning provides new opportunities for UATR, but faces challenges brought by the scarcity of reference samples and complex environmental interference. To address these issues, we proposes a multi-task balanced channel attention convolutional neural network (MT-BCA-CNN). The method integrates a channel attention mechanism with a multi-task learning strategy, constructing a shared feature extractor and multi-task classifiers to jointly optimize target classification and feature reconstruction tasks. The channel attention mechanism dynamically enhances discriminative acoustic features such as harmonic structures while suppressing noise. Experiments on the Watkins Marine Life Dataset demonstrate that MT-BCA-CNN achieves 97\% classification accuracy and 95\% $F1$-score in 27-class few-shot scenarios, significantly outperforming traditional CNN and ACNN models, as well as popular state-of-the-art UATR methods. Ablation studies confirm the synergistic benefits of multi-task learning and attention mechanisms, while a dynamic weighting adjustment strategy effectively balances task contributions. This work provides an efficient solution for few-shot underwater acoustic recognition, advancing research in marine bioacoustics and sonar signal processing.