Li Liu

Zhixin Wang, Jiaming Xu, Tianyi Zhou et al.

Effectively scaling Reinforcement Learning (RL) is crucial for enhancing the reasoning and alignment of Large Language Models. The massive data and complex execution flows inherent in these tasks require a distributed architecture capable of efficient scaling. However, to simplify programming and dependency management, mainstream frameworks often rely on a centralized architecture where a single node dispatches both control and data. This inherent coupling creates significant communication bottlenecks, severely limiting system scalability and efficiency. We present DISTFLOW, a novel, fully distributed RL framework that adopts a multi-controller paradigm. By decoupling data transmission from control dispatch, DISTFLOW establishes a parallelism-aware, decentralized Data Coordinator that leverages local caching, load balancing, and asynchronous double buffer to minimize communication overhead and mitigate straggler effects. For control logic, it introduces a task scheduler built upon Directed Acyclic Graph (DAG) that facilitates fine-grained, independent execution. Experimental results demonstrate that DISTFLOW achieves near-linear scalability up to 512 GPUs and delivers up to a 2.63x throughput improvement over state-of-the-art (SOTA) frameworks. The source code is available at: https://github.com/sii-research/siiRL.

Xinpeng Jiang, Haoyu Liu, Siqiang Luo et al.

The \emph{Single-Source Personalized PageRank} (SSPPR) query is central to graph OLAP, measuring the probability $π(s,t)$ that an $α$-decay random walk from node $s$ terminates at $t$. Despite decades of research, a significant gap remains between upper and lower bounds for its computational complexity. Existing upper bounds are $O\left(\min\left(\frac{\log(1/ε)}{ε^2}, \frac{\sqrt{m \log n}}ε, m \log \frac{1}ε\right)\right)$ for SSPPR-A and $O\left(\min\left(\frac{\log(1/n)}δ, \sqrt{m \log(n/δ)}, m \log \left(\frac{\log(n)}{mδ}\right)\right)\right)$ for SSPPR-R, with trivial lower bounds of $Ω(\min(n,1/ε))$ and $Ω(\min(n,1/δ))$. This work narrows or closes this gap. We improve the upper bounds for SSPPR-A and SSPPR-R to $O\left(\frac{1}{ε^2}\right)$ and $O\left(\min\left(\frac{\log(1/δ)}δ, m + n \log(n) \log \left(\frac{\log(n)}{mδ}\right)\right)\right)$, respectively, offering improvements by factors of $\log(1/ε)$ and $\log\left(\frac{\log(n)}{mδ}\right)$. On the lower bound side, we establish stronger results: $Ω(\min(m, 1/ε^2))$ for SSPPR-A and $Ω(\min(m, \frac{\log(1/δ)}δ))$ for SSPPR-R, strengthening theoretical foundations. Our upper and lower bounds for SSPPR-R coincide for graphs with $m \in Ω(n \log^2 n)$ and any threshold $δ, 1/δ\in O(\text{poly}(n))$, achieving theoretical optimality in most graph regimes. The SSPPR-A query attains partial optimality for large error thresholds, matching our new lower bound. This is the first optimal result for SSPPR queries. Our techniques generalize to the Single-Target Personalized PageRank (STPPR) query, improving its lower bound from $Ω(\min(n, 1/δ))$ to $Ω(\min(m, \frac{n}δ \log n))$, matching the upper bound and revealing its optimality.

Chenghua Gao, Min Li, Jianshuo Liu et al.

Video Moment Retrieval (VMR) aims to retrieve relevant moments of an untrimmed video corresponding to the query. While cross-modal interaction approaches have shown progress in filtering out query-irrelevant information in videos, they assume the precise alignment between the query semantics and the corresponding video moments, potentially overlooking the misunderstanding of the natural language semantics. To address this challenge, we propose a novel model called \textit{QD-VMR}, a query debiasing model with enhanced contextual understanding. Firstly, we leverage a Global Partial Aligner module via video clip and query features alignment and video-query contrastive learning to enhance the cross-modal understanding capabilities of the model. Subsequently, we employ a Query Debiasing Module to obtain debiased query features efficiently, and a Visual Enhancement module to refine the video features related to the query. Finally, we adopt the DETR structure to predict the possible target video moments. Through extensive evaluations of three benchmark datasets, QD-VMR achieves state-of-the-art performance, proving its potential to improve the accuracy of VMR. Further analytical experiments demonstrate the effectiveness of our proposed module. Our code will be released to facilitate future research.

Bowen Peng, Li Liu, Tianpeng Liu et al.

Transfer-based targeted adversarial attacks against black-box deep neural networks (DNNs) have been proven to be significantly more challenging than untargeted ones. The impressive transferability of current SOTA, the generative methods, comes at the cost of requiring massive amounts of additional data and time-consuming training for each targeted label. This results in limited efficiency and flexibility, significantly hindering their deployment in practical applications. In this paper, we offer a self-universal perspective that unveils the great yet underexplored potential of input transformations in pursuing this goal. Specifically, transformations universalize gradient-based attacks with intrinsic but overlooked semantics inherent within individual images, exhibiting similar scalability and comparable results to time-consuming learning over massive additional data from diverse classes. We also contribute a surprising empirical insight that one of the most fundamental transformations, simple image scaling, is highly effective, scalable, sufficient, and necessary in enhancing targeted transferability. We further augment simple scaling with orthogonal transformations and block-wise applicability, resulting in the Simple, faSt, Self-universal yet Strong Scale Transformation (S$^4$ST) for self-universal TTA. On the ImageNet-Compatible benchmark dataset, our method achieves a 19.8% improvement in the average targeted transfer success rate against various challenging victim models over existing SOTA transformation methods while only consuming 36% time for attacking. It also outperforms resource-intensive attacks by a large margin in various challenging settings.

Zichong Li, Liming Liu, Chen Liang et al. · gatech

The choice of optimizer significantly impacts the training efficiency and computational costs of large language models (LLMs). Recently, the Muon optimizer has demonstrated promising results by orthogonalizing parameter updates, improving optimization geometry through better conditioning. Despite Muon's emergence as a candidate successor to Adam, the potential for jointly leveraging their strengths has not been systematically explored. In this work, we bridge this gap by proposing NorMuon (Neuron-wise Normalized Muon), an optimizer that synergistically combines orthogonalization with neuron-level adaptive learning rates. Our analysis reveals that while Muon effectively reduces condition numbers, the resulting updates exhibit highly non-uniform neuron norms, causing certain neurons to dominate the optimization process. NorMuon addresses this imbalance by maintaining second-order momentum statistics for each neuron and applying row-wise normalization after orthogonalization, ensuring balanced parameter utilization while preserving Muon's conditioning benefits. To enable practical deployment at scale, we develop an efficient distributed implementation under the FSDP2 framework that strategically distributes orthogonalization computations across devices. Experiments across multiple model scales demonstrate that NorMuon consistently outperforms both Adam and Muon, achieving 21.74% better training efficiency than Adam and 11.31% improvement over Muon on 1.1 B pretraining setting, while maintaining a comparable memory footprint to Muon. Our findings suggest that orthogonalization and adaptive learning rates are complementary rather than competing approaches, opening new avenues for optimizer design in large-scale deep learning.

Zhengyang Ji, Shang Gao, Li Liu et al.

Biomedical visual question answering (VQA) has been widely studied and has demonstrated significant application value and potential in fields such as assistive medical diagnosis. Despite their success, current biomedical VQA models perform multimodal information interaction only at the model level within large language models (LLMs), leading to suboptimal multimodal semantic alignment when dealing with complex tasks. To address this issue, we propose BioD2C: a novel Dual-level Semantic Consistency Constraint Framework for Biomedical VQA, which achieves dual-level semantic interaction alignment at both the model and feature levels, enabling the model to adaptively learn visual features based on the question. Specifically, we firstly integrate textual features into visual features via an image-text fusion mechanism as feature-level semantic interaction, obtaining visual features conditioned on the given text; and then introduce a text-queue-based cross-modal soft semantic loss function to further align the image semantics with the question semantics. Specifically, in this work, we establish a new dataset, BioVGQ, to address inherent biases in prior datasets by filtering manually-altered images and aligning question-answer pairs with multimodal context, and train our model on this dataset. Extensive experimental results demonstrate that BioD2C achieves state-of-the-art (SOTA) performance across multiple downstream datasets, showcasing its robustness, generalizability, and potential to advance biomedical VQA research.

Li Liu, Xianghao Zhan, Rumeng Wu et al.

Electronic nose has been proven to be effective in alternative herbal medicine classification, but due to the nature of supervised learning, previous research heavily relies on the labelled training data, which are time-costly and labor-intensive to collect. To alleviate the critical dependency on the training data in real-world applications, this study aims to improve classification accuracy via data augmentation strategies. The effectiveness of five data augmentation strategies under different training data inadequacy are investigated in two scenarios: the noise-free scenario where different availabilities of unlabelled data were considered, and the noisy scenario where different levels of Gaussian noises and translational shifts were added to represent sensor drifts. The five augmentation strategies, namely noise-adding data augmentation, semi-supervised learning, classifier-based online learning, Inductive Conformal Prediction (ICP) online learning and our novel ensemble ICP online learning proposed in this study, are experimented and compared against supervised learning baseline, with Linear Discriminant Analysis (LDA) and Support Vector Machine (SVM) as the classifiers. Our novel strategy, ensemble ICP online learning, outperforms the others by showing non-decreasing classification accuracy on all tasks and a significant improvement on most simulated tasks (25out of 36 tasks,p<=0.05). Furthermore, this study provides a systematic analysis of different augmentation strategies. It shows at least one strategy significantly improved the classification accuracy with LDA (p<=0.05) and non-decreasing classification accuracy with SVM in each task. In particular, our proposed strategy demonstrated both effectiveness and robustness in boosting the classification model generalizability, which can be employed in other machine learning applications.