Pei Zhang

Xiangpeng Wei, Haoran Wei, Huan Lin et al.

Large language models (LLMs) demonstrate remarkable ability to comprehend, reason, and generate following nature language instructions. However, the development of LLMs has been primarily focused on high-resource languages, such as English, thereby limiting their applicability and research in other languages. Consequently, we present PolyLM, a multilingual LLM trained on 640 billion (B) tokens, avaliable in two model sizes: 1.7B and 13B. To enhance its multilingual capabilities, we 1) integrate bilingual data into training data; and 2) adopt a curriculum learning strategy that increases the proportion of non-English data from 30% in the first stage to 60% in the final stage during pre-training. Further, we propose a multilingual self-instruct method which automatically generates 132.7K diverse multilingual instructions for model fine-tuning. To assess the model's performance, we collect several existing multilingual tasks, including multilingual understanding, question answering, generation, and translation. Extensive experiments show that PolyLM surpasses other open-source models such as LLaMA and BLOOM on multilingual tasks while maintaining comparable performance in English. Our models, alone with the instruction data and multilingual benchmark, are available at: \url{https://modelscope.cn/models/damo/nlp_polylm_13b_text_generation}.

Jingcan Duan, Pei Zhang, Siwei Wang et al.

Graph anomaly detection (GAD) has attracted increasing attention in machine learning and data mining. Recent works have mainly focused on how to capture richer information to improve the quality of node embeddings for GAD. Despite their significant advances in detection performance, there is still a relative dearth of research on the properties of the task. GAD aims to discern the anomalies that deviate from most nodes. However, the model is prone to learn the pattern of normal samples which make up the majority of samples. Meanwhile, anomalies can be easily detected when their behaviors differ from normality. Therefore, the performance can be further improved by enhancing the ability to learn the normal pattern. To this end, we propose a normality learning-based GAD framework via multi-scale contrastive learning networks (NLGAD for abbreviation). Specifically, we first initialize the model with the contrastive networks on different scales. To provide sufficient and reliable normal nodes for normality learning, we design an effective hybrid strategy for normality selection. Finally, the model is refined with the only input of reliable normal nodes and learns a more accurate estimate of normality so that anomalous nodes can be more easily distinguished. Eventually, extensive experiments on six benchmark graph datasets demonstrate the effectiveness of our normality learning-based scheme on GAD. Notably, the proposed algorithm improves the detection performance (up to 5.89% AUC gain) compared with the state-of-the-art methods. The source code is released at https://github.com/FelixDJC/NLGAD.

Shuaiwen Leon Song, Bonnie Kruft, Minjia Zhang et al. · microsoft-research

In the upcoming decade, deep learning may revolutionize the natural sciences, enhancing our capacity to model and predict natural occurrences. This could herald a new era of scientific exploration, bringing significant advancements across sectors from drug development to renewable energy. To answer this call, we present DeepSpeed4Science initiative (deepspeed4science.ai) which aims to build unique capabilities through AI system technology innovations to help domain experts to unlock today's biggest science mysteries. By leveraging DeepSpeed's current technology pillars (training, inference and compression) as base technology enablers, DeepSpeed4Science will create a new set of AI system technologies tailored for accelerating scientific discoveries by addressing their unique complexity beyond the common technical approaches used for accelerating generic large language models (LLMs). In this paper, we showcase the early progress we made with DeepSpeed4Science in addressing two of the critical system challenges in structural biology research.

An Yang, Baosong Yang, Binyuan Hui et al.

This report introduces the Qwen2 series, the latest addition to our large language models and large multimodal models. We release a comprehensive suite of foundational and instruction-tuned language models, encompassing a parameter range from 0.5 to 72 billion, featuring dense models and a Mixture-of-Experts model. Qwen2 surpasses most prior open-weight models, including its predecessor Qwen1.5, and exhibits competitive performance relative to proprietary models across diverse benchmarks on language understanding, generation, multilingual proficiency, coding, mathematics, and reasoning. The flagship model, Qwen2-72B, showcases remarkable performance: 84.2 on MMLU, 37.9 on GPQA, 64.6 on HumanEval, 89.5 on GSM8K, and 82.4 on BBH as a base language model. The instruction-tuned variant, Qwen2-72B-Instruct, attains 9.1 on MT-Bench, 48.1 on Arena-Hard, and 35.7 on LiveCodeBench. Moreover, Qwen2 demonstrates robust multilingual capabilities, proficient in approximately 30 languages, spanning English, Chinese, Spanish, French, German, Arabic, Russian, Korean, Japanese, Thai, Vietnamese, and more, underscoring its versatility and global reach. To foster community innovation and accessibility, we have made the Qwen2 model weights openly available on Hugging Face and ModelScope, and the supplementary materials including example code on GitHub. These platforms also include resources for quantization, fine-tuning, and deployment, facilitating a wide range of applications and research endeavors.

Yiming Wang, Zhuosheng Zhang, Pei Zhang et al.

Neural-symbolic methods have demonstrated efficiency in enhancing the reasoning abilities of large language models (LLMs). However, existing methods mainly rely on syntactically mapping natural languages to complete formal languages like Python and SQL. Those methods require that reasoning tasks be convertible into programs, which cater to the computer execution mindset and deviate from human reasoning habits. To broaden symbolic methods' applicability and adaptability in the real world, we propose the Meta-Reasoning from a linguistic perspective. This method empowers LLMs to deconstruct reasoning-independent semantic information into generic symbolic representations, thereby efficiently capturing more generalized reasoning knowledge. We conduct extensive experiments on more than ten datasets encompassing conventional reasoning tasks like arithmetic, symbolic, and logical reasoning, and the more complex interactive reasoning tasks like theory-of-mind reasoning. Experimental results demonstrate that Meta-Reasoning significantly enhances in-context reasoning accuracy, learning efficiency, out-of-domain generalization, and output stability compared to the Chain-of-Thought technique. Code and data are publicly available at \url{https://github.com/Alsace08/Meta-Reasoning}.

Pei Zhang, Logan Kearney, Debsindhu Bhowmik et al.

Transformer-based large language models have remarkable potential to accelerate design optimization for applications such as drug development and materials discovery. Self-supervised pretraining of transformer models requires large-scale datasets, which are often sparsely populated in topical areas such as polymer science. State-of-the-art approaches for polymers conduct data augmentation to generate additional samples but unavoidably incurs extra computational costs. In contrast, large-scale open-source datasets are available for small molecules and provide a potential solution to data scarcity through transfer learning. In this work, we show that using transformers pretrained on small molecules and fine-tuned on polymer properties achieve comparable accuracy to those trained on augmented polymer datasets for a series of benchmark prediction tasks.

Jong Youl Choi, Pei Zhang, Kshitij Mehta et al.

Graph Convolutional Neural Network (GCNN) is a popular class of deep learning (DL) models in material science to predict material properties from the graph representation of molecular structures. Training an accurate and comprehensive GCNN surrogate for molecular design requires large-scale graph datasets and is usually a time-consuming process. Recent advances in GPUs and distributed computing open a path to reduce the computational cost for GCNN training effectively. However, efficient utilization of high performance computing (HPC) resources for training requires simultaneously optimizing large-scale data management and scalable stochastic batched optimization techniques. In this work, we focus on building GCNN models on HPC systems to predict material properties of millions of molecules. We use HydraGNN, our in-house library for large-scale GCNN training, leveraging distributed data parallelism in PyTorch. We use ADIOS, a high-performance data management framework for efficient storage and reading of large molecular graph data. We perform parallel training on two open-source large-scale graph datasets to build a GCNN predictor for an important quantum property known as the HOMO-LUMO gap. We measure the scalability, accuracy, and convergence of our approach on two DOE supercomputers: the Summit supercomputer at the Oak Ridge Leadership Computing Facility (OLCF) and the Perlmutter system at the National Energy Research Scientific Computing Center (NERSC). We present our experimental results with HydraGNN showing i) reduction of data loading time up to 4.2 times compared with a conventional method and ii) linear scaling performance for training up to 1,024 GPUs on both Summit and Perlmutter.

Jingcan Duan, Siwei Wang, Pei Zhang et al.

Graph anomaly detection (GAD) is a vital task in graph-based machine learning and has been widely applied in many real-world applications. The primary goal of GAD is to capture anomalous nodes from graph datasets, which evidently deviate from the majority of nodes. Recent methods have paid attention to various scales of contrastive strategies for GAD, i.e., node-subgraph and node-node contrasts. However, they neglect the subgraph-subgraph comparison information which the normal and abnormal subgraph pairs behave differently in terms of embeddings and structures in GAD, resulting in sub-optimal task performance. In this paper, we fulfill the above idea in the proposed multi-view multi-scale contrastive learning framework with subgraph-subgraph contrast for the first practice. To be specific, we regard the original input graph as the first view and generate the second view by graph augmentation with edge modifications. With the guidance of maximizing the similarity of the subgraph pairs, the proposed subgraph-subgraph contrast contributes to more robust subgraph embeddings despite of the structure variation. Moreover, the introduced subgraph-subgraph contrast cooperates well with the widely-adopted node-subgraph and node-node contrastive counterparts for mutual GAD performance promotions. Besides, we also conduct sufficient experiments to investigate the impact of different graph augmentation approaches on detection performance. The comprehensive experimental results well demonstrate the superiority of our method compared with the state-of-the-art approaches and the effectiveness of the multi-view subgraph pair contrastive strategy for the GAD task.

Yi Wen, Suyuan Liu, Xinhang Wan et al.

Anchor-based multi-view graph clustering (AMVGC) has received abundant attention owing to its high efficiency and the capability to capture complementary structural information across multiple views. Intuitively, a high-quality anchor graph plays an essential role in the success of AMVGC. However, the existing AMVGC methods only consider single-structure information, i.e., local or global structure, which provides insufficient information for the learning task. To be specific, the over-scattered global structure leads to learned anchors failing to depict the cluster partition well. In contrast, the local structure with an improper similarity measure results in potentially inaccurate anchor assignment, ultimately leading to sub-optimal clustering performance. To tackle the issue, we propose a novel anchor-based multi-view graph clustering framework termed Efficient Multi-View Graph Clustering with Local and Global Structure Preservation (EMVGC-LG). Specifically, a unified framework with a theoretical guarantee is designed to capture local and global information. Besides, EMVGC-LG jointly optimizes anchor construction and graph learning to enhance the clustering quality. In addition, EMVGC-LG inherits the linear complexity of existing AMVGC methods respecting the sample number, which is time-economical and scales well with the data size. Extensive experiments demonstrate the effectiveness and efficiency of our proposed method.

Qiyuan Ou, Siwei Wang, Pei Zhang et al.

Multi-view clustering has attracted growing attention owing to its capabilities of aggregating information from various sources and its promising horizons in public affairs. Up till now, many advanced approaches have been proposed in recent literature. However, there are several ongoing difficulties to be tackled. One common dilemma occurs while attempting to align the features of different views. {Moreover, due to the fact that many existing multi-view clustering algorithms stem from spectral clustering, this results to cubic time complexity w.r.t. the number of dataset. However, we propose Anchor-based Multi-view Subspace Clustering with Hierarchical Feature Descent(MVSC-HFD) to tackle the discrepancy among views through hierarchical feature descent and project to a common subspace( STAGE 1), which reveals dependency of different views. We further reduce the computational complexity to linear time cost through a unified sampling strategy in the common subspace( STAGE 2), followed by anchor-based subspace clustering to learn the bipartite graph collectively( STAGE 3). }Extensive experimental results on public benchmark datasets demonstrate that our proposed model consistently outperforms the state-of-the-art techniques.

Xian Shi, Xiong Wang, Zhifang Guo et al.

In this report, we introduce Qwen3-ASR family, which includes two powerful all-in-one speech recognition models and a novel non-autoregressive speech forced alignment model. Qwen3-ASR-1.7B and Qwen3-ASR-0.6B are ASR models that support language identification and ASR for 52 languages and dialects. Both of them leverage large-scale speech training data and the strong audio understanding ability of their foundation model Qwen3-Omni. We conduct comprehensive internal evaluation besides the open-sourced benchmarks as ASR models might differ little on open-sourced benchmark scores but exhibit significant quality differences in real-world scenarios. The experiments reveal that the 1.7B version achieves SOTA performance among open-sourced ASR models and is competitive with the strongest proprietary APIs while the 0.6B version offers the best accuracy-efficiency trade-off. Qwen3-ASR-0.6B can achieve an average TTFT as low as 92ms and transcribe 2000 seconds speech in 1 second at a concurrency of 128. Qwen3-ForcedAligner-0.6B is an LLM based NAR timestamp predictor that is able to align text-speech pairs in 11 languages. Timestamp accuracy experiments show that the proposed model outperforms the three strongest force alignment models and takes more advantages in efficiency and versatility. To further accelerate the community research of ASR and audio understanding, we release these models under the Apache 2.0 license.

Joonhee Lee, Hyunseung Shin, Hyunmi Kim et al.

Reasoning-based robotic policies using large language and vision-language models achieve strong semantic planning capabilities but mostly suffer from a high inference latency that limits practical real-time deployment. In this work, we observe that robotic reasoning workloads contain substantial temporal redundancy, where consecutive observations frequently produce identical actions and subgoals. Based on this insight, we present REIS, a human cognition inspired robotic decision-making framework that minimizes unnecessary reasoning while preserving semantic adaptability. REIS combines lightweight scene gating, KV-steered affordance routing, and deliberative reasoning to accelerate robotic control under embodied constraints. Experiments on ALFRED, and real-world robotic tasks demonstrate that REIS significantly suppresses reasoning overhead while maintaining competitive task performance.

Jun Ma, Zhenye Yang, Ruichen Zhou et al.

Driver gaze estimation serves as a fundamental metric for evaluating driver attentiveness in modern monitoring systems. Beyond being vulnerable to sudden lighting changes and sensor noise, spatial-domain models struggle to disentangle authentic gaze cues from irrelevant visual attributes. In this paper, we propose LISA, a \textbf{L}anguage-guided \textbf{I}nterference-aware \textbf{S}patial-Frequency \textbf{A}ttention framework that combines frequency-domain priors with vision-language knowledge. Observing that the amplitude spectrum remains relatively stable even under spatial perturbations, we design a dual-domain fusion mechanism. It integrates stable low-frequency semantics into high-frequency details, employing spatial attention to precisely target ocular regions. To reduce semantic ambiguity, we also introduce a training-time disentanglement strategy. Using a frozen CLIP encoder and orthogonal regularization, we explicitly separate gaze features from appearance interference. Experiments on two benchmarks show that LISA achieves state-of-the-art performance, with significantly improved robustness against occlusions and lighting variations. The code repository is available at https://github.com/Mason-bupt/LISA.

Pei Zhang, Baosong Yang, Haoran Wei et al.

Neural machine translation (NMT) is often criticized for failures that happen without awareness. The lack of competency awareness makes NMT untrustworthy. This is in sharp contrast to human translators who give feedback or conduct further investigations whenever they are in doubt about predictions. To fill this gap, we propose a novel competency-aware NMT by extending conventional NMT with a self-estimator, offering abilities to translate a source sentence and estimate its competency. The self-estimator encodes the information of the decoding procedure and then examines whether it can reconstruct the original semantics of the source sentence. Experimental results on four translation tasks demonstrate that the proposed method not only carries out translation tasks intact but also delivers outstanding performance on quality estimation. Without depending on any reference or annotated data typically required by state-of-the-art metric and quality estimation methods, our model yields an even higher correlation with human quality judgments than a variety of aforementioned methods, such as BLEURT, COMET, and BERTScore. Quantitative and qualitative analyses show better robustness of competency awareness in our model.

Chuanlin Zhang, Junkang Feng, Chenggang Cui et al.

With the growing adoption of electric vehicles (EVs), understanding user charging behavior has become critical for grid stability and transportation planning. This study investigates the behavioral heterogeneity of EV taxi drivers by analyzing the interaction between psychological traits and situational triggers within dynamic travel contexts. Leveraging large language models (LLMs) as a core simulation tool, a novel framework with statistical enhancement is developed to replicate and analyze the charging behaviors of taxi drivers. LLMs simulate personalized decision-making processes by leveraging natural language reasoning and role-playing capabilities, accounting for factors such as time sensitivity, price awareness, and range anxiety. Simulation results indicate that the framework reliably reproduces real-world charging behaviors across multiple urban environments. his fidelity arises from integrating statistical priors into the reasoning process, allowing the model to anchor its decisions in empirical behavioral patterns. Further analysis highlights the joint influence of environmental and psychological variables on charging decisions and reveals the heterogeneity of different user groups. The findings provide new insights into EV user behavior, offering a foundation for optimizing charging infrastructure, informing energy policy, and advancing the integration of EV behavioral models into smart transportation and energy management systems.

Hangrui Hu, Xinfa Zhu, Ting He et al.

In this report, we present the Qwen3-TTS series, a family of advanced multilingual, controllable, robust, and streaming text-to-speech models. Qwen3-TTS supports state-of-the-art 3-second voice cloning and description-based control, allowing both the creation of entirely novel voices and fine-grained manipulation over the output speech. Trained on over 5 million hours of speech data spanning 10 languages, Qwen3-TTS adopts a dual-track LM architecture for real-time synthesis, coupled with two speech tokenizers: 1) Qwen-TTS-Tokenizer-25Hz is a single-codebook codec emphasizing semantic content, which offers seamlessly integration with Qwen-Audio and enables streaming waveform reconstruction via a block-wise DiT. 2) Qwen-TTS-Tokenizer-12Hz achieves extreme bitrate reduction and ultra-low-latency streaming, enabling immediate first-packet emission ($97\,\mathrm{ms}$) through its 12.5 Hz, 16-layer multi-codebook design and a lightweight causal ConvNet. Extensive experiments indicate state-of-the-art performance across diverse objective and subjective benchmark (e.g., TTS multilingual test set, InstructTTSEval, and our long speech test set). To facilitate community research and development, we release both tokenizers and models under the Apache 2.0 license.

Jin Xu, Zhifang Guo, Hangrui Hu et al. · pku

We present Qwen3-Omni, a single multimodal model that, for the first time, maintains state-of-the-art performance across text, image, audio, and video without any degradation relative to single-modal counterparts. Qwen3-Omni matches the performance of same-sized single-modal models within the Qwen series and excels particularly on audio tasks. Across 36 audio and audio-visual benchmarks, Qwen3-Omni achieves open-source SOTA on 32 benchmarks and overall SOTA on 22, outperforming strong closed-source models such as Gemini-2.5-Pro, Seed-ASR, and GPT-4o-Transcribe. Qwen3-Omni adopts a Thinker-Talker MoE architecture that unifies perception and generation across text, images, audio, and video, yielding fluent text and natural real-time speech. It supports text interaction in 119 languages, speech understanding in 19 languages, and speech generation in 10 languages. To reduce first-packet latency in streaming synthesis, Talker autoregressively predicts discrete speech codecs using a multi-codebook scheme. Leveraging the representational capacity of these codebooks, we replace computationally intensive block-wise diffusion with a lightweight causal ConvNet, enabling streaming from the first codec frame. In cold-start settings, Qwen3-Omni achieves a theoretical end-to-end first-packet latency of 234 ms. To further strengthen multimodal reasoning, we introduce a Thinking model that explicitly reasons over inputs from any modality. Since the research community currently lacks a general-purpose audio captioning model, we fine-tuned Qwen3-Omni-30B-A3B to obtain Qwen3-Omni-30B-A3B-Captioner, which produces detailed, low-hallucination captions for arbitrary audio inputs. Qwen3-Omni-30B-A3B, Qwen3-Omni-30B-A3B-Thinking, and Qwen3-Omni-30B-A3B-Captioner are publicly released under the Apache 2.0 license.

Pei Zhang, Chutian Wang, Edmund Y. Lam

Bio-inspired neuromorphic cameras asynchronously record pixel brightness changes and generate sparse event streams. They can capture dynamic scenes with little motion blur and more details in extreme illumination conditions. Due to the multidimensional address-event structure, most existing vision algorithms cannot properly handle asynchronous event streams. While several event representations and processing methods have been developed to address such an issue, they are typically driven by a large number of events, leading to substantial overheads in runtime and memory. In this paper, we propose a new graph representation of the event data and couple it with a Graph Transformer to perform accurate neuromorphic classification. Extensive experiments show that our approach leads to better results and excels at the challenging realistic situations where only a small number of events and limited computational resources are available, paving the way for neuromorphic applications embedded into mobile facilities.

Alam Noor, Kai Li, Eduardo Tovar et al.

Recognizing unauthorized Unmanned Aerial Vehicles (UAVs) within designated no-fly zones throughout the day and night is of paramount importance, where the unauthorized UAVs pose a substantial threat to both civil and military aviation safety. However, recognizing UAVs day and night with dual-vision cameras is nontrivial, since red-green-blue (RGB) images suffer from a low detection rate under an insufficient light condition, such as on cloudy or stormy days, while black-and-white infrared (IR) images struggle to capture UAVs that overlap with the background at night. In this paper, we propose a new optical flow-assisted graph-pooling residual network (OF-GPRN), which significantly enhances the UAV detection rate in day and night dual visions. The proposed OF-GPRN develops a new optical fusion to remove superfluous backgrounds, which improves RGB/IR imaging clarity. Furthermore, OF-GPRN extends optical fusion by incorporating a graph residual split attention network and a feature pyramid, which refines the perception of UAVs, leading to a higher success rate in UAV detection. A comprehensive performance evaluation is conducted using a benchmark UAV catch dataset. The results indicate that the proposed OF-GPRN elevates the UAV mean average precision (mAP) detection rate to 87.8%, marking a 17.9% advancement compared to the residual graph neural network (ResGCN)-based approach.

An Yang, Anfeng Li, Baosong Yang et al. · tsinghua

In this work, we present Qwen3, the latest version of the Qwen model family. Qwen3 comprises a series of large language models (LLMs) designed to advance performance, efficiency, and multilingual capabilities. The Qwen3 series includes models of both dense and Mixture-of-Expert (MoE) architectures, with parameter scales ranging from 0.6 to 235 billion. A key innovation in Qwen3 is the integration of thinking mode (for complex, multi-step reasoning) and non-thinking mode (for rapid, context-driven responses) into a unified framework. This eliminates the need to switch between different models--such as chat-optimized models (e.g., GPT-4o) and dedicated reasoning models (e.g., QwQ-32B)--and enables dynamic mode switching based on user queries or chat templates. Meanwhile, Qwen3 introduces a thinking budget mechanism, allowing users to allocate computational resources adaptively during inference, thereby balancing latency and performance based on task complexity. Moreover, by leveraging the knowledge from the flagship models, we significantly reduce the computational resources required to build smaller-scale models, while ensuring their highly competitive performance. Empirical evaluations demonstrate that Qwen3 achieves state-of-the-art results across diverse benchmarks, including tasks in code generation, mathematical reasoning, agent tasks, etc., competitive against larger MoE models and proprietary models. Compared to its predecessor Qwen2.5, Qwen3 expands multilingual support from 29 to 119 languages and dialects, enhancing global accessibility through improved cross-lingual understanding and generation capabilities. To facilitate reproducibility and community-driven research and development, all Qwen3 models are publicly accessible under Apache 2.0.

Qwen, An Yang, Baosong Yang et al.

In this report, we introduce Qwen2.5, a comprehensive series of large language models (LLMs) designed to meet diverse needs. Compared to previous iterations, Qwen 2.5 has been significantly improved during both the pre-training and post-training stages. In terms of pre-training, we have scaled the high-quality pre-training datasets from the previous 7 trillion tokens to 18 trillion tokens. This provides a strong foundation for common sense, expert knowledge, and reasoning capabilities. In terms of post-training, we implement intricate supervised finetuning with over 1 million samples, as well as multistage reinforcement learning. Post-training techniques enhance human preference, and notably improve long text generation, structural data analysis, and instruction following. To handle diverse and varied use cases effectively, we present Qwen2.5 LLM series in rich sizes. Open-weight offerings include base and instruction-tuned models, with quantized versions available. In addition, for hosted solutions, the proprietary models currently include two mixture-of-experts (MoE) variants: Qwen2.5-Turbo and Qwen2.5-Plus, both available from Alibaba Cloud Model Studio. Qwen2.5 has demonstrated top-tier performance on a wide range of benchmarks evaluating language understanding, reasoning, mathematics, coding, human preference alignment, etc. Specifically, the open-weight flagship Qwen2.5-72B-Instruct outperforms a number of open and proprietary models and demonstrates competitive performance to the state-of-the-art open-weight model, Llama-3-405B-Instruct, which is around 5 times larger. Qwen2.5-Turbo and Qwen2.5-Plus offer superior cost-effectiveness while performing competitively against GPT-4o-mini and GPT-4o respectively. Additionally, as the foundation, Qwen2.5 models have been instrumental in training specialized models such as Qwen2.5-Math, Qwen2.5-Coder, QwQ, and multimodal models.

Pei Zhang, Shijie Lin, Zhou Ge et al.

Quasi-bimodal objects, such as text, road signs, and barcodes, play a basic yet vital role in daily visual communication. By boiling these down to clear silhouettes, binarization uses a minimal language to convey essential vision cues for maximum downstream efficiency. The catch is that frame-based imaging often struggles on mobile platforms like drones, self-driving cars, and underwater vehicles. In these dynamic scenes, rapid motion and harsh lighting can make it blind, causing severe motion blur and erasing crucial details. To overcome the limits, neuromorphic vision via event cameras, featuring microsecond-level temporal resolution and high dynamic range, steps in as a natural solution. Building upon this event-driven sensing paradigm, we introduce a simple yet effective dual-modal approach that harnesses the synergy between frames and events to achieve real-time, high-frame-rate binarization on CPU-only devices. Extensive evaluations present that it earns competitive performance against leading techniques in reducing motion blur, while delivering impressive improvements under challenging illumination. Besides, our asynchronous workflow bypasses event scarcity that breaks traditional time-binning reconstruction, maintaining clear target shapes even at extreme kilohertz frame rates. Its binary results further serve as reliable representations that facilitate a range of downstream tasks. This work paves the way towards lightweight perception and interaction in embodied intelligence on resource-constrained edge platforms.

Wenhao Liu, Zhenyi Lu, Xinyu Hu et al.

High-quality math datasets are crucial for advancing the reasoning abilities of large language models (LLMs). However, existing datasets often suffer from three key issues: outdated and insufficient challenging content, neglecting human-like reasoning, and limited reliability due to single-LLM generation. To address these, we introduce STORM-BORN, an ultra-challenging dataset of mathematical derivations sourced from cutting-edge academic papers, which includes dense human-like approximations and heuristic cues. To ensure the reliability and quality, we propose a novel human-in-the-loop, multi-agent data generation framework, integrating reasoning-dense filters, multi-agent collaboration, and human mathematicians' evaluations. We curated a set of 2,000 synthetic samples and deliberately selected the 100 most difficult problems. Even most advanced models like GPT-o1 solved fewer than 5% of them. Fine-tuning on STORM-BORN boosts accuracy by 7.84% (LLaMA3-8B) and 9.12% (Qwen2.5-7B). As AI approaches mathematician-level reasoning, STORM-BORN provides both a high-difficulty benchmark and a human-like reasoning training resource. Our code and dataset are publicly available at https://github.com/lwhere/STORM-BORN.

Hunor Csala, Sebastian De Pascuale, Paul Laiu et al.

Accurate modeling of scrape-off layer (SOL) and divertor-edge dynamics is vital for designing plasma-facing components in fusion devices. High-fidelity edge fluid/neutral codes such as SOLPS-ITER capture SOL physics with high accuracy, but their computational cost limits broad parameter scans and long transient studies. We present transformer-based, autoregressive surrogates for efficient prediction of 2D, time-dependent plasma edge state fields. Trained on SOLPS-ITER spatiotemporal data, the surrogates forecast electron temperature, electron density, and radiated power over extended horizons. We evaluate model variants trained with increasing autoregressive horizons (1-100 steps) on short- and long-horizon prediction tasks. Longer-horizon training systematically improves rollout stability and mitigates error accumulation, enabling stable predictions over hundreds to thousands of steps and reproducing key dynamical features such as the motion of high-radiation regions. Measured end-to-end wall-clock times show the surrogate is orders of magnitude faster than SOLPS-ITER, enabling rapid parameter exploration. Prediction accuracy degrades when the surrogate enters physical regimes not represented in the training dataset, motivating future work on data enrichment and physics-informed constraints. Overall, this approach provides a fast, accurate surrogate for computationally intensive plasma edge simulations, supporting rapid scenario exploration, control-oriented studies, and progress toward real-time applications in fusion devices.

Xinyu Zhang, Pei Zhang, Shuang Luo et al.

Cultural competence, defined as the ability to understand and adapt to multicultural contexts, is increasingly vital for large language models (LLMs) in global environments. While several cultural benchmarks exist to assess LLMs' cultural competence, current evaluations suffer from fragmented taxonomies, domain specificity, and heavy reliance on manual data annotation. To address these limitations, we introduce CultureSynth, a novel framework comprising (1) a comprehensive hierarchical multilingual cultural taxonomy covering 12 primary and 130 secondary topics, and (2) a Retrieval-Augmented Generation (RAG)-based methodology leveraging factual knowledge to synthesize culturally relevant question-answer pairs. The CultureSynth-7 synthetic benchmark contains 19,360 entries and 4,149 manually verified entries across 7 languages. Evaluation of 14 prevalent LLMs of different sizes reveals clear performance stratification led by ChatGPT-4o-Latest and Qwen2.5-72B-Instruct. The results demonstrate that a 3B-parameter threshold is necessary for achieving basic cultural competence, models display varying architectural biases in knowledge processing, and significant geographic disparities exist across models. We believe that CultureSynth offers a scalable framework for developing culturally aware AI systems while reducing reliance on manual annotation\footnote{Benchmark is available at https://github.com/Eyr3/CultureSynth.}.

Fei Zhang, Pei Zhang, Baosong Yang et al.

This paper presents the first study on adapting the visual in-context learning (V-ICL) paradigm to optical character recognition tasks, specifically focusing on text removal and segmentation. Most existing V-ICL generalists employ a reasoning-as-reconstruction approach: they turn to using a straightforward image-label compositor as the prompt and query input, and then masking the query label to generate the desired output. This direct prompt confines the model to a challenging single-step reasoning process. To address this, we propose a task-chaining compositor in the form of image-removal-segmentation, providing an enhanced prompt that elicits reasoning with enriched intermediates. Additionally, we introduce context-aware aggregation, integrating the chained prompt pattern into the latent query representation, thereby strengthening the model's in-context reasoning. We also consider the issue of visual heterogeneity, which complicates the selection of homogeneous demonstrations in text recognition. Accordingly, this is effectively addressed through a simple self-prompting strategy, preventing the model's in-context learnability from devolving into specialist-like, context-free inference. Collectively, these insights culminate in our ConText model, which achieves new state-of-the-art across both in- and out-of-domain benchmarks. The code is available at https://github.com/Ferenas/ConText.

Massimiliano Lupo Pasini, Jong Youl Choi, Pei Zhang et al.

Graph foundation models using graph neural networks promise sustainable, efficient atomistic modeling. To tackle challenges of processing multi-source, multi-fidelity data during pre-training, recent studies employ multi-task learning, in which shared message passing layers initially process input atomistic structures regardless of source, then route them to multiple decoding heads that predict data-specific outputs. This approach stabilizes pre-training and enhances a model's transferability to unexplored chemical regions. Preliminary results on approximately four million structures are encouraging, yet questions remain about generalizability to larger, more diverse datasets and scalability on supercomputers. We propose a multi-task parallelism method that distributes each head across computing resources with GPU acceleration. Implemented in the open-source HydraGNN architecture, our method was trained on over 24 million structures from five datasets and tested on the Perlmutter, Aurora, and Frontier supercomputers, demonstrating efficient scaling on all three highly heterogeneous super-computing architectures.

Xiao Zhang, Xiangyu Han, Xiwen Lai et al.

Today's unsupervised image segmentation algorithms often segment suboptimally. Modern graph-cut based approaches rely on high-dimensional attention maps from Transformer-based foundation models, typically employing a relaxed Normalized Cut solved recursively via the Fiedler vector (the eigenvector of the second smallest eigenvalue). Consequently, they still lag behind supervised methods in both mask generation speed and segmentation accuracy. We present a regularized fractional alternating cut (Falcon), an optimization-based K-way Normalized Cut without relying on recursive eigenvector computations, achieving substantially improved speed and accuracy. Falcon operates in two stages: (1) a fast K-way Normalized Cut solved by extending into a fractional quadratic transformation, with an alternating iterative procedure and regularization to avoid local minima; and (2) refinement of the resulting masks using complementary low-level information, producing high-quality pixel-level segmentations. Experiments show that Falcon not only surpasses existing state-of-the-art methods by an average of 2.5% across six widely recognized benchmarks (reaching up to 4.3\% improvement on Cityscapes), but also reduces runtime by around 30% compared to prior graph-based approaches. These findings demonstrate that the semantic information within foundation-model attention can be effectively harnessed by a highly parallelizable graph cut framework. Consequently, Falcon can narrow the gap between unsupervised and supervised segmentation, enhancing scalability in real-world applications and paving the way for dense prediction-based vision pre-training in various downstream tasks. The code is released in https://github.com/KordingLab/Falcon.

Zhipeng Qian, Pei Zhang, Baosong Yang et al.

This paper introduces AnyTrans, an all-encompassing framework for the task-Translate AnyText in the Image (TATI), which includes multilingual text translation and text fusion within images. Our framework leverages the strengths of large-scale models, such as Large Language Models (LLMs) and text-guided diffusion models, to incorporate contextual cues from both textual and visual elements during translation. The few-shot learning capability of LLMs allows for the translation of fragmented texts by considering the overall context. Meanwhile, the advanced inpainting and editing abilities of diffusion models make it possible to fuse translated text seamlessly into the original image while preserving its style and realism. Additionally, our framework can be constructed entirely using open-source models and requires no training, making it highly accessible and easily expandable. To encourage advancement in the TATI task, we have meticulously compiled a test dataset called MTIT6, which consists of multilingual text image translation data from six language pairs.

Juntao Zhang, Shaogeng Liu, Kun Bian et al.

Mamba is an efficient State Space Model (SSM) with linear computational complexity. Although SSMs are not suitable for handling non-causal data, Vision Mamba (ViM) methods still demonstrate good performance in tasks such as image classification and object detection. Recent studies have shown that there is a rich theoretical connection between state space models and attention variants. We propose a novel separable self attention method, for the first time introducing some excellent design concepts of Mamba into separable self-attention. To ensure a fair comparison with ViMs, we introduce VMINet, a simple yet powerful prototype architecture, constructed solely by stacking our novel attention modules with the most basic down-sampling layers. Notably, VMINet differs significantly from the conventional Transformer architecture. Our experiments demonstrate that VMINet has achieved competitive results on image classification and high-resolution dense prediction tasks.Code is available at: https://github.com/yws-wxs/VMINet.

Massimiliano Lupo Pasini, Pei Zhang, Samuel Temple Reeve et al.

We introduce a multi-tasking graph convolutional neural network, HydraGNN, to simultaneously predict both global and atomic physical properties and demonstrate with ferromagnetic materials. We train HydraGNN on an open-source ab initio density functional theory (DFT) dataset for iron-platinum (FePt) with a fixed body centered tetragonal (BCT) lattice structure and fixed volume to simultaneously predict the mixing enthalpy (a global feature of the system), the atomic charge transfer, and the atomic magnetic moment across configurations that span the entire compositional range. By taking advantage of underlying physical correlations between material properties, multi-task learning (MTL) with HydraGNN provides effective training even with modest amounts of data. Moreover, this is achieved with just one architecture instead of three, as required by single-task learning (STL). The first convolutional layers of the HydraGNN architecture are shared by all learning tasks and extract features common to all material properties. The following layers discriminate the features of the different properties, the results of which are fed to the separate heads of the final layer to produce predictions. Numerical results show that HydraGNN effectively captures the relation between the configurational entropy and the material properties over the entire compositional range. Overall, the accuracy of simultaneous MTL predictions is comparable to the accuracy of the STL predictions. In addition, the computational cost of training HydraGNN for MTL is much lower than the original DFT calculations and also lower than training separate STL models for each property.

Qian Chen, Yafeng Chen, Yanni Chen et al.

Recent advancements in large language models (LLMs) and multimodal speech-text models have laid the groundwork for seamless voice interactions, enabling real-time, natural, and human-like conversations. Previous models for voice interactions are categorized as native and aligned. Native models integrate speech and text processing in one framework but struggle with issues like differing sequence lengths and insufficient pre-training. Aligned models maintain text LLM capabilities but are often limited by small datasets and a narrow focus on speech tasks. In this work, we introduce MinMo, a Multimodal Large Language Model with approximately 8B parameters for seamless voice interaction. We address the main limitations of prior aligned multimodal models. We train MinMo through multiple stages of speech-to-text alignment, text-to-speech alignment, speech-to-speech alignment, and duplex interaction alignment, on 1.4 million hours of diverse speech data and a broad range of speech tasks. After the multi-stage training, MinMo achieves state-of-the-art performance across various benchmarks for voice comprehension and generation while maintaining the capabilities of text LLMs, and also facilitates full-duplex conversation, that is, simultaneous two-way communication between the user and the system. Moreover, we propose a novel and simple voice decoder that outperforms prior models in voice generation. The enhanced instruction-following capabilities of MinMo supports controlling speech generation based on user instructions, with various nuances including emotions, dialects, and speaking rates, and mimicking specific voices. For MinMo, the speech-to-text latency is approximately 100ms, full-duplex latency is approximately 600ms in theory and 800ms in practice. The MinMo project web page is https://funaudiollm.github.io/minmo, and the code and models will be released soon.

Pei Zhang, Yunkai Liang, Kaiqiang Wang

Underwater environments impose severe constraints on conventional imaging systems and demand solutions that balance high-quality sensing with strict resource efficiency. While emerging event cameras offer a promising alternative, their potential in aquatic scenarios remains largely unexplored. Through the lens of neuromorphic vision, this work pioneers the investigation of motion fields that serve as key media for agile underwater perception. Built upon spiking neural networks, we introduce a self-supervised framework to estimate per-pixel optical flow from asynchronous event streams, elegantly bypassing the long-standing bottleneck of underwater data scarcity. Extensive evaluations demonstrate that our method achieves competitive visual and quantitative results against leading techniques while operating with superior computational efficiency. By bridging neuromorphic sensing and aquatic intelligence, this work opens new frontiers for lightweight, real-time, and low-cost perception on resource-constrained underwater edge platforms.

Yiming Wang, Pei Zhang, Siyuan Huang et al.

Test-time scaling enhances large language model performance by allocating additional compute resources during inference. Best-of-N (BoN) sampling serves as a common sampling-based scaling technique, broadening the search space in parallel to find better solutions from the model distribution. However, its cost-performance trade-off is still underexplored. Two main challenges limit the efficiency of BoN sampling: (1) Generating N full samples consumes substantial GPU memory, reducing inference capacity under limited resources. (2) Reward models add extra memory and latency overhead, and training strong reward models introduces potential training data costs. Although some studies have explored efficiency improvements, none have addressed both challenges at once. To address this gap, we propose Self-Truncation Best-of-N (ST-BoN), a decoding method that avoids fully generating all N samples and eliminates the need for reward models. It leverages early sampling consistency in the model's internal states to identify the most promising path and truncate suboptimal ones. In terms of cost, ST-BoN reduces dynamic GPU memory usage by over 80% and inference latency by 50%. In terms of cost-performance trade-off, ST-BoN achieves the same performance as Full-BoN while saving computational cost by 70%-80%, and under the same cost, it can improve accuracy by 3-4 points.

Yiming Wang, Pei Zhang, Baosong Yang et al.

LLM self-evaluation relies on the LLM's own ability to estimate response correctness, which can greatly improve its deployment reliability. In this research track, we propose the Chain-of-Embedding (CoE) in the latent space to enable LLMs to perform output-free self-evaluation. CoE consists of all progressive hidden states produced during the inference time, which can be treated as the latent thinking path of LLMs. We find that when LLMs respond correctly and incorrectly, their CoE features differ, these discrepancies assist us in estimating LLM response correctness. Experiments in four diverse domains and seven LLMs fully demonstrate the effectiveness of our method. Meanwhile, its label-free design intent without any training and millisecond-level computational cost ensures real-time feedback in large-scale scenarios. More importantly, we provide interesting insights into LLM response correctness from the perspective of hidden state changes inside LLMs.

Yiming Wang, Pei Zhang, Jialong Tang et al.

In this paper, we introduce PolyMath, a multilingual mathematical reasoning benchmark covering 18 languages and 4 easy-to-hard difficulty levels. Our benchmark ensures difficulty comprehensiveness, language diversity, and high-quality translation, making it a highly discriminative multilingual mathematical benchmark in the era of reasoning LLMs. We conduct a comprehensive evaluation for advanced LLMs and find that even Qwen-3-235B-A22B-Thinking and Gemini-2.5-pro, achieve only 54.6 and 52.2 benchmark scores, with about 40% accuracy under the highest level From a language perspective, our benchmark reveals several key challenges of LLMs in multilingual reasoning: (1) Reasoning performance varies widely across languages for current LLMs; (2) Input-output language consistency is low in reasoning LLMs and may be correlated with performance; (3) The thinking length differs significantly by language for current LLMs. Additionally, we demonstrate that controlling the output language in the instructions has the potential to affect reasoning performance, especially for some low-resource languages, suggesting a promising direction for improving multilingual capabilities in LLMs.

Yiming Wang, Pei Zhang, Baosong Yang et al.

Real-world data deviating from the independent and identically distributed (i.i.d.) assumption of in-distribution training data poses security threats to deep networks, thus advancing out-of-distribution (OOD) detection algorithms. Detection methods in generative language models (GLMs) mainly focus on uncertainty estimation and embedding distance measurement, with the latter proven to be most effective in traditional linguistic tasks like summarization and translation. However, another complex generative scenario mathematical reasoning poses significant challenges to embedding-based methods due to its high-density feature of output spaces, but this feature causes larger discrepancies in the embedding shift trajectory between different samples in latent spaces. Hence, we propose a trajectory-based method TV score, which uses trajectory volatility for OOD detection in mathematical reasoning. Experiments show that our method outperforms all traditional algorithms on GLMs under mathematical reasoning scenarios and can be extended to more applications with high-density features in output spaces, such as multiple-choice questions.

Pei Zhang, Zhaobo Hua, Jinliang Ding

Mammals can generate autonomous behaviors in various complex environments through the coordination and interaction of activities at different levels of their central nervous system. In this paper, we propose a novel hierarchical learning control framework by mimicking the hierarchical structure of the central nervous system along with their coordination and interaction behaviors. The framework combines the active and passive control systems to improve both the flexibility and reliability of the control system as well as to achieve more diverse autonomous behaviors of robots. Specifically, the framework has a backbone of independent neural network controllers at different levels and takes a three-level dual descending pathway structure, inspired from the functionality of the cerebral cortex, cerebellum, and spinal cord. We comprehensively validated the proposed approach through the simulation as well as the experiment of a hexapod robot in various complex environments, including obstacle crossing and rapid recovery after partial damage. This study reveals the principle that governs the autonomous behavior in the central nervous system and demonstrates the effectiveness of the hierarchical control approach with the salient features of the hierarchical learning control architecture and combination of active and passive control systems.

Yikang Liu, Yeting Shen, Hongao Zhu et al.

Whether and how language models (LMs) acquire the syntax of natural languages has been widely evaluated under the minimal pair paradigm. However, a lack of wide-coverage benchmarks in languages other than English has constrained systematic investigations into the issue. Addressing it, we first introduce ZhoBLiMP, the most comprehensive benchmark of linguistic minimal pairs for Chinese to date, with 118 paradigms, covering 15 linguistic phenomena. We then train 20 LMs of different sizes (14M to 1.4B) on Chinese corpora of various volumes (100M to 3B tokens) and evaluate them along with 14 off-the-shelf LLMs on ZhoBLiMP. The overall results indicate that Chinese grammar can be mostly learned by models with around 500M parameters, trained on 1B tokens with one epoch, showing limited benefits for further scaling. Most (N=95) linguistic paradigms are of easy or medium difficulty for LMs, while there are still 13 paradigms that remain challenging even for models with up to 32B parameters. In regard to how LMs acquire Chinese grammar, we observe a U-shaped learning pattern in several phenomena, similar to those observed in child language acquisition.

Pei Zhang, Zhaobo Hua, Jinliang Ding

The development of intelligent robots requires control policies that can handle dynamic environments and evolving tasks. Pre-training reinforcement learning has emerged as an effective approach to address these demands by enabling robots to acquire reusable motor skills. However, they often rely on large datasets or expert-designed goal spaces, limiting adaptability. Additionally, these methods need help to generate dynamic and diverse skills in high-dimensional state spaces, reducing their effectiveness for downstream tasks. In this paper, we propose CMS-PRL, a pre-training reinforcement learning method inspired by the Central Motor System (CMS). First, we introduce a fusion reward mechanism that combines the basic motor reward with mutual information reward, promoting the discovery of dynamic skills during pre-training without reliance on external data. Second, we design a skill encoding method inspired by the motor program of the basal ganglia, providing rich and continuous skill instructions during pre-training. Finally, we propose a skill activity function to regulate motor skill activity, enabling the generation of skills with different activity levels, thereby enhancing the robot's flexibility in downstream tasks. We evaluate the model on four types of robots in a challenging set of sparse-reward tasks. Experimental results demonstrate that CMS-PRL generates diverse, reusable motor skills to solve various downstream tasks and outperforms baseline methods, particularly in high-degree-of-freedom robots and complex tasks.

Haiquan Zhao, Chenhan Yuan, Fei Huang et al.

As large language models (LLMs) become more capable and widely used, ensuring the safety of their outputs is increasingly critical. Existing guardrail models, though useful in static evaluation settings, face two major limitations in real-world applications: (1) they typically output only binary "safe/unsafe" labels, which can be interpreted inconsistently across diverse safety policies, rendering them incapable of accommodating varying safety tolerances across domains; and (2) they require complete model outputs before performing safety checks, making them fundamentally incompatible with streaming LLM inference, thereby preventing timely intervention during generation and increasing exposure to harmful partial outputs. To address these challenges, we present Qwen3Guard, a series of multilingual safety guardrail models with two specialized variants: Generative Qwen3Guard, which casts safety classification as an instruction-following task to enable fine-grained tri-class judgments (safe, controversial, unsafe); and Stream Qwen3Guard, which introduces a token-level classification head for real-time safety monitoring during incremental text generation. Both variants are available in three sizes (0.6B, 4B, and 8B parameters) and support up to 119 languages and dialects, providing comprehensive, scalable, and low-latency safety moderation for global LLM deployments. Evaluated across English, Chinese, and multilingual benchmarks, Qwen3Guard achieves state-of-the-art performance in both prompt and response safety classification. All models are released under the Apache 2.0 license for public use.

Pei Zhang, M. Paul Laiu, Matthew Norman et al.

Accurate representation of the multiscale features in spatiotemporal physical systems using vision transformer (ViT) architectures requires extremely long, computationally prohibitive token sequences. To address this issue, we propose two adaptive tokenization schemes that dynamically adjust patch sizes based on local features: one ensures convergent behavior to uniform patch refinement, while the other offers better computational efficiency. Moreover, we present a set of spatiotemporal attention schemes, where the temporal or axial spatial dimensions are decoupled, and evaluate their computational and data efficiencies. We assess the performance of the proposed multiscale adaptive model, MATEY, in a sequence of experiments. The results show that adaptive tokenization schemes achieve improved accuracy without significantly increasing the length of the token sequence. Compared to a full spatiotemporal attention scheme or a scheme that decouples only the temporal dimension, we find that fully decoupled axial attention is less efficient and expressive, requiring more training time and model weights to achieve the same accuracy. Finally, we demonstrate in two fine-tuning tasks featuring different physics that models pretrained on PDEBench data outperform the ones trained from scratch, especially in the low data regime with frozen attention.

Qiyuan Ou, Pei Zhang, Sihang Zhou et al.

Late fusion multi-view clustering (LFMVC) has become a rapidly growing class of methods in the multi-view clustering (MVC) field, owing to its excellent computational speed and clustering performance. One bottleneck faced by existing late fusion methods is that they are usually aligned to the average kernel function, which makes the clustering performance highly dependent on the quality of datasets. Another problem is that they require subsequent k-means clustering after obtaining the consensus partition matrix to get the final discrete labels, and the resulting separation of the label learning and cluster structure optimization processes limits the integrity of these models. To address the above issues, we propose an integrated framework named One-Step Late Fusion Multi-view Clustering with Compressed Subspace (OS-LFMVC-CS). Specifically, we use the consensus subspace to align the partition matrix while optimizing the partition fusion, and utilize the fused partition matrix to guide the learning of discrete labels. A six-step iterative optimization approach with verified convergence is proposed. Sufficient experiments on multiple datasets validate the effectiveness and efficiency of our proposed method.

Suhang Wu, Jialong Tang, Chengyi Yang et al.

Direct speech translation (ST) has garnered increasing attention nowadays, yet the accurate translation of terminology within utterances remains a great challenge. In this regard, current studies mainly concentrate on leveraging various translation knowledge into ST models. However, these methods often struggle with interference from irrelevant noise and can not fully utilize the translation knowledge. To address these issues, in this paper, we propose a novel Locate-and-Focus method for terminology translation. It first effectively locates the speech clips containing terminologies within the utterance to construct translation knowledge, minimizing irrelevant information for the ST model. Subsequently, it associates the translation knowledge with the utterance and hypothesis from both audio and textual modalities, allowing the ST model to better focus on translation knowledge during translation. Experimental results across various datasets demonstrate that our method effectively locates terminologies within utterances and enhances the success rate of terminology translation, while maintaining robust general translation performance.

Sungmin Lee, Minju Kang, Joonhee Lee et al.

Question-answering (QA) interfaces powered by large language models (LLMs) present a promising direction for improving interactivity with HVAC system insights, particularly for non-expert users. However, enabling accurate, real-time, and context-aware interactions with HVAC systems introduces unique challenges, including the integration of frequently updated sensor data, domain-specific knowledge grounding, and coherent multi-stage reasoning. In this paper, we present JARVIS, a two-stage LLM-based QA framework tailored for sensor data-driven HVAC system interaction. JARVIS employs an Expert-LLM to translate high-level user queries into structured execution instructions, and an Agent that performs SQL-based data retrieval, statistical processing, and final response generation. To address HVAC-specific challenges, JARVIS integrates (1) an adaptive context injection strategy for efficient HVAC and deployment-specific information integration, (2) a parameterized SQL builder and executor to improve data access reliability, and (3) a bottom-up planning scheme to ensure consistency across multi-stage response generation. We evaluate JARVIS using real-world data collected from a commercial HVAC system and a ground truth QA dataset curated by HVAC experts to demonstrate its effectiveness in delivering accurate and interpretable responses across diverse queries. Results show that JARVIS consistently outperforms baseline and ablation variants in both automated and user-centered assessments, achieving high response quality and accuracy.

Shouxu Lin, Zimeng Pan, Yuhang Yao et al.

Multi-Model Federated Learning (MMFL) is an emerging direction in Federated Learning (FL) where multiple models are trained in parallel, generally on various datasets. Optimizing the models' accuracies and training times in the MMFL setting requires adapting to data and system heterogeneity across clients as in single-model FL; these challenges are amplified in the MMFL setting due to additional heterogeneity across models. Neither existing solutions nor naïve extensions of single-model FL frameworks efficiently address these challenges. To bridge this gap, we propose FLAMMABLE, a comprehensive MMFL training framework. FLAMMABLE optimizes model training by intelligently adapting client batch sizes while engaging them to train multiple carefully chosen models, depending on their system capabilities, in each training round. To evaluate FLAMMABLE, we develop the first benchmark platform for the MMFL setting, which may enable future reproducible MMFL research. Extensive evaluations on multiple datasets and models show that FLAMMABLE boosts the MMFL time-to-accuracy performance by 1.1$\sim$10.0$\times$ while improving the final model accuracy by 1.3$\sim$5.4\% compared to several known baselines.

Pei Zhang, Andong Chen, Xi Chen et al.

Large language models (LLMs) have expanded from text to speech, giving rise to Speech Large Models (SLMs) that support recognition, translation, and synthesis. A key challenge is aligning speech and text representations, which becomes harder in multilingual settings. Existing methods often freeze LLM parameters and train encoders on multilingual data, but this forces cross-language convergence and limits performance. We introduce Progressive Alignment Representation Training (PART), a multi-stage and multi-task framework that separates within-language from cross-language alignment. During cross-language training, LLM parameters are dynamically activated, and text-based tasks are later introduced to enhance multilingual understanding. Experiments on CommonVoice 15, Fleurs, Wenetspeech, and CoVoST2 show that PART surpasses conventional approaches, with analysis confirming its ability to balance language-specific distinctions and cross-language generalization. These results demonstrate PART's effectiveness and generality for multilingual speech modality alignment.

Pei Zhang, Yiming Wang, Jialong Tang et al.

Simultaneous speech-to-text translation (Simul-S2TT) aims to translate speech into target text in real time, outputting translations while receiving source speech input, rather than waiting for the entire utterance to be spoken. Simul-S2TT research often modifies model architectures to implement read-write strategies. However, with the rise of large audio-language models (LALMs), a key challenge is how to directly activate Simul-S2TT capabilities in base models without additional architectural changes. In this paper, we introduce {\bf Simul}taneous {\bf S}elf-{\bf A}ugmentation ({\bf SimulSA}), a strategy that utilizes LALMs' inherent capabilities to obtain simultaneous data by randomly truncating speech and constructing partially aligned translation. By incorporating them into offline SFT data, SimulSA effectively bridges the distribution gap between offline translation during pretraining and simultaneous translation during inference. Experimental results demonstrate that augmenting only about {\bf 1\%} of the simultaneous data, compared to the full offline SFT data, can significantly activate LALMs' Simul-S2TT capabilities without modifications to model architecture or decoding strategy.

Wesley Brewer, Murali Meena Gopalakrishnan, Matthias Maiterth et al.

With the end of Moore's law and Dennard scaling, efficient training increasingly requires rethinking data volume. Can we train better models with significantly less data via intelligent subsampling? To explore this, we develop SICKLE, a sparse intelligent curation framework for efficient learning, featuring a novel maximum entropy (MaxEnt) sampling approach, scalable training, and energy benchmarking. We compare MaxEnt with random and phase-space sampling on large direct numerical simulation (DNS) datasets of turbulence. Evaluating SICKLE at scale on Frontier, we show that subsampling as a preprocessing step can, in many cases, improve model accuracy and substantially lower energy consumption, with observed reductions of up to 38x.

Junqi Yin, Mijanur Palash, M. Paul Laiu et al.

Turbulence plays a crucial role in multiphysics applications, including aerodynamics, fusion, and combustion. Accurately capturing turbulence's multiscale characteristics is essential for reliable predictions of multiphysics interactions, but remains a grand challenge even for exascale supercomputers and advanced deep learning models. The extreme-resolution data required to represent turbulence, ranging from billions to trillions of grid points, pose prohibitive computational costs for models based on architectures like vision transformers. To address this challenge, we introduce a multiscale hierarchical Turbulence Transformer that reduces sequence length from billions to a few millions and a novel RingX sequence parallelism approach that enables scalable long-context learning. We perform scaling and science runs on the Frontier supercomputer. Our approach demonstrates excellent performance up to 1.1 EFLOPS on 32,768 AMD GPUs, with a scaling efficiency of 94%. To our knowledge, this is the first AI model for turbulence that can capture small-scale eddies down to the dissipative range.