Wei Yuan

OneRec Team, Biao Yang, Boyang Ding et al.

Generative recommendation models in the OneRec family have been widely deployed in many real-world services, such as short-video, live-streaming, advertising, and e-commerce. However, these generative models can only benefit from the scaling advantage, while their reasoning ability is hard to activate, since we cannot construct meaningful Chain-of-Thought (CoT) sequences consisting of itemic tokens only. Inspired by the success of the reasoning-style ``think before answer'' paradigm in the LLM field, we conduct preliminary studies (i.e., OneRec-Think, OpenOneRec) to explore reasoning capability in generative recommendation. Nevertheless, we notice an unexpected phenomenon: the thinking mode does not show advantages over the non-thinking mode. Drawing insights from recent findings on CoT robustness in multi-modal language models, we argue that effective reasoning in recommendation rests on two factors: perception, the ability to ground itemic tokens in their underlying language semantics, and cognition, the ability to reorganize a user's behavior sequence into coherent latent interest points. We therefore propose OneReason, which includes: (1) strong itemic token perception in pre-training, (2) a three-level cognition-enhanced CoT format for recommendation tasks in SFT, and (3) a specialize-then-unify training recipe in RL to enhance the thinking ability.

Wei Yuan, Jianwei Huang, Ying Jun Zhang

This paper considers the problem of charging station pricing and plug-in electric vehicles (PEVs) station selection. When a PEV needs to be charged, it selects a charging station by considering the charging prices, waiting times, and travel distances. Each charging station optimizes its charging price based on the prediction of the PEVs' charging station selection decisions and the other station's pricing decision, in order to maximize its profit. To obtain insights of such a highly coupled system, we consider a one-dimensional system with two competing charging stations and Poisson arriving PEVs. We propose a multi-leader-multi-follower Stackelberg game model, in which the charging stations (leaders) announce their charging prices in Stage I, and the PEVs (followers) make their charging station selections in Stage II. We show that there always exists a unique charging station selection equilibrium in Stage II, and such equilibrium depends on the charging stations' service capacities and the price difference between them. We then characterize the sufficient conditions for the existence and uniqueness of the pricing equilibrium in Stage I. We also develop a low complexity algorithm that efficiently computes the pricing equilibrium and the subgame perfect equilibrium of the two-stage Stackelberg game.

Yueyuan Li, Wei Yuan, Songan Zhang et al.

Simulators play a crucial role in autonomous driving, offering significant time, cost, and labor savings. Over the past few years, the number of simulators for autonomous driving has grown substantially. However, there is a growing concern about the validity of algorithms developed and evaluated in simulators, indicating a need for a thorough analysis of the development status of the simulators. To bridge the gap in research, this paper analyzes the evolution of simulators and explains how the functionalities and utilities have developed. Then, the existing simulators are categorized based on their task applicability, providing researchers with a taxonomy to swiftly assess a simulator's suitability for specific tasks. Recommendations for select simulators are presented, considering factors such as accessibility, maintenance status, and quality. Recognizing potential hazards in simulators that could impact the confidence of simulation experiments, the paper dedicates substantial effort to identifying and justifying critical issues in actively maintained open-source simulators. Moreover, the paper reviews potential solutions to address these issues, serving as a guide for enhancing the credibility of simulators.

Zipei Fan, Xiaojie Yang, Wei Yuan et al.

Knowing "what is happening" and "what will happen" of the mobility in a city is the building block of a data-driven smart city system. In recent years, mobility digital twin that makes a virtual replication of human mobility and predicting or simulating the fine-grained movements of the subjects in a virtual space at a metropolitan scale in near real-time has shown its great potential in modern urban intelligent systems. However, few studies have provided practical solutions. The main difficulties are four-folds. 1) The daily variation of human mobility is hard to model and predict; 2) the transportation network enforces a complex constraints on human mobility; 3) generating a rational fine-grained human trajectory is challenging for existing machine learning models; and 4) making a fine-grained prediction incurs high computational costs, which is challenging for an online system. Bearing these difficulties in mind, in this paper we propose a two-stage human mobility predictor that stratifies the coarse and fine-grained level predictions. In the first stage, to encode the daily variation of human mobility at a metropolitan level, we automatically extract citywide mobility trends as crowd contexts and predict long-term and long-distance movements at a coarse level. In the second stage, the coarse predictions are resolved to a fine-grained level via a probabilistic trajectory retrieval method, which offloads most of the heavy computations to the offline phase. We tested our method using a real-world mobile phone GPS dataset in the Kanto area in Japan, and achieved good prediction accuracy and a time efficiency of about 2 min in predicting future 1h movements of about 220K mobile phone users on a single machine to support more higher-level analysis of mobility prediction.

Zhaofeng Zhong, Wei Yuan, Tong Chen et al.

Recent large reasoning models (LRMs) have made substantial progress in complex reasoning tasks, yet they often generate lengthy reasoning paths for every query, incurring unnecessary computation and latency. Existing speed-up approaches typically rely on retraining the model or designing sophisticated prompting, which are either prohibitively expensive or highly sensitive to the input and prompt formulation. In this work, we study model merging as a lightweight alternative for efficient reasoning: by combining a long chain-of-thought (Long-CoT) reasoning model with a Short-CoT instruction model, we obtain an adaptive reasoner without training from scratch or requiring large-scale additional data. Building on this idea, we propose Reasoning Pattern Alignment Merging (RPAM), a layer-wise model merging framework based on feature alignment to facilitate query-adaptive reasoning. RPAM first constructs a small pattern-labeled calibration set that assigns each query an appropriate reasoning pattern. It then optimizes layer-wise merging coefficients by aligning the merged model's intermediate representations with those of the selected model, while a contrastive objective explicitly pushes them away from the non-selected model. Experiments on seven widely used reasoning benchmarks show that RPAM substantially reduces inference cost while maintaining strong performance. Upon article acceptance, we will provide open-source code to reproduce experiments for RPAM.

Weihang Ran, Wei Yuan, Xiaodan Shi et al.

Building outline extracted from high-resolution aerial images can be used in various application fields such as change detection and disaster assessment. However, traditional CNN model cannot recognize contours very precisely from original images. In this paper, we proposed a CNN and Transformer based model together with active contour model to deal with this problem. We also designed a triple-branch decoder structure to handle different features generated by encoder. Experiment results show that our model outperforms other baseline model on two datasets, achieving 91.1% mIoU on Vaihingen and 83.8% on Bing huts.

Ailin Huang, Ang Li, Aobo Kong et al.

We introduce Step 3.5 Flash, a sparse Mixture-of-Experts (MoE) model that bridges frontier-level agentic intelligence and computational efficiency. We focus on what matters most when building agents: sharp reasoning and fast, reliable execution. Step 3.5 Flash pairs a 196B-parameter foundation with 11B active parameters for efficient inference. It is optimized with interleaved 3:1 sliding-window/full attention and Multi-Token Prediction (MTP-3) to reduce the latency and cost of multi-round agentic interactions. To reach frontier-level intelligence, we design a scalable reinforcement learning framework that combines verifiable signals with preference feedback, while remaining stable under large-scale off-policy training, enabling consistent self-improvement across mathematics, code, and tool use. Step 3.5 Flash demonstrates strong performance across agent, coding, and math tasks, achieving 85.4% on IMO-AnswerBench, 86.4% on LiveCodeBench-v6 (2024.08-2025.05), 88.2% on tau2-Bench, 69.0% on BrowseComp (with context management), and 51.0% on Terminal-Bench 2.0, comparable to frontier models such as GPT-5.2 xHigh and Gemini 3.0 Pro. By redefining the efficiency frontier, Step 3.5 Flash provides a high-density foundation for deploying sophisticated agents in real-world industrial environments.

Kaibing Chen, Dong Shen, Hanwen Zhong et al.

In the field of multi-modal language models, the majority of methods are built on an architecture similar to LLaVA. These models use a single-layer ViT feature as a visual prompt, directly feeding it into the language models alongside textual tokens. However, when dealing with long sequences of visual signals or inputs such as videos, the self-attention mechanism of language models can lead to significant computational overhead. Additionally, using single-layer ViT features makes it challenging for large language models to perceive visual signals fully. This paper proposes an efficient multi-modal language model to minimize computational costs while enabling the model to perceive visual signals as comprehensively as possible. Our method primarily includes: (1) employing cross-attention to image-text interaction similar to Flamingo. (2) utilize hierarchical ViT features. (3) introduce the Mixture of Experts (MoE) mechanism to enhance model effectiveness. Our model achieves competitive scores on public multi-modal benchmarks and performs well in tasks such as image captioning and video captioning.

Shutong Qiao, Wei Yuan, Tong Chen et al.

Semantic ID learning is a key interface in Generative Recommendation (GR) models, mapping items to discrete identifiers grounded in side information, most commonly via a pretrained text encoder. However, these text encoders are primarily optimized for well-formed natural language. In real-world recommendation data, item descriptions are often symbolic and attribute-centric, containing numerals, units, and abbreviations. These text encoders can break these signals into fragmented tokens, weakening semantic coherence and distorting relationships among attributes. Worse still, when moving to multimodal GR, relying on standard text encoders introduces an additional obstacle: text and image embeddings often exhibit mismatched geometric structures, making cross-modal fusion less effective and less stable. In this paper, we revisit representation design for Semantic ID learning by treating text as a visual signal. We conduct a systematic empirical study of OCR-based text representations, obtained by rendering item descriptions into images and encoding them with vision-based OCR models. Experiments across four datasets and two generative backbones show that OCR-text consistently matches or surpasses standard text embeddings for Semantic ID learning in both unimodal and multimodal settings. Furthermore, we find that OCR-based Semantic IDs remain robust under extreme spatial-resolution compression, indicating strong robustness and efficiency in practical deployments.

Chen Hu, Haikuo Du, Heng Wang et al.

As LLMs shift toward autonomous agents, Deep Research has emerged as a pivotal metric. However, existing academic benchmarks like BrowseComp often fail to meet real-world demands for open-ended research, which requires robust skills in intent recognition, long-horizon decision-making, and cross-source verification. To address this, we introduce Step-DeepResearch, a cost-effective, end-to-end agent. We propose a Data Synthesis Strategy Based on Atomic Capabilities to reinforce planning and report writing, combined with a progressive training path from agentic mid-training to SFT and RL. Enhanced by a Checklist-style Judger, this approach significantly improves robustness. Furthermore, to bridge the evaluation gap in the Chinese domain, we establish ADR-Bench for realistic deep research scenarios. Experimental results show that Step-DeepResearch (32B) scores 61.4% on Scale AI Research Rubrics. On ADR-Bench, it significantly outperforms comparable models and rivals SOTA closed-source models like OpenAI and Gemini DeepResearch. These findings prove that refined training enables medium-sized models to achieve expert-level capabilities at industry-leading cost-efficiency.

Yicheng Di, Wei Yuan, Tieke He et al.

Vision-language pretrained models offer strong transferable representations, yet adapting them in privacy-sensitive multi-party settings is challenging due to the high communication cost of federated optimization and the limited local data on clients. Federated prompt learning mitigates this issue by keeping the VLPM backbone frozen and collaboratively training lightweight prompt parameters. However, existing approaches typically enforce a unified prompt structure and length across clients, which is inadequate under practical client heterogeneity in both data distributions and system resources, and may further introduce conflicts between globally shared and locally optimal knowledge. To address these challenges, we propose \textbf{SDFed}, a heterogeneous federated prompt learning framework that bridges Local-Global Discrepancy via Subspace Refinement and Divergence Control. SDFed maintains a fixed-length global prompt for efficient aggregation while allowing each client to learn a variable-length local prompt to better match its data characteristics and capacity. To mitigate local-global conflicts and facilitate effective knowledge transfer, SDFed introduces a subspace refinement method for local prompts and an information retention and divergence control strategy that preserves key local information while maintaining appropriate separability between global and local representations. Extensive experiments on several datasets demonstrate that SDFed consistently improves performance and robustness in heterogeneous federated settings.

Sixiao Zhang, Cheng Long, Wei Yuan et al.

Recommender systems embody significant commercial value and represent crucial intellectual property. However, the integrity of these systems is constantly challenged by malicious actors seeking to steal their underlying models. Safeguarding against such threats is paramount to upholding the rights and interests of the model owner. While model watermarking has emerged as a potent defense mechanism in various domains, its direct application to recommender systems remains unexplored and non-trivial. In this paper, we address this gap by introducing Autoregressive Out-of-distribution Watermarking (AOW), a novel technique tailored specifically for recommender systems. Our approach entails selecting an initial item and querying it through the oracle model, followed by the selection of subsequent items with small prediction scores. This iterative process generates a watermark sequence autoregressively, which is then ingrained into the model's memory through training. To assess the efficacy of the watermark, the model is tasked with predicting the subsequent item given a truncated watermark sequence. Through extensive experimentation and analysis, we demonstrate the superior performance and robust properties of AOW. Notably, our watermarking technique exhibits high-confidence extraction capabilities and maintains effectiveness even in the face of distillation and fine-tuning processes.

Liang Qu, Jianxin Li, Wei Yuan et al.

Federated recommender systems (FedRS) have emerged as a paradigm for protecting user privacy by keeping interaction data on local devices while coordinating model training through a central server. However, most existing federated recommender systems adopt a one-size-fits-all assumption on user privacy, where all users are required to keep their data strictly local. This setting overlooks users who are willing to share their data with the server in exchange for better recommendation performance. Although several recent studies have explored personalized user data sharing in FedRS, they assume static user privacy preferences and cannot handle user requests to remove previously shared data and its corresponding influence on the trained model. To address this limitation, we propose FedShare, a federated learn-unlearn framework for recommender systems with personalized user data sharing. FedShare not only allows users to control how much interaction data is shared with the server, but also supports data unsharing requests by removing the influence of the unshared data from the trained model. Specifically, FedShare leverages shared data to construct a server-side high-order user-item graph and uses contrastive learning to jointly align local and global representations. In the unlearning phase, we design a contrastive unlearning mechanism that selectively removes representations induced by the unshared data using a small number of historical embedding snapshots, avoiding the need to store large amounts of historical gradient information as required by existing federated recommendation unlearning methods. Extensive experiments on three public datasets demonstrate that FedShare achieves strong recommendation performance in both the learning and unlearning phases, while significantly reducing storage overhead in the unlearning phase compared with state-of-the-art baselines.

Lihao Liu, Yan Wang, Biao Yang et al.

Multimodal Large Language Models (MLLMs) have shown remarkable success in comprehension tasks such as visual description and visual question answering. However, their direct application to embedding-based tasks like retrieval remains challenging due to the discrepancy between output formats and optimization objectives. Previous approaches often employ contrastive fine-tuning to adapt MLLMs for retrieval, but at the cost of losing their generative capabilities. We argue that both generative and embedding tasks fundamentally rely on shared cognitive mechanisms, specifically cross-modal representation alignment and contextual comprehension. To this end, we propose CREM (Compression-driven Representation Enhanced Model), with a unified framework that enhances multimodal representations for retrieval while preserving generative ability. Specifically, we introduce a compression-based prompt design with learnable chorus tokens to aggregate multimodal semantics and a compression-driven training strategy that integrates contrastive and generative objectives through compression-aware attention. Extensive experiments demonstrate that CREM achieves state-of-the-art retrieval performance on MMEB while maintaining strong generative performance on multiple comprehension benchmarks. Our findings highlight that generative supervision can further improve the representational quality of MLLMs under the proposed compression-driven paradigm.

Yuchen Gui, Boyi Wei, Wei Yuan et al.

Sampling is an important process in many GNN structures in order to train larger datasets with a smaller computational complexity. However, compared to other processes in GNN (such as aggregate, backward propagation), the sampling process still costs tremendous time, which limits the speed of training. To reduce the time of sampling, hardware acceleration is an ideal choice. However, state of the art GNN acceleration proposal did not specify how to accelerate the sampling process. What's more, directly accelerating traditional sampling algorithms will make the structure of the accelerator very complicated. In this work, we made two contributions: (1) Proposed a new neighbor sampler: CONCAT Sampler, which can be easily accelerated on hardware level while guaranteeing the test accuracy. (2) Designed a CONCAT-sampler-accelerator based on FPGA, with which the neighbor sampling process boosted to about 300-1000 times faster compared to the sampling process without it.

Da Li, Yuxiao Luo, Keping Bi et al.

Vision-language models advance multimodal representation learning by acquiring transferable semantic embeddings, thereby substantially enhancing performance across a range of vision-language tasks, including cross-modal retrieval, clustering, and classification. An effective embedding is expected to comprehensively preserve the semantic content of the input while simultaneously emphasizing features that are discriminative for downstream tasks. Recent approaches demonstrate that VLMs can be adapted into competitive embedding models via large-scale contrastive learning, enabling the simultaneous optimization of two complementary objectives. We argue that the two aforementioned objectives can be decoupled: a comprehensive understanding of the input facilitates the embedding model in achieving superior performance in downstream tasks via contrastive learning. In this paper, we propose CoMa, a compressed pre-training phase, which serves as a warm-up stage for contrastive learning. Experiments demonstrate that with only a small amount of pre-training data, we can transform a VLM into a competitive embedding model. CoMa achieves new state-of-the-art results among VLMs of comparable size on the MMEB, realizing optimization in both efficiency and effectiveness.

Qizhe Wu, Huawen Liang, Yuchen Gui et al.

General matrix-matrix multiplication (GEMM) is a cornerstone of AI computations, making tensor processing engines (TPEs) increasingly critical in GPUs and domain-specific architectures. Existing architectures primarily optimize dataflow or operand reuse strategies. However, considering the interaction between matrix multiplication and multiply-accumulators (MACs) offers greater optimization potential. This work introduces a novel hardware perspective on matrix multiplication, focusing on the bit-weight dimension of MACs. We propose a finer-grained TPE notation using matrix triple loops as an example, introducing new methods for designing and optimizing PE microarchitectures. Based on this notation and its transformations, we propose four optimization techniques that improve timing, area, and power consumption. Implementing our design in RTL using the SMIC-28nm process, we evaluate its effectiveness across four classic TPE architectures: systolic array, 3D-Cube, multiplier-adder tree, and 2D-Matrix. Our techniques achieve area efficiency improvements of 1.27x, 1.28x, 1.56x, and 1.44x, and energy efficiency gains of 1.04x, 1.56x, 1.49x, and 1.20x, respectively. Applied to a bit-slice architecture, our approach achieves a 12.10x improvement in energy efficiency and 2.85x in area efficiency compared to Laconic. Our Verilog HDL code, along with timing, area, and power reports, is available at https://github.com/wqzustc/High-Performance-Tensor-Processing-Engines

Xiaobo Xing, Wei Yuan, Tong Chen et al.

Modeling semantic and structural information from tabular data remains a core challenge for effective table understanding. Existing Table-as-Text approaches flatten tables for large language models (LLMs), but lose crucial structural cues, while Table-as-Image methods preserve structure yet struggle with fine-grained semantics. Recent Table-as-Multimodality strategies attempt to combine textual and visual views, but they (1) statically process both modalities for every query-table pair within a large multimodal LLMs (MLLMs), inevitably introducing redundancy and even conflicts, and (2) depend on costly fine-tuning of MLLMs. In light of this, we propose TableDART, a training-efficient framework that integrates multimodal views by reusing pretrained single-modality models. TableDART introduces a lightweight 2.59M-parameter MLP gating network that dynamically selects the optimal path (either Text-only, Image-only, or Fusion) for each table-query pair, effectively reducing redundancy and conflicts from both modalities. In addition, we propose a novel agent to mediate cross-modal knowledge integration by analyzing outputs from text- and image-based models, either selecting the best result or synthesizing a new answer through reasoning. This design avoids the prohibitive costs of full MLLM fine-tuning. Extensive experiments on seven benchmarks show that TableDART establishes new state-of-the-art performance among open-source models, surpassing the strongest baseline by an average of 4.02%. The code is available at: https://anonymous.4open.science/r/TableDART-C52B

Hezheng Lin, Xing Cheng, Xiangyu Wu et al.

Since Transformer has found widespread use in NLP, the potential of Transformer in CV has been realized and has inspired many new approaches. However, the computation required for replacing word tokens with image patches for Transformer after the tokenization of the image is vast(e.g., ViT), which bottlenecks model training and inference. In this paper, we propose a new attention mechanism in Transformer termed Cross Attention, which alternates attention inner the image patch instead of the whole image to capture local information and apply attention between image patches which are divided from single-channel feature maps capture global information. Both operations have less computation than standard self-attention in Transformer. By alternately applying attention inner patch and between patches, we implement cross attention to maintain the performance with lower computational cost and build a hierarchical network called Cross Attention Transformer(CAT) for other vision tasks. Our base model achieves state-of-the-arts on ImageNet-1K, and improves the performance of other methods on COCO and ADE20K, illustrating that our network has the potential to serve as general backbones. The code and models are available at \url{https://github.com/linhezheng19/CAT}.

Wei Yuan, Kai-Xin Gao

Many adaptive optimization methods have been proposed and used in deep learning, in which Adam is regarded as the default algorithm and widely used in many deep learning frameworks. Recently, many variants of Adam, such as Adabound, RAdam and Adabelief, have been proposed and show better performance than Adam. However, these variants mainly focus on changing the stepsize by making differences on the gradient or the square of it. Motivated by the fact that suitable damping is important for the success of powerful second-order optimizers, we discuss the impact of the constant $ε$ for Adam in this paper. Surprisingly, we can obtain better performance than Adam simply changing the position of $ε$. Based on this finding, we propose a new variant of Adam called EAdam, which doesn't need extra hyper-parameters or computational costs. We also discuss the relationships and differences between our method and Adam. Finally, we conduct extensive experiments on various popular tasks and models. Experimental results show that our method can bring significant improvement compared with Adam. Our code is available at https://github.com/yuanwei2019/EAdam-optimizer.

Kwai Keye Team, Biao Yang, Bin Wen et al.

While Multimodal Large Language Models (MLLMs) demonstrate remarkable capabilities on static images, they often fall short in comprehending dynamic, information-dense short-form videos, a dominant medium in today's digital landscape. To bridge this gap, we introduce \textbf{Kwai Keye-VL}, an 8-billion-parameter multimodal foundation model engineered for leading-edge performance in short-video understanding while maintaining robust general-purpose vision-language abilities. The development of Keye-VL rests on two core pillars: a massive, high-quality dataset exceeding 600 billion tokens with a strong emphasis on video, and an innovative training recipe. This recipe features a four-stage pre-training process for solid vision-language alignment, followed by a meticulous two-phase post-training process. The first post-training stage enhances foundational capabilities like instruction following, while the second phase focuses on stimulating advanced reasoning. In this second phase, a key innovation is our five-mode ``cold-start'' data mixture, which includes ``thinking'', ``non-thinking'', ``auto-think'', ``think with image'', and high-quality video data. This mixture teaches the model to decide when and how to reason. Subsequent reinforcement learning (RL) and alignment steps further enhance these reasoning capabilities and correct abnormal model behaviors, such as repetitive outputs. To validate our approach, we conduct extensive evaluations, showing that Keye-VL achieves state-of-the-art results on public video benchmarks and remains highly competitive on general image-based tasks (Figure 1). Furthermore, we develop and release the \textbf{KC-MMBench}, a new benchmark tailored for real-world short-video scenarios, where Keye-VL shows a significant advantage.

Xinchen Luo, Jiangxia Cao, Tianyu Sun et al.

In recent years, with the significant evolution of multi-modal large models, many recommender researchers realized the potential of multi-modal information for user interest modeling. In industry, a wide-used modeling architecture is a cascading paradigm: (1) first pre-training a multi-modal model to provide omnipotent representations for downstream services; (2) The downstream recommendation model takes the multi-modal representation as additional input to fit real user-item behaviours. Although such paradigm achieves remarkable improvements, however, there still exist two problems that limit model performance: (1) Representation Unmatching: The pre-trained multi-modal model is always supervised by the classic NLP/CV tasks, while the recommendation models are supervised by real user-item interaction. As a result, the two fundamentally different tasks' goals were relatively separate, and there was a lack of consistent objective on their representations; (2) Representation Unlearning: The generated multi-modal representations are always stored in cache store and serve as extra fixed input of recommendation model, thus could not be updated by recommendation model gradient, further unfriendly for downstream training. Inspired by the two difficulties challenges in downstream tasks usage, we introduce a quantitative multi-modal framework to customize the specialized and trainable multi-modal information for different downstream models.

Sida Tian, Can Zhang, Wei Yuan et al.

In recent years, remarkable advancements in artificial intelligence-generated content (AIGC) have been achieved in the fields of image synthesis and text generation, generating content comparable to that produced by humans. However, the quality of AI-generated music has not yet reached this standard, primarily due to the challenge of effectively controlling musical emotions and ensuring high-quality outputs. This paper presents a generalized symbolic music generation framework, XMusic, which supports flexible prompts (i.e., images, videos, texts, tags, and humming) to generate emotionally controllable and high-quality symbolic music. XMusic consists of two core components, XProjector and XComposer. XProjector parses the prompts of various modalities into symbolic music elements (i.e., emotions, genres, rhythms and notes) within the projection space to generate matching music. XComposer contains a Generator and a Selector. The Generator generates emotionally controllable and melodious music based on our innovative symbolic music representation, whereas the Selector identifies high-quality symbolic music by constructing a multi-task learning scheme involving quality assessment, emotion recognition, and genre recognition tasks. In addition, we build XMIDI, a large-scale symbolic music dataset that contains 108,023 MIDI files annotated with precise emotion and genre labels. Objective and subjective evaluations show that XMusic significantly outperforms the current state-of-the-art methods with impressive music quality. Our XMusic has been awarded as one of the nine Highlights of Collectibles at WAIC 2023. The project homepage of XMusic is https://xmusic-project.github.io.

Jing Jia, Wei Yuan, Sifan Liu et al.

Can a diffusion model trained on bedrooms recover human faces? Diffusion models are widely used as priors for inverse problems, but standard approaches usually assume a high-fidelity model trained on data that closely match the unknown signal. In practice, one often must use a mismatched or low-fidelity diffusion prior. Surprisingly, these weak priors often perform nearly as well as full-strength, in-domain baselines. We study when and why inverse solvers are robust to weak diffusion priors. Through extensive experiments, we find that weak priors succeed when measurements are highly informative (e.g., many observed pixels), and we identify regimes where they fail. Our theory, based on Bayesian consistency, gives conditions under which high-dimensional measurements make the posterior concentrate near the true signal. These results provide a principled justification on when weak diffusion priors can be used reliably.

Biao Yang, Bin Wen, Boyang Ding et al.

In recent years, the development of Large Language Models (LLMs) has significantly advanced, extending their capabilities to multimodal tasks through Multimodal Large Language Models (MLLMs). However, video understanding remains a challenging area due to the dynamic and information-dense nature of videos. Existing models struggle with the trade-off between spatial resolution and temporal coverage when processing video content. We present Keye-VL-1.5, which addresses fundamental challenges in video comprehension through three key innovations. First, we introduce a novel Slow-Fast video encoding strategy that dynamically allocates computational resources based on inter-frame similarity, processing key frames with significant visual changes at higher resolution (Slow pathway) while handling relatively static frames with increased temporal coverage at lower resolution (Fast pathway). Second, we implement a progressive four-stage pre-training methodology that systematically extends the model's context length from 8K to 128K tokens, enabling processing of longer videos and more complex visual content. Third, we develop a comprehensive post-training pipeline focusing on reasoning enhancement and human preference alignment, incorporating a 5-step chain-of-thought data construction process, iterative GSPO-based reinforcement learning with progressive prompt hinting for difficult cases, and alignment training. Through extensive evaluation on public benchmarks and rigorous internal human assessment, Keye-VL-1.5 demonstrates significant improvements over existing models, particularly excelling in video understanding tasks while maintaining competitive performance on general multimodal benchmarks.

Wei Yuan, Guanhua Ye, Xiangyu Zhao et al.

Time series forecasting plays a critical role in various real-world applications, including energy consumption prediction, disease transmission monitoring, and weather forecasting. Although substantial progress has been made in time series forecasting, most existing methods rely on a centralized training paradigm, where large amounts of data are collected from distributed devices (e.g., sensors, wearables) to a central cloud server. However, this paradigm has overloaded communication networks and raised privacy concerns. Federated learning, a popular privacy-preserving technique, enables collaborative model training across distributed data sources. However, directly applying federated learning to time series forecasting often yields suboptimal results, as time series data generated by different devices are inherently heterogeneous. In this paper, we propose a novel framework, Fed-TREND, to address data heterogeneity by generating informative synthetic data as auxiliary knowledge carriers. Specifically, Fed-TREND generates two types of synthetic data. The first type of synthetic data captures the representative distribution information from clients' uploaded model updates and enhances clients' local training consensus. The second kind of synthetic data extracts long-term influence insights from global model update trajectories and is used to refine the global model after aggregation. Fed-TREND is compatible with most time series forecasting models and can be seamlessly integrated into existing federated learning frameworks to improve prediction performance. Extensive experiments on eight datasets, using several federated learning baselines and four popular time series forecasting models, demonstrate the effectiveness and generalizability of Fed-TREND.

Shilong Yuan, Wei Yuan, Hongzhi Yin et al.

While language models have made many milestones in text inference and classification tasks, they remain susceptible to adversarial attacks that can lead to unforeseen outcomes. Existing works alleviate this problem by equipping language models with defense patches. However, these defense strategies often rely on impractical assumptions or entail substantial sacrifices in model performance. Consequently, enhancing the resilience of the target model using such defense mechanisms is a formidable challenge. This paper introduces an innovative model for robust text inference and classification, built upon diffusion models (ROIC-DM). Benefiting from its training involving denoising stages, ROIC-DM inherently exhibits greater robustness compared to conventional language models. Moreover, ROIC-DM can attain comparable, and in some cases, superior performance to language models, by effectively incorporating them as advisory components. Extensive experiments conducted with several strong textual adversarial attacks on three datasets demonstrate that (1) ROIC-DM outperforms traditional language models in robustness, even when the latter are fortified with advanced defense mechanisms; (2) ROIC-DM can achieve comparable and even better performance than traditional language models by using them as advisors.

Hongzhi You, Yijun Cao, Wei Yuan et al.

From a perspective of feature matching, optical flow estimation for event cameras involves identifying event correspondences by comparing feature similarity across accompanying event frames. In this work, we introduces an effective and robust high-dimensional (HD) feature descriptor for event frames, utilizing Vector Symbolic Architectures (VSA). The topological similarity among neighboring variables within VSA contributes to the enhanced representation similarity of feature descriptors for flow-matching points, while its structured symbolic representation capacity facilitates feature fusion from both event polarities and multiple spatial scales. Based on this HD feature descriptor, we propose a novel feature matching framework for event-based optical flow, encompassing both model-based (VSA-Flow) and self-supervised learning (VSA-SM) methods. In VSA-Flow, accurate optical flow estimation validates the effectiveness of HD feature descriptors. In VSA-SM, a novel similarity maximization method based on the HD feature descriptor is proposed to learn optical flow in a self-supervised way from events alone, eliminating the need for auxiliary grayscale images. Evaluation results demonstrate that our VSA-based method achieves superior accuracy in comparison to both model-based and self-supervised learning methods on the DSEC benchmark, while remains competitive among both methods on the MVSEC benchmark. This contribution marks a significant advancement in event-based optical flow within the feature matching methodology.

Haotian Lan, Yao Gao, Yujun Cheng et al.

Social media's rise establishes user-generated content (UGC) as pivotal for travel decisions, yet analytical methods lack scalability. This study introduces a dual-method LLM framework: unsupervised expectation extraction from UGC paired with survey-informed supervised fine-tuning. Findings reveal leisure/social expectations drive engagement more than foundational natural/emotional factors. By establishing LLMs as precision tools for expectation quantification, we advance tourism analytics methodology and propose targeted strategies for experience personalization and social travel promotion. The framework's adaptability extends to consumer behavior research, demonstrating computational social science's transformative potential in marketing optimization.

Bowen Song, Zecheng Zhang, Zhaoxu Luo et al.

Diffusion models have emerged as powerful tools for generative tasks, producing high-quality outputs across diverse domains. However, how the generated data responds to the initial noise perturbation in diffusion models remains under-explored, which hinders understanding the controllability of the sampling process. In this work, we first observe an interesting phenomenon: the relationship between the change of generation outputs and the scale of initial noise perturbation is highly linear through the diffusion ODE sampling. Then we provide both theoretical and empirical study to justify this linearity property of this input-output (noise-generation data) relationship. Inspired by these new insights, we propose a novel Controllable and Constrained Sampling method (CCS) together with a new controller algorithm for diffusion models to sample with desired statistical properties while preserving good sample quality. We perform extensive experiments to compare our proposed sampling approach with other methods on both sampling controllability and sampled data quality. Results show that our CCS method achieves more precisely controlled sampling while maintaining superior sample quality and diversity.

Qi Li, Wendong Huang, Qichen Ye et al.

The E-commerce advertising platforms typically sell commercial traffic through either second-price auction (SPA) or first-price auction (FPA). SPA was historically prevalent due to its dominant strategy incentive-compatible (DSIC) for bidders with quasi-linear utilities, especially when budgets are not a binding constraint, while FPA has gained more prominence for offering higher revenue potential to publishers and avoiding the possibility for discriminatory treatment in personalized reserve prices. Meanwhile, on the demand side, advertisers are increasingly adopting platform-wide marketing solutions akin to QuanZhanTui, shifting from spending budgets solely on commercial traffic to bidding on the entire traffic for the purpose of maximizing overall sales. For automated bidding systems, such a trend poses a critical challenge: determining optimal strategies across heterogeneous auction channels to fulfill diverse advertiser objectives, such as maximizing return (MaxReturn) or meeting target return on ad spend (TargetROAS). To overcome this challenge, this work makes two key contributions. First, we derive an efficient solution for optimal bidding under FPA channels, which takes into account the presence of organic traffic - traffic can be won for free. Second, we introduce a marginal cost alignment (MCA) strategy that provably secures bidding efficiency across heterogeneous auction mechanisms. To validate performance of our developed framework, we conduct comprehensive offline experiments on public datasets and large-scale online A/B testing, which demonstrate consistent improvements over existing methods.

Shutong Qiao, Wei Yuan, Junliang Yu et al.

Recommender systems (RSs) are now fundamental to various online platforms, but their dependence on user-contributed data leaves them vulnerable to shilling attacks that can manipulate item rankings by injecting fake users. Although widely studied, most existing attack models fail to meet two critical objectives simultaneously: achieving strong adversarial promotion of target items while maintaining realistic behavior to evade detection. As a result, the true severity of shilling threats that manage to reconcile the two objectives remains underappreciated. To expose this overlooked vulnerability, we present DLDA, a diffusion-based attack framework that can generate highly effective yet indistinguishable fake users by enabling fine-grained control over target promotion. Specifically, DLDA operates in a pre-aligned collaborative embedding space, where it employs a conditional latent diffusion process to iteratively synthesize fake user profiles with precise target item control. To evade detection, DLDA introduces a dispersive regularization mechanism that promotes variability and realism in generated behavioral patterns. Extensive experiments on three real-world datasets and five popular RS models demonstrate that, compared to prior attacks, DLDA consistently achieves stronger item promotion while remaining harder to detect. These results highlight that modern RSs are more vulnerable than previously recognized, underscoring the urgent need for more robust defenses.

Ruiqi Yang, Jianxu Wang, Wei Yuan et al.

This study explores the application of machine learning-based genetic linguistics for identifying heavy metal response genes in rice (Oryza sativa). By integrating convolutional neural networks and random forest algorithms, we developed a hybrid model capable of extracting and learning meaningful features from gene sequences, such as k-mer frequencies and physicochemical properties. The model was trained and tested on datasets of genes, achieving high predictive performance (precision: 0.89, F1-score: 0.82). RNA-seq and qRT-PCR experiments conducted on rice leaves which exposed to Hg0, revealed differential expression of genes associated with heavy metal responses, which validated the model's predictions. Co-expression network analysis identified 103 related genes, and a literature review indicated that these genes are highly likely to be involved in heavy metal-related biological processes. By integrating and comparing the analysis results with those of differentially expressed genes (DEGs), the validity of the new machine learning method was further demonstrated. This study highlights the efficacy of combining machine learning with genetic linguistics for large-scale gene prediction. It demonstrates a cost-effective and efficient approach for uncovering molecular mechanisms underlying heavy metal responses, with potential applications in developing stress-tolerant crop varieties.

Lingzhi Wang, Tong Chen, Wei Yuan et al.

Recent legislation of the "right to be forgotten" has led to the interest in machine unlearning, where the learned models are endowed with the function to forget information about specific training instances as if they have never existed in the training set. Previous work mainly focuses on computer vision scenarios and largely ignores the essentials of unlearning in NLP field, where text data contains more explicit and sensitive personal information than images. In this paper, we propose a general unlearning framework called KGA to induce forgetfulness. Different from previous work that tries to recover gradients or forces models to perform close to one specific distribution, KGA maintains distribution differences (i.e., knowledge gap). This relaxes the distribution assumption. Furthermore, we first apply the unlearning method to various NLP tasks (i.e., classification, translation, response generation) and propose several unlearning evaluation metrics with pertinence. Experiments on large-scale datasets show that KGA yields comprehensive improvements over baselines, where extensive analyses further validate the effectiveness of KGA and provide insight into unlearning for NLP tasks.

Zhibo Wang, He Wang, Shuaifan Jin et al.

Face recognition service providers protect face privacy by extracting compact and discriminative facial features (representations) from images, and storing the facial features for real-time recognition. However, such features can still be exploited to recover the appearance of the original face by building a reconstruction network. Although several privacy-preserving methods have been proposed, the enhancement of face privacy protection is at the expense of accuracy degradation. In this paper, we propose an adversarial features-based face privacy protection (AdvFace) approach to generate privacy-preserving adversarial features, which can disrupt the mapping from adversarial features to facial images to defend against reconstruction attacks. To this end, we design a shadow model which simulates the attackers' behavior to capture the mapping function from facial features to images and generate adversarial latent noise to disrupt the mapping. The adversarial features rather than the original features are stored in the server's database to prevent leaked features from exposing facial information. Moreover, the AdvFace requires no changes to the face recognition network and can be implemented as a privacy-enhancing plugin in deployed face recognition systems. Extensive experimental results demonstrate that AdvFace outperforms the state-of-the-art face privacy-preserving methods in defending against reconstruction attacks while maintaining face recognition accuracy.

Beihao Xia, Conghao Wong, Qinmu Peng et al.

Trajectory prediction aims to predict the movement trend of the agents like pedestrians, bikers, vehicles. It is helpful to analyze and understand human activities in crowded spaces and widely applied in many areas such as surveillance video analysis and autonomous driving systems. Thanks to the success of deep learning, trajectory prediction has made significant progress. The current methods are dedicated to studying the agents' future trajectories under the social interaction and the sceneries' physical constraints. Moreover, how to deal with these factors still catches researchers' attention. However, they ignore the \textbf{Semantic Shift Phenomenon} when modeling these interactions in various prediction sceneries. There exist several kinds of semantic deviations inner or between social and physical interactions, which we call the "\textbf{Gap}". In this paper, we propose a \textbf{C}ontextual \textbf{S}emantic \textbf{C}onsistency \textbf{Net}work (\textbf{CSCNet}) to predict agents' future activities with powerful and efficient context constraints. We utilize a well-designed context-aware transfer to obtain the intermediate representations from the scene images and trajectories. Then we eliminate the differences between social and physical interactions by aligning activity semantics and scene semantics to cross the Gap. Experiments demonstrate that CSCNet performs better than most of the current methods quantitatively and qualitatively.

Wei Yuan, Hongzhi Yin, Tieke He et al.

Question Generation (QG), as a challenging Natural Language Processing task, aims at generating questions based on given answers and context. Existing QG methods mainly focus on building or training models for specific QG datasets. These works are subject to two major limitations: (1) They are dedicated to specific QG formats (e.g., answer-extraction or multi-choice QG), therefore, if we want to address a new format of QG, a re-design of the QG model is required. (2) Optimal performance is only achieved on the dataset they were just trained on. As a result, we have to train and keep various QG models for different QG datasets, which is resource-intensive and ungeneralizable. To solve the problems, we propose a model named Unified-QG based on lifelong learning techniques, which can continually learn QG tasks across different datasets and formats. Specifically, we first build a format-convert encoding to transform different kinds of QG formats into a unified representation. Then, a method named \emph{STRIDER} (\emph{S}imilari\emph{T}y \emph{R}egular\emph{I}zed \emph{D}ifficult \emph{E}xample \emph{R}eplay) is built to alleviate catastrophic forgetting in continual QG learning. Extensive experiments were conducted on $8$ QG datasets across $4$ QG formats (answer-extraction, answer-abstraction, multi-choice, and boolean QG) to demonstrate the effectiveness of our approach. Experimental results demonstrate that our Unified-QG can effectively and continually adapt to QG tasks when datasets and formats vary. In addition, we verify the ability of a single trained Unified-QG model in improving $8$ Question Answering (QA) systems' performance through generating synthetic QA data.

Conghao Wong, Beihao Xia, Ziming Hong et al.

Understanding and forecasting future trajectories of agents are critical for behavior analysis, robot navigation, autonomous cars, and other related applications. Previous methods mostly treat trajectory prediction as time sequence generation. Different from them, this work studies agents' trajectories in a "vertical" view, i.e., modeling and forecasting trajectories from the spectral domain. Different frequency bands in the trajectory spectrums could hierarchically reflect agents' motion preferences at different scales. The low-frequency and high-frequency portions could represent their coarse motion trends and fine motion variations, respectively. Accordingly, we propose a hierarchical network V$^2$-Net, which contains two sub-networks, to hierarchically model and predict agents' trajectories with trajectory spectrums. The coarse-level keypoints estimation sub-network first predicts the "minimal" spectrums of agents' trajectories on several "key" frequency portions. Then the fine-level spectrum interpolation sub-network interpolates the spectrums to reconstruct the final predictions. Experimental results display the competitiveness and superiority of V$^2$-Net on both ETH-UCY benchmark and the Stanford Drone Dataset.

Conghao Wong, Beihao Xia, Qinmu Peng et al.

Trajectory prediction aims to forecast agents' possible future locations considering their observations along with the video context. It is strongly needed by many autonomous platforms like tracking, detection, robot navigation, and self-driving cars. Whether it is agents' internal personality factors, interactive behaviors with the neighborhood, or the influence of surroundings, they all impact agents' future planning. However, many previous methods model and predict agents' behaviors with the same strategy or feature distribution, making them challenging to make predictions with sufficient style differences. This paper proposes the Multi-Style Network (MSN), which utilizes style proposal and stylized prediction using two sub-networks, to provide multi-style predictions in a novel categorical way adaptively. The proposed network contains a series of style channels, and each channel is bound to a unique and specific behavior style. We use agents' end-point plannings and their interaction context as the basis for the behavior classification, so as to adaptively learn multiple diverse behavior styles through these channels. Then, we assume that the target agents may plan their future behaviors according to each of these categorized styles, thus utilizing different style channels to make predictions with significant style differences in parallel. Experiments show that the proposed MSN outperforms current state-of-the-art methods up to 10% quantitatively on two widely used datasets, and presents better multi-style characteristics qualitatively.

Shiming Chen, Peng Zhang, Guo-Sen Xie et al.

Dynamic texture (DT) exhibits statistical stationarity in the spatial domain and stochastic repetitiveness in the temporal dimension, indicating that different frames of DT possess a high similarity correlation that is critical prior knowledge. However, existing methods cannot effectively learn a promising synthesis model for high-dimensional DT from a small number of training data. In this paper, we propose a novel DT synthesis method, which makes full use of similarity prior knowledge to address this issue. Our method bases on the proposed kernel similarity embedding, which not only can mitigate the high-dimensionality and small sample issues, but also has the advantage of modeling nonlinear feature relationship. Specifically, we first raise two hypotheses that are essential for DT model to generate new frames using similarity correlation. Then, we integrate kernel learning and extreme learning machine into a unified synthesis model to learn kernel similarity embedding for representing DT. Extensive experiments on DT videos collected from the internet and two benchmark datasets, i.e., Gatech Graphcut Textures and Dyntex, demonstrate that the learned kernel similarity embedding can effectively exhibit the discriminative representation for DT. Accordingly, our method is capable of preserving the long-term temporal continuity of the synthesized DT sequences with excellent sustainability and generalization. Meanwhile, it effectively generates realistic DT videos with fast speed and low computation, compared with the state-of-the-art methods. The code and more synthesis videos are available at our project page https://shiming-chen.github.io/Similarity-page/Similarit.html.

Zhiling Guo, Hiroaki Shengoku, Guangming Wu et al.

The automatic digitizing of paper maps is a significant and challenging task for both academia and industry. As an important procedure of map digitizing, the semantic segmentation section mainly relies on manual visual interpretation with low efficiency. In this study, we select urban planning maps as a representative sample and investigate the feasibility of utilizing U-shape fully convolutional based architecture to perform end-to-end map semantic segmentation. The experimental results obtained from the test area in Shibuya district, Tokyo, demonstrate that our proposed method could achieve a very high Jaccard similarity coefficient of 93.63% and an overall accuracy of 99.36%. For implementation on GPGPU and cuDNN, the required processing time for the whole Shibuya district can be less than three minutes. The results indicate the proposed method can serve as a viable tool for urban planning map semantic segmentation task with high accuracy and efficiency.

Xinge You, Jiamiao Xu, Wei Yuan et al.

Cross-view classification that means to classify samples from heterogeneous views is a significant yet challenging problem in computer vision. A promising approach to handle this problem is the multi-view subspace learning (MvSL), which intends to find a common subspace for multi-view data. Despite the satisfactory results achieved by existing methods, the performance of previous work will be dramatically degraded when multi-view data lies on nonlinear manifolds. To circumvent this drawback, we propose Multi-view Common Component Discriminant Analysis (MvCCDA) to handle view discrepancy, discriminability and nonlinearity in a joint manner. Specifically, our MvCCDA incorporates supervised information and local geometric information into the common component extraction process to learn a discriminant common subspace and to discover the nonlinear structure embedded in multi-view data. We develop a kernel method of MvCCDA to further boost the performance of MvCCDA. Beyond kernel extension, optimization and complexity analysis of MvCCDA are also presented for completeness. Our MvCCDA is competitive with the state-of-the-art MvSL based methods on four benchmark datasets, demonstrating its superiority.

Zi Wang, Wei Yuan, Giovanni Montana

Gene expression levels in a population vary extensively across tissues. Such heterogeneity is caused by genetic variability and environmental factors, and is expected to be linked to disease development. The abundance of experimental data now enables the identification of features of gene expression profiles that are shared across tissues, and those that are tissue-specific. While most current research is concerned with characterising differential expression by comparing mean expression profiles across tissues, it is also believed that a significant difference in a gene expression's variance across tissues may also be associated to molecular mechanisms that are important for tissue development and function. We propose a sparse multi-view matrix factorisation (sMVMF) algorithm to jointly analyse gene expression measurements in multiple tissues, where each tissue provides a different "view" of the underlying organism. The proposed methodology can be interpreted as an extension of principal component analysis in that it provides the means to decompose the total sample variance in each tissue into the sum of two components: one capturing the variance that is shared across tissues, and one isolating the tissue-specific variances. sMVMF has been used to jointly model mRNA expression profiles in three tissues - adipose, skin and LCL - which are available for a large and well-phenotyped twins cohort, TwinsUK. Using sMVMF, we are able to prioritise genes based on whether their variation patterns are specific to each tissue. Furthermore, using DNA methylation profiles available, we provide supporting evidence that adipose-specific gene expression patterns may be driven by epigenetic effects.