Ziyi Zhao

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.

Ziyi Zhao, Sena Kiciroglu, Hugues Vinzant et al.

Unsupervised self-rehabilitation exercises and physical training can cause serious injuries if performed incorrectly. We introduce a learning-based framework that identifies the mistakes made by a user and proposes corrective measures for easier and safer individual training. Our framework does not rely on hard-coded, heuristic rules. Instead, it learns them from data, which facilitates its adaptation to specific user needs. To this end, we use a Graph Convolutional Network (GCN) architecture acting on the user's pose sequence to model the relationship between the body joints trajectories. To evaluate our approach, we introduce a dataset with 3 different physical exercises. Our approach yields 90.9% mistake identification accuracy and successfully corrects 94.2% of the mistakes.

Lei Wang, Ziyi Zhao, Hanwei Liu et al.

With the introduction of ChatGPT, the public's perception of AI-generated content (AIGC) has begun to reshape. Artificial intelligence has significantly reduced the barrier to entry for non-professionals in creative endeavors, enhancing the efficiency of content creation. Recent advancements have seen significant improvements in the quality of symbolic music generation, which is enabled by the use of modern generative algorithms to extract patterns implicit in a piece of music based on rule constraints or a musical corpus. Nevertheless, existing literature reviews tend to present a conventional and conservative perspective on future development trajectories, with a notable absence of thorough benchmarking of generative models. This paper provides a survey and analysis of recent intelligent music generation techniques, outlining their respective characteristics and discussing existing methods for evaluation. Additionally, the paper compares the different characteristics of music generation techniques in the East and West as well as analysing the field's development prospects.

Ziyi Zhao, Chen Luo, Jin Zhao et al.

The number of daily small Unmanned Aircraft Systems (sUAS) operations in uncontrolled low altitude airspace is expected to reach into the millions. UAS Traffic Management (UTM) is an emerging concept aiming at the safe and efficient management of such very dense traffic, but few studies are addressing the policies to accommodate such demand and the required ground infrastructure in suburban or urban environments. Searching for the optimal air traffic management policy is a combinatorial optimization problem with intractable complexity when the number of sUAS and the constraints increases. As the demands on the airspace increase and traffic patterns get complicated, it is difficult to forecast the potential low altitude airspace hotspots and the corresponding ground resource requirements. This work presents a Multi-agent Air Traffic and Resource Usage Simulation (MATRUS) framework that aims for fast evaluation of different air traffic management policies and the relationship between policy, environment and resulting traffic patterns. It can also be used as a tool to decide the resource distribution and launch site location in the planning of a next-generation smart city. As a case study, detailed comparisons are provided for the sUAS flight time, conflict ratio, cellular communication resource usage, for a managed (centrally coordinated) and unmanaged (free flight) traffic scenario.

Qiuyang Mang, Wenhao Chai, Zhifei Li et al.

We introduce FrontierCS, a benchmark of 156 open-ended problems across diverse areas of computer science, designed and reviewed by experts, including CS PhDs and top-tier competitive programming participants and problem setters. Unlike existing benchmarks that focus on tasks with known optimal solutions, FrontierCS targets problems where the optimal solution is unknown, but the quality of a solution can be objectively evaluated. Models solve these tasks by implementing executable programs rather than outputting a direct answer. FrontierCS includes algorithmic problems, which are often NP-hard variants of competitive programming problems with objective partial scoring, and research problems with the same property. For each problem we provide an expert reference solution and an automatic evaluator. Combining open-ended design, measurable progress, and expert curation, FrontierCS provides a benchmark at the frontier of computer-science difficulty. Empirically, we find that frontier reasoning models still lag far behind human experts on both the algorithmic and research tracks, that increasing reasoning budgets alone does not close this gap, and that models often over-optimize for generating merely workable code instead of discovering high-quality algorithms and system designs.

Yingjie Zhou, Xilei Zhu, Siyu Ren et al.

Speech-driven Talking Human (TH) generation, commonly known as "Talker," currently faces limitations in multi-subject driving capabilities. Extending this paradigm to "Multi-Talker," capable of animating multiple subjects simultaneously, introduces richer interactivity and stronger immersion in audiovisual communication. However, current Multi-Talkers still exhibit noticeable quality degradation caused by technical limitations, resulting in suboptimal user experiences. To address this challenge, we construct THQA-MT, the first large-scale Multi-Talker-generated Talking Human Quality Assessment dataset, consisting of 5,492 Multi-Talker-generated THs (MTHs) from 15 representative Multi-Talkers using 400 real portraits collected online. Through subjective experiments, we analyze perceptual discrepancies among different Multi-Talkers and identify 12 common types of distortion. Furthermore, we introduce EvalTalker, a novel TH quality assessment framework. This framework possesses the ability to perceive global quality, human characteristics, and identity consistency, while integrating Qwen-Sync to perceive multimodal synchrony. Experimental results demonstrate that EvalTalker achieves superior correlation with subjective scores, providing a robust foundation for future research on high-quality Multi-Talker generation and evaluation.

Zhongping Ji, Feifei Che, Hanshuo Liu et al.

In this paper, we propose a technique for making humans in photographs protrude like reliefs. Unlike previous methods which mostly focus on the face and head, our method aims to generate art works that describe the whole body activity of the character. One challenge is that there is no ground-truth for supervised deep learning. We introduce a sigmoid variant function to manipulate gradients tactfully and train our neural networks by equipping with a loss function defined in gradient domain. The second challenge is that actual photographs often across different light conditions. We used image-based rendering technique to address this challenge and acquire rendering images and depth data under different lighting conditions. To make a clear division of labor in network modules, a two-scale architecture is proposed to create high-quality relief from a single photograph. Extensive experimental results on a variety of scenes show that our method is a highly effective solution for generating digital 2.5D artwork from photographs.

Jinzhao Zhou, Yiqun Duan, Ziyi Zhao et al.

Decoding linguistic information from non-invasive brain signals using EEG has gained increasing research attention due to its vast applicational potential. Recently, a number of works have adopted a generative-based framework to decode electroencephalogram (EEG) signals into sentences by utilizing the power generative capacity of pretrained large language models (LLMs). However, this approach has several drawbacks that hinder the further development of linguistic applications for brain-computer interfaces (BCIs). Specifically, the ability of the EEG encoder to learn semantic information from EEG data remains questionable, and the LLM decoder's tendency to generate sentences based on its training memory can be hard to avoid. These issues necessitate a novel approach for converting EEG signals into sentences. In this paper, we propose a novel two-step pipeline that addresses these limitations and enhances the validity of linguistic EEG decoding research. We first confirm that word-level semantic information can be learned from EEG data recorded during natural reading by training a Conformer encoder via a masked contrastive objective for word-level classification. To achieve sentence decoding results, we employ a training-free retrieval method to retrieve sentences based on the predictions from the EEG encoder. Extensive experiments and ablation studies were conducted in this paper for a comprehensive evaluation of the proposed approach. Visualization of the top prediction candidates reveals that our model effectively groups EEG segments into semantic categories with similar meanings, thereby validating its ability to learn patterns from unspoken EEG recordings. Despite the exploratory nature of this work, these results suggest that our method holds promise for providing more reliable solutions for converting EEG signals into text.

Meituan LongCat Team, Bairui Wang, Bayan et al.

We introduce LongCat-Flash-Omni, a state-of-the-art open-source omni-modal model with 560 billion parameters, excelling at real-time audio-visual interaction. By adopting a curriculum-inspired progressive training strategy that transitions from simpler to increasingly complex modality sequence modeling tasks, LongCat-Flash-Omni attains comprehensive multimodal capabilities while maintaining strong unimodal capability. Building upon LongCat-Flash, which adopts a high-performance Shortcut-connected Mixture-of-Experts (MoE) architecture with zero-computation experts, LongCat-Flash-Omni integrates efficient multimodal perception and speech reconstruction modules. Despite its immense size of 560B parameters (with 27B activated), LongCat-Flash-Omni achieves low-latency real-time audio-visual interaction. For training infrastructure, we developed a modality-decoupled parallelism scheme specifically designed to manage the data and model heterogeneity inherent in large-scale multimodal training. This innovative approach demonstrates exceptional efficiency by sustaining over 90% of the throughput achieved by text-only training. Extensive evaluations show that LongCat-Flash-Omni achieves state-of-the-art performance on omni-modal benchmarks among open-source models. Furthermore, it delivers highly competitive results across a wide range of modality-specific tasks, including text, image, and video understanding, as well as audio understanding and generation. We provide a comprehensive overview of the model architecture design, training procedures, and data strategies, and open-source the model to foster future research and development in the community.

Ziyi Zhao, Jinzhao Zhou, Xiaowei Jiang et al.

Decoding linguistic information from electroencephalography (EEG) remains challenging due to the brain's distributed and nonlinear organization. We present BrainStack, a functionally guided neuro-mixture-of-experts (Neuro-MoE) framework that models the brain's modular functional architecture through anatomically partitioned expert networks. Each functional region is represented by a specialized expert that learns localized neural dynamics, while a transformer-based global expert captures cross-regional dependencies. A learnable routing gate adaptively aggregates these heterogeneous experts, enabling context-dependent expert coordination and selective fusion. To promote coherent representation across the hierarchy, we introduce cross-regional distillation, where the global expert provides top-down regularization to the regional experts. We further release SilentSpeech-EEG (SS-EEG), a large-scale benchmark comprising over 120 hours of EEG recordings from 12 subjects performing 24 silent words, the largest dataset of its kind. Experiments demonstrate that BrainStack consistently outperforms state-of-the-art models, achieving superior accuracy and generalization across subjects. Our results establish BrainStack as a functionally modular, neuro-inspired MoE paradigm that unifies neuroscientific priors with adaptive expert routing, paving the way for scalable and interpretable brain-language decoding.

Abrar Majeedi, Zhiyuan Ruan, Ziyi Zhao et al.

Multimodal large language models (MLLMs) have achieved impressive performance on visual perception and reasoning tasks with RGB imagery, yet they remain fragile under common degradations, such as fog, blur, or low-light conditions. Infrared (IR) imaging, a well-established complement to RGB, offers inherent robustness in these conditions, but its integration into MLLMs remains underexplored. To bridge this gap, we propose DUALVISION, a lightweight fusion module that efficiently incorporates IR-RGB information into MLLMs via patch-level localized cross-attention. To support training and evaluation and to facilitate future research, we also introduce DV-204K, a dataset of ~25K publicly available aligned IR-RGB image pairs with 204K modality-specific QA annotations, and DV-500, a benchmark of 500 IR-RGB image pairs with 500 QA pairs designed for evaluating cross-modal reasoning. Leveraging these datasets, we benchmark both open- and closed-source MLLMs and demonstrate that DUALVISION delivers strong empirical performance under a wide range of visual degradations. Our code and dataset are available at https://abrarmajeedi.github.io/dualvision.

Jianglin Lu, Yuanwei Wu, Ziyi Zhao et al.

Complex image restoration aims to recover high-quality images from inputs affected by multiple degradations such as blur, noise, rain, and compression artifacts. Recent restoration agents, powered by vision-language models and large language models, offer promising restoration capabilities but suffer from significant efficiency bottlenecks due to reflection, rollback, and iterative tool searching. Moreover, their performance heavily depends on degradation recognition models that require extensive annotations for training, limiting their applicability in label-free environments. To address these limitations, we propose a policy optimization-based restoration framework that learns an lightweight agent to determine tool-calling sequences. The agent operates in a sequential decision process, selecting the most appropriate restoration operation at each step to maximize final image quality. To enable training within label-free environments, we introduce a novel reward mechanism driven by multimodal large language models, which act as human-aligned evaluator and provide perceptual feedback for policy improvement. Once trained, our agent executes a deterministic restoration plans without redundant tool invocations, significantly accelerating inference while maintaining high restoration quality. Extensive experiments show that despite using no supervision, our method matches SOTA performance on full-reference metrics and surpasses existing approaches on no-reference metrics across diverse degradation scenarios.

Jinzhao Zhou, Zehong Cao, Yiqun Duan et al.

This paper explores silent speech decoding in active brain-computer interface (BCI) systems, which offer more natural and flexible communication than traditional BCI applications. We collected a new silent speech dataset of over 120 hours of electroencephalogram (EEG) recordings from 12 subjects, capturing 24 commonly used English words for language model pretraining and decoding. Following the recent success of pretraining large models with self-supervised paradigms to enhance EEG classification performance, we propose Large Brain Language Model (LBLM) pretrained to decode silent speech for active BCI. To pretrain LBLM, we propose Future Spectro-Temporal Prediction (FSTP) pretraining paradigm to learn effective representations from unlabeled EEG data. Unlike existing EEG pretraining methods that mainly follow a masked-reconstruction paradigm, our proposed FSTP method employs autoregressive modeling in temporal and frequency domains to capture both temporal and spectral dependencies from EEG signals. After pretraining, we finetune our LBLM on downstream tasks, including word-level and semantic-level classification. Extensive experiments demonstrate significant performance gains of the LBLM over fully-supervised and pretrained baseline models. For instance, in the difficult cross-session setting, our model achieves 47.0\% accuracy on semantic-level classification and 39.6\% in word-level classification, outperforming baseline methods by 5.4\% and 7.3\%, respectively. Our research advances silent speech decoding in active BCI systems, offering an innovative solution for EEG language model pretraining and a new dataset for fundamental research.

Yang Shen, Zhenyi Yi, Ziyi Zhao et al.

As AI agents evolve, the community is rapidly shifting from single Large Language Models (LLMs) to Multi-Agent Systems (MAS) to overcome cognitive bottlenecks in automated research. However, the optimal multi-agent coordination framework for these autonomous agents remains largely unexplored. In this paper, we present a systematic empirical study investigating the comparative efficacy of distinct multi-agent structures for automated machine learning optimization. Utilizing a rigorously controlled, execution-based testbed equipped with Git worktree isolation and explicit global memory, we benchmark a single-agent baseline against two multi-agent paradigms: a subagent architecture (parallel exploration with post-hoc consolidation) and an agent team architecture (experts with pre-execution handoffs). By evaluating these systems under strictly fixed computational time budgets, our findings reveal a fundamental trade-off between operational stability and theoretical deliberation. The subagent mode functions as a highly resilient, high-throughput search engine optimal for broad, shallow optimizations under strict time constraints. Conversely, the agent team topology exhibits higher operational fragility due to multi-author code generation but achieves the deep theoretical alignment necessary for complex architectural refactoring given extended compute budgets. These empirical insights provide actionable guidelines for designing future autoresearch systems, advocating for dynamically routed architectures that adapt their collaborative structures to real-time task complexity.

Xiaowei Jiang, Charles Zhou, Yiqun Duan et al.

Brain-computer interfaces (BCIs) present a promising avenue by translating neural activity directly into text, eliminating the need for physical actions. However, existing non-invasive BCI systems have not successfully covered the entire alphabet, limiting their practicality. In this paper, we propose a novel non-invasive EEG-based BCI system with Curriculum-based Neural Spelling Framework, which recognizes all 26 alphabet letters by decoding neural signals associated with handwriting first, and then apply a Generative AI (GenAI) to enhance spell-based neural language decoding tasks. Our approach combines the ease of handwriting with the accessibility of EEG technology, utilizing advanced neural decoding algorithms and pre-trained large language models (LLMs) to translate EEG patterns into text with high accuracy. This system show how GenAI can improve the performance of typical spelling-based neural language decoding task, and addresses the limitations of previous methods, offering a scalable and user-friendly solution for individuals with communication impairments, thereby enhancing inclusive communication options.

Xinkai Wang, Shuo Zhang, Ziyi Zhao et al.

While tangible user interface has shown its power in naturally interacting with rigid or soft objects, users cannot conveniently use different types of granular materials as the interaction media. We introduce DipMe as a smart device to recognize the types of granular media in real time, which can be used to connect the granular materials in the physical world with various virtual content. Other than vision-based solutions, we propose a dip operation of our device and exploit the haptic signals to recognize different types of granular materials. With modern machine learning tools, we find the haptic signals from different granular media are distinguishable by DipMe. With the online granular object recognition, we build several tangible interactive applications, demonstrating the effects of DipMe in perceiving granular materials and its potential in developing a tangible user interface with granular objects as the new media.

Ziyi Zhao, Qingchuan Li, Yuxuan Xu

Path signatures embed trajectories into tensor algebra and constitute a universal, non-parametric representation of paths; however, in the standard form, they collapse temporal structure into a single global object, which limits their suitability for decision-making problems that require step-wise reactivity. We propose the Incremental Signature Contribution (ISC) method, which decomposes truncated path signatures into a temporally ordered sequence of elements in the tensor-algebra space, corresponding to incremental contributions induced by last path increments. This reconstruction preserves the algebraic structure and expressivity of signatures, while making their internal temporal evolution explicit, enabling processing signature-based representations via sequential modeling approaches. In contrast to full signatures, ISC is inherently sensitive to instantaneous trajectory updates, which is critical for sensitive and stability-requiring control dynamics. Building on this representation, we introduce ISC-Transformer (ISCT), an offline reinforcement learning model that integrates ISC into a standard Transformer architecture without further architectural modification. We evaluate ISCT on HalfCheetah, Walker2d, Hopper, and Maze2d, including settings with delayed rewards and downgraded datasets. The results demonstrate that ISC method provides a theoretically grounded and practically effective alternative to path processing for temporally sensitive control tasks.

Ziyi Zhao, Haowen Fang, Zhao Jin et al.

The trajectory prediction is a critical and challenging problem in the design of an autonomous driving system. Many AI-oriented companies, such as Google Waymo, Uber and DiDi, are investigating more accurate vehicle trajectory prediction algorithms. However, the prediction performance is governed by lots of entangled factors, such as the stochastic behaviors of surrounding vehicles, historical information of self-trajectory, and relative positions of neighbors, etc. In this paper, we propose a novel graph-based information sharing network (GISNet) that allows the information sharing between the target vehicle and its surrounding vehicles. Meanwhile, the model encodes the historical trajectory information of all the vehicles in the scene. Experiments are carried out on the public NGSIM US-101 and I-80 Dataset and the prediction performance is measured by the Root Mean Square Error (RMSE). The quantitative and qualitative experimental results show that our model significantly improves the trajectory prediction accuracy, by up to 50.00%, compared to existing models.

Ziyi Zhao, Zhao Jin, Wentian Bai et al.

The number of daily sUAS operations in uncontrolled low altitude airspace is expected to reach into the millions in a few years. Therefore, UAS density prediction has become an emerging and challenging problem. In this paper, a deep learning-based UAS instantaneous density prediction model is presented. The model takes two types of data as input: 1) the historical density generated from the historical data, and 2) the future sUAS mission information. The architecture of our model contains four components: Historical Density Formulation module, UAS Mission Translation module, Mission Feature Extraction module, and Density Map Projection module. The training and testing data are generated by a python based simulator which is inspired by the multi-agent air traffic resource usage simulator (MATRUS) framework. The quality of prediction is measured by the correlation score and the Area Under the Receiver Operating Characteristics (AUROC) between the predicted value and simulated value. The experimental results demonstrate outstanding performance of the deep learning-based UAS density predictor. Compared to the baseline models, for simplified traffic scenario where no-fly zones and safe distance among sUASs are not considered, our model improves the prediction accuracy by more than 15.2% and its correlation score reaches 0.947. In a more realistic scenario, where the no-fly zone avoidance and the safe distance among sUASs are maintained using A* routing algorithm, our model can still achieve 0.823 correlation score. Meanwhile, the AUROC can reach 0.951 for the hot spot prediction.

Haowen Fang, Amar Shrestha, Ziyi Zhao et al.

The recent discovered spatial-temporal information processing capability of bio-inspired Spiking neural networks (SNN) has enabled some interesting models and applications. However designing large-scale and high-performance model is yet a challenge due to the lack of robust training algorithms. A bio-plausible SNN model with spatial-temporal property is a complex dynamic system. Each synapse and neuron behave as filters capable of preserving temporal information. As such neuron dynamics and filter effects are ignored in existing training algorithms, the SNN downgrades into a memoryless system and loses the ability of temporal signal processing. Furthermore, spike timing plays an important role in information representation, but conventional rate-based spike coding models only consider spike trains statistically, and discard information carried by its temporal structures. To address the above issues, and exploit the temporal dynamics of SNNs, we formulate SNN as a network of infinite impulse response (IIR) filters with neuron nonlinearity. We proposed a training algorithm that is capable to learn spatial-temporal patterns by searching for the optimal synapse filter kernels and weights. The proposed model and training algorithm are applied to construct associative memories and classifiers for synthetic and public datasets including MNIST, NMNIST, DVS 128 etc.; and their accuracy outperforms state-of-art approaches.

Ziyi Zhao, Krittaphat Pugdeethosapol, Sheng Lin et al.

The topic modeling discovers the latent topic probability of the given text documents. To generate the more meaningful topic that better represents the given document, we proposed a new feature extraction technique which can be used in the data preprocessing stage. The method consists of three steps. First, it generates the word/word-pair from every single document. Second, it applies a two-way TF-IDF algorithm to word/word-pair for semantic filtering. Third, it uses the K-means algorithm to merge the word pairs that have the similar semantic meaning. Experiments are carried out on the Open Movie Database (OMDb), Reuters Dataset and 20NewsGroup Dataset. The mean Average Precision score is used as the evaluation metric. Comparing our results with other state-of-the-art topic models, such as Latent Dirichlet allocation and traditional Restricted Boltzmann Machines. Our proposed data preprocessing can improve the generated topic accuracy by up to 12.99\%.

Wenya Zhu, Xiankai Lu, Tao Xu et al.

Structure information is ubiquitous in natural scene images and it plays an important role in scene representation. In this paper, third order structure statistics (TOSS) and fourth order structure statistics (FOSS) are exploited to encode higher order structure information. Afterwards, based on the radial and normal slice of TOSS and FOSS, we propose the high order structure feature: third order structure feature (TOSF) and fourth order structure feature (FOSF). It is well known that scene images are well characterized by particular arrangements of their local structures, we divide the scene image into the non-overlapping sub-regions and compute the proposed higher order structural features among them. Then a scene classification is performed by using SVM classifier with these higher order structure features. The experimental results show that higher order structure statistics can deliver image structure information well and its spatial envelope has strong discriminative ability.