arXiv subjects

Chenhao Xu

Publications and source records attributed to Chenhao Xu.

16 recordsLinked to original sources

arXiv2026

DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence

We present a preview version of DeepSeek-V4 series, including two strong Mixture-of-Experts (MoE) language models -- DeepSeek-V4-Pro with 1.6T parameters (49B activated) and DeepSeek-V4-Flash with 284B parameters (13B activated) -- both supporting a context length of one million tokens. DeepSeek-V4 series incorporate several key upgrades in architecture and optimization: (1) a hybrid attention architecture that combines Compressed Sparse Attention (CSA) and Heavily Compressed Attention (HCA) to improve long-context efficiency; (2) Manifold-Constrained Hyper-Connections (mHC) that enhance conventional residual connections; (3) and the Muon optimizer for faster convergence and greater training stability. We pre-train both models on more than 32T diverse and high-quality tokens, followed by a comprehensive post-training pipeline that unlocks and further enhances their capabilities. DeepSeek-V4-Pro-Max, the maximum reasoning effort mode of DeepSeek-V4-Pro, redefines the state-of-the-art for open models, outperforming its predecessors in core tasks. Meanwhile, DeepSeek-V4 series are highly efficient in long-context scenarios. In the one-million-token context setting, DeepSeek-V4-Pro requires only 27% of single-token inference FLOPs and 10% of KV cache compared with DeepSeek-V3.2. This enables us to routinely support one-million-token contexts, thereby making long-horizon tasks and further test-time scaling more feasible. The model checkpoints are available at https://huggingface.co/collections/deepseek-ai/deepseek-v4.

cs.CL↗

arXiv2025

DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models

We introduce DeepSeek-V3.2, a model that harmonizes high computational efficiency with superior reasoning and agent performance. The key technical breakthroughs of DeepSeek-V3.2 are as follows: (1) DeepSeek Sparse Attention (DSA): We introduce DSA, an efficient attention mechanism that substantially reduces computational complexity while preserving model performance in long-context scenarios. (2) Scalable Reinforcement Learning Framework: By implementing a robust reinforcement learning protocol and scaling post-training compute, DeepSeek-V3.2 performs comparably to GPT-5. Notably, our high-compute variant, DeepSeek-V3.2-Speciale, surpasses GPT-5 and exhibits reasoning proficiency on par with Gemini-3.0-Pro, achieving gold-medal performance in both the 2025 International Mathematical Olympiad (IMO) and the International Olympiad in Informatics (IOI). (3) Large-Scale Agentic Task Synthesis Pipeline: To integrate reasoning into tool-use scenarios, we developed a novel synthesis pipeline that systematically generates training data at scale. This methodology facilitates scalable agentic post-training, yielding substantial improvements in generalization and instruction-following robustness within complex, interactive environments.

cs.CL↗

arXiv2025

Efficient Learning-based Graph Simulation for Temporal Graphs

Graph simulation has recently received a surge of attention in graph processing and analytics. In real-life applications, e.g. social science, biology, and chemistry, many graphs are composed of a series of evolving graphs (i.e., temporal graphs). While most of the existing graph generators focus on static graphs, the temporal information of the graphs is ignored. In this paper, we focus on simulating temporal graphs, which aim to reproduce the structural and temporal properties of the observed real-life temporal graphs. In this paper, we first give an overview of the existing temporal graph generators, including recently emerged learning-based approaches. Most of these learning-based methods suffer from one of the limitations: low efficiency in training or slow generating, especially for temporal random walk-based methods. Therefore, we propose an efficient learning-based approach to generate graph snapshots, namely temporal graph autoencoder (TGAE). Specifically, we propose an attention-based graph encoder to encode temporal and structural characteristics on sampled ego-graphs. And we proposed an ego-graph decoder that can achieve a good trade-off between simulation quality and efficiency in temporal graph generation. Finally, the experimental evaluation is conducted among our proposed TGAE and representative temporal graph generators on real-life temporal graphs and synthesized graphs. It is reported that our proposed approach outperforms the state-of-the-art temporal graph generators by means of simulation quality and efficiency.

cs.LG↗

arXiv2025

An evaluation of LLMs for generating movie reviews: GPT-4o, Gemini-2.0 and DeepSeek-V3

Large language models (LLMs) have been prominent in various tasks, including text generation and summarisation. The applicability of LLMs to the generation of product reviews is gaining momentum, paving the way for the generation of movie reviews. In this study, we propose a framework that generates movie reviews using three LLMs (GPT-4o, DeepSeek-V3, and Gemini-2.0), and evaluate their performance by comparing the generated outputs with IMDb user reviews. We use movie subtitles and screenplays as input to the LLMs and investigate how they affect the quality of reviews generated. We review the LLM-based movie reviews in terms of vocabulary, sentiment polarity, similarity, and thematic consistency in comparison to IMDB user reviews. The results demonstrate that LLMs are capable of generating syntactically fluent and structurally complete movie reviews. Nevertheless, there is still a noticeable gap in emotional richness and stylistic coherence between LLM-generated and IMDb reviews, suggesting that further refinement is needed to improve the overall quality of movie review generation. We provided a survey-based analysis where participants were told to distinguish between LLM and IMDb user reviews. The results show that LLM-generated reviews are difficult to distinguish from IMDB user reviews. We found that DeepSeek-V3 produced the most balanced reviews, closely matching IMDb reviews. GPT-4o overemphasised positive emotions, while Gemini-2.0 captured negative emotions better but showed excessive emotional intensity.

cs.CL↗

arXiv2025

Distributed Retrieval-Augmented Generation

As large language models (LLMs) become increasingly adopted on edge devices, Retrieval-Augmented Generation (RAG) is gaining prominence as a solution to address factual deficiencies and hallucinations by integrating external knowledge. However, centralized RAG architectures face significant challenges in data privacy and scalability. For instance, smart healthcare services often rely on collecting sensitive patient data and building a centralized knowledge base to provide better diagnosis and treatment advice, while privacy concerns significantly impede this process. Besides, maintaining a comprehensive and continuously updated knowledge base is costly, particularly in response to regional epidemics and rapidly mutating viruses. To address these challenges, this paper introduces Distributed Retrieval-Augmented Generation (DRAG), a novel framework that improves data privacy by eliminating the need for a centralized knowledge base and restoring data control to owners. DRAG incorporates a Topic-Aware Random Walk (TARW) algorithm that leverages LLMs to extract query topics and facilitate targeted peer discovery within a peer-to-peer network, enabling efficient knowledge retrieval in decentralized environments. Extensive experiments across three diverse datasets and LLMs demonstrate that DRAG with TARW achieves near-centralized RAG performance by using half as many messages as flooding. The code is available at https://github.com/xuchenhao001/DRAG.

cs.DC↗

arXiv2025

BQSched: A Non-intrusive Scheduler for Batch Concurrent Queries via Reinforcement Learning

Most large enterprises build predefined data pipelines and execute them periodically to process operational data using SQL queries for various tasks. A key issue in minimizing the overall makespan of these pipelines is the efficient scheduling of concurrent queries within the pipelines. Existing tools mainly rely on simple heuristic rules due to the difficulty of expressing the complex features and mutual influences of queries. The latest reinforcement learning (RL) based methods have the potential to capture these patterns from feedback, but it is non-trivial to apply them directly due to the large scheduling space, high sampling cost, and poor sample utilization. Motivated by these challenges, we propose BQSched, a non-intrusive Scheduler for Batch concurrent Queries via reinforcement learning. Specifically, BQSched designs an attention-based state representation to capture the complex query patterns, and proposes IQ-PPO, an auxiliary task-enhanced proximal policy optimization (PPO) algorithm, to fully exploit the rich signals of Individual Query completion in logs. Based on the RL framework above, BQSched further introduces three optimization strategies, including adaptive masking to prune the action space, scheduling gain-based query clustering to deal with large query sets, and an incremental simulator to reduce sampling cost. To our knowledge, BQSched is the first non-intrusive batch query scheduler via RL. Extensive experiments show that BQSched can significantly improve the efficiency and stability of batch query scheduling, while also achieving remarkable scalability and adaptability in both data and queries. For example, across all DBMSs and scales tested, BQSched reduces the overall makespan of batch queries on TPC-DS benchmark by an average of 34% and 13%, compared with the commonly used heuristic strategy and the adapted RL-based scheduler, respectively.

cs.DB↗

arXiv2024

ODC-SA Net: Orthogonal Direction Enhancement and Scale Aware Network for Polyp Segmentation

Accurate polyp segmentation is crucial for the early detection and prevention of colorectal cancer. However, the existing polyp detection methods sometimes ignore multi-directional features and drastic changes in scale. To address these challenges, we design an Orthogonal Direction Enhancement and Scale Aware Network (ODC-SA Net) for polyp segmentation. The Orthogonal Direction Convolutional (ODC) block can extract multi-directional features using transposed rectangular convolution kernels through forming an orthogonal feature vector basis, which solves the issue of random feature direction changes and reduces computational load. Additionally, the Multi-scale Fusion Attention (MSFA) mechanism is proposed to emphasize scale changes in both spatial and channel dimensions, enhancing the segmentation accuracy for polyps of varying sizes. Extraction with Re-attention Module (ERA) is used to re-combinane effective features, and Structures of Shallow Reverse Attention Mechanism (SRA) is used to enhance polyp edge with low level information. A large number of experiments conducted on public datasets have demonstrated that the performance of this model is superior to state-of-the-art methods.

cs.CV↗

arXiv2024

Mask-TS Net: Mask Temperature Scaling Uncertainty Calibration for Polyp Segmentation

Lots of popular calibration methods in medical images focus on classification, but there are few comparable studies on semantic segmentation. In polyp segmentation of medical images, we find most diseased area occupies only a small portion of the entire image, resulting in previous models being not well-calibrated for lesion regions but well-calibrated for background, despite their seemingly better Expected Calibration Error (ECE) scores overall. Therefore, we proposed four-branches calibration network with Mask-Loss and Mask-TS strategies to more focus on the scaling of logits within potential lesion regions, which serves to mitigate the influence of background interference. In the experiments, we compare the existing calibration methods with the proposed Mask Temperature Scaling (Mask-TS). The results indicate that the proposed calibration network outperforms other methods both qualitatively and quantitatively.

cs.CV↗

arXiv2024

HSViT: Horizontally Scalable Vision Transformer

Due to its deficiency in prior knowledge (inductive bias), Vision Transformer (ViT) requires pre-training on large-scale datasets to perform well. Moreover, the growing layers and parameters in ViT models impede their applicability to devices with limited computing resources. To mitigate the aforementioned challenges, this paper introduces a novel horizontally scalable vision transformer (HSViT) scheme. Specifically, a novel image-level feature embedding is introduced to ViT, where the preserved inductive bias allows the model to eliminate the need for pre-training while outperforming on small datasets. Besides, a novel horizontally scalable architecture is designed, facilitating collaborative model training and inference across multiple computing devices. The experimental results depict that, without pre-training, HSViT achieves up to 10% higher top-1 accuracy than state-of-the-art schemes on small datasets, while providing existing CNN backbones up to 3.1% improvement in top-1 accuracy on ImageNet. The code is available at https://github.com/xuchenhao001/HSViT.

cs.CV↗

arXiv2023

Deep Learning Techniques for Video Instance Segmentation: A Survey

Video instance segmentation, also known as multi-object tracking and segmentation, is an emerging computer vision research area introduced in 2019, aiming at detecting, segmenting, and tracking instances in videos simultaneously. By tackling the video instance segmentation tasks through effective analysis and utilization of visual information in videos, a range of computer vision-enabled applications (e.g., human action recognition, medical image processing, autonomous vehicle navigation, surveillance, etc) can be implemented. As deep-learning techniques take a dominant role in various computer vision areas, a plethora of deep-learning-based video instance segmentation schemes have been proposed. This survey offers a multifaceted view of deep-learning schemes for video instance segmentation, covering various architectural paradigms, along with comparisons of functional performance, model complexity, and computational overheads. In addition to the common architectural designs, auxiliary techniques for improving the performance of deep-learning models for video instance segmentation are compiled and discussed. Finally, we discuss a range of major challenges and directions for further investigations to help advance this promising research field.

cs.CV↗

arXiv2023

Asynchronous Federated Learning on Heterogeneous Devices: A Survey

Federated learning (FL) is a kind of distributed machine learning framework, where the global model is generated on the centralized aggregation server based on the parameters of local models, addressing concerns about privacy leakage caused by the collection of local training data. With the growing computational and communication capacities of edge and IoT devices, applying FL on heterogeneous devices to train machine learning models is becoming a prevailing trend. Nonetheless, the synchronous aggregation strategy in the classic FL paradigm, particularly on heterogeneous devices, encounters limitations in resource utilization due to the need to wait for slow devices before aggregation in each training round. Furthermore, the uneven distribution of data across devices (i.e. data heterogeneity) in real-world scenarios adversely impacts the accuracy of the global model. Consequently, many asynchronous FL (AFL) approaches have been introduced across various application contexts to enhance efficiency, performance, privacy, and security. This survey comprehensively analyzes and summarizes existing AFL variations using a novel classification scheme, including device heterogeneity, data heterogeneity, privacy, and security on heterogeneous devices, as well as applications on heterogeneous devices. Finally, this survey reveals rising challenges and presents potentially promising research directions in this under-investigated domain.

cs.DC↗

arXiv2023

SCEI: A Smart-Contract Driven Edge Intelligence Framework for IoT Systems

Federated learning (FL) enables collaborative training of a shared model on edge devices while maintaining data privacy. FL is effective when dealing with independent and identically distributed (iid) datasets, but struggles with non-iid datasets. Various personalized approaches have been proposed, but such approaches fail to handle underlying shifts in data distribution, such as data distribution skew commonly observed in real-world scenarios (e.g., driver behavior in smart transportation systems changing across time and location). Additionally, trust concerns among unacquainted devices and security concerns with the centralized aggregator pose additional challenges. To address these challenges, this paper presents a dynamically optimized personal deep learning scheme based on blockchain and federated learning. Specifically, the innovative smart contract implemented in the blockchain allows distributed edge devices to reach a consensus on the optimal weights of personalized models. Experimental evaluations using multiple models and real-world datasets demonstrate that the proposed scheme achieves higher accuracy and faster convergence compared to traditional federated and personalized learning approaches.

cs.LG↗

arXiv2023

RPDP: An Efficient Data Placement based on Residual Performance for P2P Storage Systems

Storage systems using Peer-to-Peer (P2P) architecture are an alternative to the traditional client-server systems. They offer better scalability and fault tolerance while at the same time eliminate the single point of failure. The nature of P2P storage systems (which consist of heterogeneous nodes) introduce however data placement challenges that create implementation trade-offs (e.g., between performance and scalability). Existing Kademlia-based DHT data placement method stores data at closest node, where the distance is measured by bit-wise XOR operation between data and a given node. This approach is highly scalable because it does not require global knowledge for placing data nor for the data retrieval. It does not however consider the heterogeneous performance of the nodes, which can result in imbalanced resource usage affecting the overall latency of the system. Other works implement criteria-based selection that addresses heterogeneity of nodes, however often cause subsequent data retrieval to require global knowledge of where the data stored. This paper introduces Residual Performance-based Data Placement (RPDP), a novel data placement method based on dynamic temporal residual performance of data nodes. RPDP places data to most appropriate selected nodes based on their throughput and latency with the aim to achieve lower overall latency by balancing data distribution with respect to the individual performance of nodes. RPDP relies on Kademlia-based DHT with modified data structure to allow data subsequently retrieved without the need of global knowledge. The experimental results indicate that RPDP reduces the overall latency of the baseline Kademlia-based P2P storage system (by 4.87%) and it also reduces the variance of latency among the nodes, with minimal impact to the data retrieval complexity.

cs.DC↗

arXiv2023

AICons: An AI-Enabled Consensus Algorithm Driven by Energy Preservation and Fairness

Blockchain has been used in several domains. However, this technology still has major limitations that are largely related to one of its core components, namely the consensus protocol/algorithm. Several solutions have been proposed in literature and some of them are based on the use of Machine Learning (ML) methods. The ML-based consensus algorithms usually waste the work done by the (contributing/participating) nodes, as only winners' ML models are considered/used, resulting in low energy efficiency. To reduce energy waste and improve scalability, this paper proposes an AI-enabled consensus algorithm (named AICons) driven by energy preservation and fairness of rewarding nodes based on their contribution. In particular, the local ML models trained by all nodes are utilised to generate a global ML model for selecting winners, which reduces energy waste. Considering the fairness of the rewards, we innovatively designed a utility function for the Shapley value evaluation equation to evaluate the contribution of each node from three aspects, namely ML model accuracy, energy consumption, and network bandwidth. The three aspects are combined into a single Shapley value to reflect the contribution of each node in a blockchain system. Extensive experiments were carried out to evaluate fairness, scalability, and profitability of the proposed solution. In particular, AICons has an evenly distributed reward-contribution ratio across nodes, handling 38.4 more transactions per second, and allowing nodes to get more profit to support a bigger network than the state-of-the-art schemes.

cs.DC↗

arXiv2022

An Efficient and Reliable Asynchronous Federated Learning Scheme for Smart Public Transportation

Since the traffic conditions change over time, machine learning models that predict traffic flows must be updated continuously and efficiently in smart public transportation. Federated learning (FL) is a distributed machine learning scheme that allows buses to receive model updates without waiting for model training on the cloud. However, FL is vulnerable to poisoning or DDoS attacks since buses travel in public. Some work introduces blockchain to improve reliability, but the additional latency from the consensus process reduces the efficiency of FL. Asynchronous Federated Learning (AFL) is a scheme that reduces the latency of aggregation to improve efficiency, but the learning performance is unstable due to unreasonably weighted local models. To address the above challenges, this paper offers a blockchain-based asynchronous federated learning scheme with a dynamic scaling factor (DBAFL). Specifically, the novel committee-based consensus algorithm for blockchain improves reliability at the lowest possible cost of time. Meanwhile, the devised dynamic scaling factor allows AFL to assign reasonable weights to stale local models. Extensive experiments conducted on heterogeneous devices validate outperformed learning performance, efficiency, and reliability of DBAFL.

cs.LG↗

arXiv2017

Learning convolutional neural network to maximize Pos@Top performance measure

In the machine learning problems, the performance measure is used to evaluate the machine learning models. Recently, the number positive data points ranked at the top positions (Pos@Top) has been a popular performance measure in the machine learning community. In this paper, we propose to learn a convolutional neural network (CNN) model to maximize the Pos@Top performance measure. The CNN model is used to represent the multi-instance data point, and a classifier function is used to predict the label from the its CNN representation. We propose to minimize the loss function of Pos@Top over a training set to learn the filters of CNN and the classifier parameter. The classifier parameter vector is solved by the Lagrange multiplier method, and the filters are updated by the gradient descent method alternately in an iterative algorithm. Experiments over benchmark data sets show that the proposed method outperforms the state-of-the-art Pos@Top maximization methods.

cs.CV↗