arXiv subjects

Chao Feng

Publications and source records attributed to Chao Feng.

At least 55 records · Page 3Linked to original sources

arXiv2025

Understanding and Optimizing Agentic Workflows via Shapley value

Agentic workflows have become the dominant paradigm for building complex AI systems, orchestrating specialized components, such as planning, reasoning, action execution, and reflection, to tackle sophisticated real-world tasks. However, systematically analyzing and optimizing these workflows remains challenging due to intricate component interdependencies and the lack of principled attribution methods. In this work, we introduce ShapleyFlow, the first framework that employs cooperative game theory to analyze and optimize agentic workflows. By applying the Shapley value to evaluate all possible component configurations, ShapleyFlow enables fine-grained attribution of each component's contribution and facilitates the identification of task-specific optimal configurations. Through a constructed dataset evaluated across 7 scenarios, such as navigation, math and OS, we demonstrate 3 key contributions: (1) Theoretical Framework: a principled game-theoretic approach for the attribution of contributions in agentic workflows. (2) Optimal Workflow Discovery: ShapleyFlow identifies task-specific component configurations that consistently outperform workflows relying on a single LLM across all tested tasks. (3) Comprehensive Analysis: we construct and analyze over 1,500 tasks, providing actionable insights and design guidelines for optimizing workflows across multiple domains.

cs.AI↗

arXiv2025

SAIL-Embedding Technical Report: Omni-modal Embedding Foundation Model

Multimodal embedding models aim to yield informative unified representations that empower diverse cross-modal tasks. Despite promising developments in the evolution from CLIP-based dual-tower architectures to large vision-language models, prior works still face unavoidable challenges in real-world applications and business scenarios, such as the limited modality support, unstable training mechanisms, and industrial domain gaps. In this work, we introduce SAIL-Embedding, an omni-modal embedding foundation model that addresses these issues through tailored training strategies and architectural design. In the optimization procedure, we propose a multi-stage training scheme to boost the multifaceted effectiveness of representation learning. Specifically, the content-aware progressive training aims to enhance the model's adaptability to diverse downstream tasks and master enriched cross-modal proficiency. The collaboration-aware recommendation enhancement training further adapts multimodal representations for recommendation scenarios by distilling knowledge from sequence-to-item and ID-to-item embeddings while mining user historical interests. Concurrently, we develop the stochastic specialization and dataset-driven pattern matching to strengthen model training flexibility and generalizability. Experimental results show that SAIL-Embedding achieves SOTA performance compared to other methods in different retrieval tasks. In online experiments across various real-world scenarios integrated with our model, we observe a significant increase in Lifetime (LT), which is a crucial indicator for the recommendation experience. For instance, the model delivers the 7-day LT gain of +0.5% in the Douyin-Selected scenario. For the Douyin feed rank model, the match features produced by SAIL-Embedding yield a +0.1% AUC gain.

cs.IR↗

arXiv2025

Masked Diffusion Captioning for Visual Feature Learning

We learn visual features by captioning images with an image-conditioned masked diffusion language model, a formulation we call masked diffusion captioning (MDC). During training, text tokens in each image-caption pair are masked at a randomly chosen ratio, and a decoder conditioned on visual features is trained to reconstruct the original text. After training, the learned visual features can be applied to downstream vision tasks. Unlike autoregressive captioning, the strength of the visual learning signal in MDC does not depend on each token's position in the sequence, reducing the need for auxiliary objectives. Linear probing experiments across a variety of academic-scale models and datasets show that the learned visual features are competitive with those produced by autoregressive and contrastive approaches.

cs.CV↗

arXiv2025

DUET: Dual Model Co-Training for Entire Space CTR Prediction

The pre-ranking stage plays a pivotal role in large-scale recommender systems but faces an intrinsic trade-off between model expressiveness and computational efficiency. Owing to the massive candidate pool and strict latency constraints, industry systems often rely on lightweight two-tower architectures, which are computationally efficient yet limited in estimation capability. As a result, they struggle to capture the complex synergistic and suppressive relationships among candidate items, which are essential for producing contextually coherent and diverse recommendation lists. Moreover, this simplicity further amplifies the Sample Selection Bias (SSB) problem, as coarse-grained models trained on biased exposure data must generalize to a much larger candidate space with distinct distributions. To address these issues, we propose \textbf{DUET} (\textbf{DU}al Model Co-Training for \textbf{E}ntire Space C\textbf{T}R Prediction), a set-wise pre-ranking framework that achieves expressive modeling under tight computational budgets. Instead of scoring items independently, DUET performs set-level prediction over the entire candidate subset in a single forward pass, enabling information-aware interactions among candidates while amortizing the computational cost across the set. Moreover, a dual model co-training mechanism extends supervision to unexposed items via mutual pseudo-label refinement, effectively mitigating SSB. Validated through extensive offline experiments and online A/B testing, DUET consistently outperforms state-of-the-art baselines and achieves improvements across multiple core business metrics. At present, DUET has been fully deployed in Kuaishou and Kuaishou Lite Apps, serving the main traffic for hundreds of millions of users.

cs.IR↗

arXiv2025

Attosecond Waveform Synthesis through Echo-enabled Harmonic Generation Free-electron Lasers

Attosecond pulse trains (APTs) are indispensable for probing electron dynamics at their intrinsic timescales. High harmonic generation (HHG) has long been a successful and widely used technique in producing extreme ultraviolet APTs. While in the soft X-ray regime, HHG suffers from low conversion efficiency and lacking flexibility in the waveform and spectrum control. Here in this study, based on the waveform synthesis of the echo-enabled harmonic generation (EEHG) free-electron laser (FEL), we propose a novel method that can generate soft X-ray APTs with high temporal and spectral tunability. EEHG scheme is well-known in forming phase-locked high harmonic bunching combs down to the soft X-ray regime. And We can facilitate successive FEL lasing in several different harmonic numbers and then perform the coherent waveform synthesis to generate soft X-ray APTs. Three-dimensional simulation results indicate that reproducible generation of coherent APTs in the soft X-ray regime with peak power of 3.5 GW and micropulse duration of 160 as can be achieved by five-harmonic synthesis. And the micropulse duration together with the spectral components of the APTs can be adjusted easily according to the programmable high harmonic bunching combs in the EEHG scheme. This method has the potential to establish a robust platform for attosecond sciences in the soft X-ray regime, which can enable unprecedented studies of electron dynamics in physical and chemical reactions, biological systems and quantum materials.

physics.acc-ph↗

arXiv2025

Intense and Tunable Multi-color Terahertz Radiation from Laser-Shaped Electron Beams

High-power multi-color terahertz (THz) radiation exhibits extraordinary scientific application prospects at various scientific frontiers, for its capacity to deliver THz excitation at multiple frequencies simultaneously. However, the generation of high-power multi-color THz radiation with tunable frequencies remains a challenge for existing techniques. Here, a technique by combining the multi-laser pulses frequency beating and coherent undulator amplification is proposed for generating high-power multi-color THz radiation with tunable frequency. Numerical simulations indicate that the proposed technique can produce multi-color THz radiation with three to six distinguished colors and a peak power up to hundreds of MW, and the temporally separated two-color pulses can also be produced by employing undulators with different resonance. Due to the intrinsic properties of the proposed technique, the THz frequencies, the color number and the frequency interval can be effectively controlled by simply adjusting the beating laser. This method paves the way for advanced application of THz pump-THz probe experiments for selective excitation of atomic multi-level systems and molecular fingerprint recognition.

physics.acc-ph↗

arXiv2025

Boosting Multi-modal Keyphrase Prediction with Dynamic Chain-of-Thought in Vision-Language Models

Multi-modal keyphrase prediction (MMKP) aims to advance beyond text-only methods by incorporating multiple modalities of input information to produce a set of conclusive phrases. Traditional multi-modal approaches have been proven to have significant limitations in handling the challenging absence and unseen scenarios. Additionally, we identify shortcomings in existing benchmarks that overestimate model capability due to significant overlap in training tests. In this work, we propose leveraging vision-language models (VLMs) for the MMKP task. Firstly, we use two widely-used strategies, e.g., zero-shot and supervised fine-tuning (SFT) to assess the lower bound performance of VLMs. Next, to improve the complex reasoning capabilities of VLMs, we adopt Fine-tune-CoT, which leverages high-quality CoT reasoning data generated by a teacher model to finetune smaller models. Finally, to address the "overthinking" phenomenon, we propose a dynamic CoT strategy which adaptively injects CoT data during training, allowing the model to flexibly leverage its reasoning capabilities during the inference stage. We evaluate the proposed strategies on various datasets and the experimental results demonstrate the effectiveness of the proposed approaches. The code is available at https://github.com/bytedance/DynamicCoT.

cs.CV↗

arXiv2025

Benchmarking Chinese Commonsense Reasoning with a Multi-hop Reasoning Perspective

While Large Language Models (LLMs) have demonstrated advanced reasoning capabilities, their comprehensive evaluation in general Chinese-language contexts remains understudied. To bridge this gap, we propose Chinese Commonsense Multi-hop Reasoning (CCMOR), a novel benchmark designed to evaluate LLMs' ability to integrate Chinese-specific factual knowledge with multi-step logical reasoning. Specifically, we first construct a domain-balanced seed set from existing QA datasets, then develop an LLM-powered pipeline to generate multi-hop questions anchored on factual unit chains. To ensure the quality of resulting dataset, we implement a human-in-the-loop verification system, where domain experts systematically validate and refine the generated questions. Using CCMOR, we evaluate state-of-the-art LLMs, demonstrating persistent limitations in LLMs' ability to process long-tail knowledge and execute knowledge-intensive reasoning. Notably, retrieval-augmented generation substantially mitigates these knowledge gaps, yielding significant performance gains.

cs.CL↗

arXiv2025

ColNet: Collaborative Optimization in Decentralized Federated Multi-task Learning Systems

The integration of Federated Learning (FL) and Multi-Task Learning (MTL) has been explored to address client heterogeneity, with Federated Multi-Task Learning (FMTL) treating each client as a distinct task. However, most existing research focuses on data heterogeneity (e.g., addressing non-IID data) rather than task heterogeneity, where clients solve fundamentally different tasks. Additionally, much of the work relies on centralized settings with a server managing the federation, leaving the more challenging domain of decentralized FMTL largely unexplored. Thus, this work bridges this gap by proposing ColNet, a framework designed for heterogeneous tasks in decentralized federated environments. ColNet partitions models into a backbone and task-specific heads, and uses adaptive clustering based on model and data sensitivity to form task-coherent client groups. Backbones are averaged within groups, and group leaders perform hyper-conflict-averse cross-group aggregation. Across datasets and federations, ColNet outperforms competing schemes under label and task heterogeneity and shows robustness to poisoning attacks.

cs.LG↗

arXiv2025

Conf-Profile: A Confidence-Driven Reasoning Paradigm for Label-Free User Profiling

User profiling, as a core technique for user understanding, aims to infer structural attributes from user information. Large Language Models (LLMs) provide a promising avenue for user profiling, yet the progress is hindered by the lack of comprehensive benchmarks. To bridge this gap, we propose ProfileBench, an industrial benchmark derived from a real-world video platform, encompassing heterogeneous user data and a well-structured profiling taxonomy. However, the profiling task remains challenging due to the difficulty of collecting large-scale ground-truth labels, and the heterogeneous and noisy user information can compromise the reliability of LLMs. To approach label-free and reliable user profiling, we propose a Confidence-driven Profile reasoning framework Conf-Profile, featuring a two-stage paradigm. We first synthesize high-quality labels by leveraging advanced LLMs with confidence hints, followed by confidence-weighted voting for accuracy improvement and confidence calibration for a balanced distribution. The multiple profile results, rationales, and confidence scores are aggregated and distilled into a lightweight LLM. We further enhance the reasoning ability via confidence-guided unsupervised reinforcement learning, which exploits confidence for difficulty filtering, quasi-ground truth voting, and reward weighting. Experimental results demonstrate that Conf-Profile delivers substantial performance through the two-stage training, improving F1 by 13.97 on Qwen3-8B.

cs.AI↗

arXiv2025

Generate the browsing process for short-video recommendation

This paper proposes a generative method to dynamically simulate users' short video watching journey for watch time prediction in short video recommendation. Unlike existing methods that rely on multimodal features for video content understanding, our method simulates users' sustained interest in watching short videos by learning collaborative information, using interest changes from existing positive and negative feedback videos and user interaction behaviors to implicitly model users' video watching journey. By segmenting videos based on duration and adopting a Transformer-like architecture, our method can capture sequential dependencies between segments while mitigating duration bias. Extensive experiments on industrial-scale and public datasets demonstrate that our method achieves state-of-the-art performance on watch time prediction tasks. The method has been deployed on Kuaishou Lite, achieving a significant improvement of +0.13\% in APP duration, and reaching an XAUC of 83\% for single video watch time prediction on industrial-scale streaming training sets, far exceeding other methods. The proposed method provides a scalable and effective solution for video recommendation through segment-level modeling and user engagement feedback.

cs.IR↗

arXiv2025

SAIL-VL2 Technical Report

We introduce SAIL-VL2, an open-suite vision-language foundation model (LVM) for comprehensive multimodal understanding and reasoning. As the successor to SAIL-VL, SAIL-VL2 achieves state-of-the-art performance at the 2B and 8B parameter scales across diverse image and video benchmarks, demonstrating strong capabilities from fine-grained perception to complex reasoning. Its effectiveness is driven by three core innovations. First, a large-scale data curation pipeline with scoring and filtering strategies enhances both quality and distribution across captioning, OCR, QA, and video data, improving training efficiency. Second, a progressive training framework begins with a powerful pre-trained vision encoder (SAIL-ViT), advances through multimodal pre-training, and culminates in a thinking-fusion SFT-RL hybrid paradigm that systematically strengthens model capabilities. Third, architectural advances extend beyond dense LLMs to efficient sparse Mixture-of-Experts (MoE) designs. With these contributions, SAIL-VL2 demonstrates competitive performance across 106 datasets and achieves state-of-the-art results on challenging reasoning benchmarks such as MMMU and MathVista. Furthermore, on the OpenCompass leaderboard, SAIL-VL2-2B ranks first among officially released open-source models under the 4B parameter scale, while serving as an efficient and extensible foundation for the open-source multimodal community.

cs.CV↗

arXiv2025

Generating 10 GW-class isolated zeptosecond x-ray bursts through bootstrapping two plasma-based accelerator stages

A method is proposed to generate coherent, intense zeptosecond x-ray pulses by using the electron beam produced by a laser wakefield accelerator (LWFA) stage as the driver for a beam driven plasma wakefield accelerator (PWFA) stage. The LWFA injector stage requires a readily available 100 TW laser driver to produce a GeV class self-injected electron beam. This beam is focused and used as the driver in a PWFA stage that utilizes a solid-density plasma with a modulated density downramp. The concept is shown to be capable of producing an ultra-short electron beam with unprecedented density ($10^{26}~\mathrm{cm^{-3}}$) and brightness ($ 10^{24}~\ampere/\meter^2/\radian ^2$), that is also pre-bunched on 0.1 Angstrom scales. By colliding this pre-bunched extreme beam with an optical undulator, an intense zeptosecond pulse can be emitted if the spatially focusing region is matched with the lasing region of the beam. Multi-dimensional particle-in-cell simulations are performed of the entire concept to demonstrate that an isolated 10 GW-class zeptosecond pulse with a full-width-half-maximum duration of 700 zs can be generated in a tapered laser pulse. Such an intense zeptosecond pulse generation scheme may provide an essential probe for nuclear physics and quantum electrodynamics processes.

physics.acc-ph↗

arXiv2025

Benchmarking Vision-Language Models on Chinese Ancient Documents: From OCR to Knowledge Reasoning

Chinese ancient documents, invaluable carriers of millennia of Chinese history and culture, hold rich knowledge across diverse fields but face challenges in digitization and understanding, i.e., traditional methods only scan images, while current Vision-Language Models (VLMs) struggle with their visual and linguistic complexity. Existing document benchmarks focus on English printed texts or simplified Chinese, leaving a gap for evaluating VLMs on ancient Chinese documents. To address this, we present AncientDoc, the first benchmark for Chinese ancient documents, designed to assess VLMs from OCR to knowledge reasoning. AncientDoc includes five tasks (page-level OCR, vernacular translation, reasoning-based QA, knowledge-based QA, linguistic variant QA) and covers 14 document types, over 100 books, and about 3,000 pages. Based on AncientDoc, we evaluate mainstream VLMs using multiple metrics, supplemented by a human-aligned large language model for scoring.

cs.CL↗

arXiv2025

GreenDFL: a Framework for Assessing the Sustainability of Decentralized Federated Learning Systems

Decentralized Federated Learning (DFL) is an emerging paradigm that enables collaborative model training without centralized data and model aggregation, enhancing privacy and resilience. However, its sustainability remains underexplored, as energy consumption and carbon emissions vary across different system configurations. Understanding the environmental impact of DFL is crucial for optimizing its design and deployment. This work aims to develop a comprehensive and operational framework for assessing the sustainability of DFL systems. To address it, this work provides a systematic method for quantifying energy consumption and carbon emissions, offering insights into improving the sustainability of DFL. This work proposes GreenDFL, a fully implementable framework that has been integrated into a real-world DFL platform. GreenDFL systematically analyzes the impact of various factors, including hardware accelerators, model architecture, communication medium, data distribution, network topology, and federation size, on the sustainability of DFL systems. Besides, a sustainability-aware aggregation algorithm (GreenDFL-SA) and a node selection algorithm (GreenDFL-SN) are developed to optimize energy efficiency and reduce carbon emissions in DFL training. Empirical experiments are conducted on multiple datasets, measuring energy consumption and carbon emissions at different phases of the DFL lifecycle. The proposed GreenDFL provides a comprehensive and practical approach for assessing the sustainability of DFL systems. Furthermore, it offers best practices for improving environmental efficiency in DFL, making sustainability considerations more actionable in real-world deployments.

cs.CY↗

arXiv2025

Assessing the Sustainability and Trustworthiness of Federated Learning Models

Artificial intelligence (AI) increasingly influences critical decision-making across sectors. Federated Learning (FL), as a privacy-preserving collaborative AI paradigm, not only enhances data protection but also holds significant promise for intelligent network management, including distributed monitoring, adaptive control, and edge intelligence. Although the trustworthiness of FL systems has received growing attention, the sustainability dimension remains insufficiently explored, despite its importance for scalable real-world deployment. To address this gap, this work introduces sustainability as a distinct pillar within a comprehensive trustworthy FL taxonomy, consistent with AI-HLEG guidelines. This pillar includes three key aspects: hardware efficiency, federation complexity, and the carbon intensity of energy sources. Experiments using the FederatedScope framework under diverse scenarios, including varying participants, system complexity, hardware, and energy configurations, validate the practicality of the approach. Results show that incorporating sustainability into FL evaluation supports environmentally responsible deployment, enabling more efficient, adaptive, and trustworthy network services and management AI models.

cs.CY↗

arXiv2025

From Models to Network Topologies: A Topology Inference Attack in Decentralized Federated Learning

Federated Learning (FL) is widely recognized as a privacy-preserving Machine Learning paradigm due to its model-sharing mechanism that avoids direct data exchange. Nevertheless, model training leaves exploitable traces that can be used to infer sensitive information. In Decentralized FL (DFL), the topology, defining how participants are connected, plays a crucial role in shaping the model's privacy, robustness, and convergence. However, the topology introduces an unexplored vulnerability: attackers can exploit it to infer participant relationships and launch targeted attacks. This work uncovers the hidden risks of DFL topologies by proposing a novel Topology Inference Attack that infers the topology solely from model behavior. A taxonomy of topology inference attacks is introduced, categorizing them by the attacker's capabilities and knowledge. Practical attack strategies are designed for various scenarios, and experiments are conducted to identify key factors influencing attack success. The results demonstrate that analyzing only the model of each node can accurately infer the DFL topology, highlighting a critical privacy risk in DFL systems. These findings offer insights for improving privacy preservation in DFL environments.

cs.LG↗

arXiv2025

LVFace: Progressive Cluster Optimization for Large Vision Models in Face Recognition

Vision Transformers (ViTs) have revolutionized large-scale visual modeling, yet remain underexplored in face recognition (FR) where CNNs still dominate. We identify a critical bottleneck: CNN-inspired training paradigms fail to unlock ViT's potential, leading to suboptimal performance and convergence instability.To address this challenge, we propose LVFace, a ViT-based FR model that integrates Progressive Cluster Optimization (PCO) to achieve superior results. Specifically, PCO sequentially applies negative class sub-sampling (NCS) for robust and fast feature alignment from random initialization, feature expectation penalties for centroid stabilization, performing cluster boundary refinement through full-batch training without NCS constraints. LVFace establishes a new state-of-the-art face recognition baseline, surpassing leading approaches such as UniFace and TopoFR across multiple benchmarks. Extensive experiments demonstrate that LVFace delivers consistent performance gains, while exhibiting scalability to large-scale datasets and compatibility with mainstream VLMs and LLMs. Notably, LVFace secured 1st place in the ICCV 2021 Masked Face Recognition (MFR)-Ongoing Challenge (March 2025), proving its efficacy in real-world scenarios. Project is available at https://github.com/bytedance/LVFace.

cs.CV↗