arXiv subjects

Yuchen Jiang

Publications and source records attributed to Yuchen Jiang.

At least 19 recordsLinked to original sources

arXiv2026

Design and commissioning of a windowless gas-target system for high-current beams at JUNA

Windowless gas targets avoid the beam-energy loss and straggling introduced by entrance foils and are therefore well suited for direct measurements of low-energy nuclear reactions. A windowless gas-target system designed for operation with milliampere beams has been developed for the Jinping Underground Nuclear Astrophysics facility (JUNA). The system combines three-stage differential pumping, closed-loop gas recovery and purification, a constant-temperature power-compensation calorimeter, and a position-resolved target-thickness monitor based on secondary elastic scattering. Stable operation was achieved over a target-pressure range of 1-3 mbar, with pressure fluctuations below 1% during 8 h of continuous circulation, while the accelerator-side pressure was maintained at approximately $10^{-4}$ Pa. The closed-loop gas-circulation system maintained stable target conditions, while gas-transport calculations indicated that the axial pressure nonuniformity remained within approximately 1.6% under representative operating conditions. Calorimeter measurements were consistent with the thermal calculations, supporting the sensitivity correction used for beam-power determination. Beam commissioning with $^{14}\mathrm{N}(p,\gamma)^{15}\mathrm{O}$ and $^{12}\mathrm{C}(p,\gamma)^{13}\mathrm{N}$ at the 600 kV Cockcroft-Walton accelerator of the China Institute of Atomic Energy (CIAE) demonstrated stable operation of the gas-target and $\gamma$-ray detection systems and provided information on the influences of reaction position and beam heating. These results demonstrate the operating stability and diagnostic capability of the system for future high-current, low-energy nuclear-reaction measurements at JUNA.

astro-ph.GA↗

arXiv2026

Teacher Retains Full Tokens, Student Merges Efficiently: TM20K for E-Commerce Sequence Modeling in Ad Recommendation

Benefiting from ultra-long behavior sequence modeling, existing recommender systems bring users a better experience via simultaneously considering their long-term and short-term interests. Nevertheless, extended sequence lengths introduce substantial burdens on training efficiency and serving throughput. Prior approaches typically utilize search-based or cluster-based compression on ultra-long sequences at the cost of fine-grained information, or rely on various lightweight target attention structures incapable of sufficient sequential feature extraction. In this paper, we balance the effectiveness and efficiency for ultra-long sequence modeling via full transformer modeling accompanied with a two-stage knowledge distillation framework. First, both teacher and student models take the full attention mechanism rather than pure target-sequence attention for effective sequence scaling. For student models, we propose several simple yet well-motivated token merge approaches, significantly compressing the sequence length while maintaining an acceptable performance. Then, a one-time teacher is heavily trained with full sequence tokens, further boosting the performance of student models via knowledge distillation. The proposed paradigm named TM20K has been successfully deployed in ByteDance's e-commerce advertising recommender system that extends the e-commerce sequence length to 20K, delivering substantial improvements in key business metrics (e.g., ADSS +1.036\%) while keeping the training and serving cost nearly the same as the online state-of-the-art model (e.g., serving latency only +5.6\%).

cs.IR↗

arXiv2026

PhysAgent: Automating Physics-Based 4D Synthesis via Trajectory-Grounded Multi-Agent Feedback

Achieving fully automated, physically plausible 3D motion synthesis is a core objective in graphics and generative AI. However, configuring complex environmental force fields still relies entirely on manual expert intervention, creating a severe bottleneck for large-scale simulation data generation. Existing automated methods primarily focus on material optimization and exhibit severe modality gaps and technical flaws when applied to the vastly more complex force field optimization space: naive Large Language Models (LLMs) lack underlying simulation feedback, causing severe physical inaccuracies, while traditional Score Distillation Sampling (SDS) suffers from sluggish gradients, local optima entrapment, and a mathematical inability to dynamically switch discrete force fields. To address this, we propose PhysAgent, the first simulator-in-the-loop multi-agent framework that leverages multimodal inputs for automated, physically grounded 4D synthesis. By decoupling intrinsic materials from extrinsic dynamics, PhysAgent utilizes a Semantic Agent equipped with an externalized Force Field Skill module to master simulation rules and generate valid initializations. Subsequently, the Refine Agents, driven by Trajectory-Grounded Multi-Agent Feedback, leverage vision foundation models to extract dense point trajectories from rendered frames. By converting these explicit motion trajectories into structured textual descriptors, the agent harnesses LLM commonsense reasoning to execute zero-shot macroscopic leaps, effectively escaping local optima and dynamically switching discrete force fields. Extensive experiments demonstrate that PhysAgent rapidly generates stable, diverse physical scenes from arbitrary multimodal prompts, significantly outperforming existing baselines in both generation diversity and physical accuracy.

cs.RO↗

arXiv2026

HACK++: Towards More Effective Head-Aware Key-Value Compression for Efficient Visual Autoregressive Modeling

Visual Autoregressive (VAR) models adopt a next-scale prediction paradigm, offering high-quality generation with substantially fewer decoding steps. However, existing VAR models suffer from significant attention complexity and severe memory overhead due to the accumulation of key-value (KV) caches across scales. In this paper, we tackle this challenge by introducing KV cache compression into the next-scale paradigm. We begin with an in-depth analysis of VAR attention and observe that attention heads can be stably divided into two functionally distinct categories: Contextual Heads focus on maintaining semantic consistency, while Structural Heads preserve spatial coherence. Their functional divergence makes existing one-size-fits-all compression methods perform poorly on VAR models. We further find that the two head types differ markedly in their reliance on historical scales, and that this reliance shifts across layers and generation steps, arguing for an adaptive cache budget allocation. To address these challenges, we propose HACK++, a training-free Head-Aware key-value Compression frameworK for VAR models. From a one-time offline calibration, HACK++ classifies head types and derives head-specific priors. At inference, it decouples attention from cache compression under independent budgets, bounding the current-scale attention cost while compressing the accumulated cache far more aggressively, via pattern-specific strategies and a reliance-aware budget allocation. Extensive experiments on multiple VAR models across text-to-image, class-conditional, and unified understanding-and-generation tasks validate the effectiveness and generalizability of HACK++. For example, on Infinity-2B/8B, HACK++ maintains near-lossless generation with only a 30% attention budget and a 10% cache budget, and remains robust even under a 1% cache budget.

cs.CV↗

arXiv2026

Rec-Distill: An Industrial Distillation Pipeline for Large-Scale Recommendation Models

Large recommendation models have demonstrated substantial potential gains under scaling laws, yet these gains are difficult to realize in industrial recommendation systems because real-world deployment requires lightweight models with strict serving efficiency and latency guarantees. This creates a fundamental gap between offline model scaling and online deployment. In this work, we present Rec-Distill, an industrial distillation pipeline that transfers the performance gains of large-scale recommendation modeling to efficient serving models. Rec-Distill combines large-teacher scaling with student-side transfer optimization through decoupled training, black-box distillation, debiasing mechanism, and a hybrid batch-streaming pipeline for dynamic recommendation environments. Across multiple recommendation and advertising scenarios on real-world platforms, our framework scales teacher models up to 24B dense parameters and 20K behavior sequence length, while enabling lightweight students to recover a substantial portion of teacher gains, with distillation transferability exceeding 60% in the best setting. Extensive offline and online experiments further show that these transferred gains consistently translate into measurable business improvements under industrial constraints. These results demonstrate that Rec-Distill provides a practical framework for distilling large-scale recommendation models into deployable, cost-efficient serving systems, while also establishing a reliable path toward scaling recommendation models to even larger regimes in the future.

cs.IR↗

arXiv2026

Verifiable Model-Free Safety Filters via Reinforcement Learning

This paper presents a reinforcement learning approach of a model-free safety filter, drawing inspiration from the framework of model-based Predictive Safety Filters (PSFs). Similar to conventional PSFs, our method adopts a Quadratic Programming (QP) formulation by representing the filter as an unrolled QP solver network. However, unlike existing PSFs that derive QP parameters explicitly from system models, we learn these parameters directly through Deep Reinforcement Learning (DRL), thereby eliminating the dependency on accurate system identification. Furthermore, compared to traditional neural network-based methods, this QP structure allows us to furnish a formal certificate for the persistent safety of the learned filter. Numerical results demonstrate that our method outperforms both conventional model-based PSFs and RL-trained Multi-Layer Perceptron (MLP) baselines in terms of safety guarantees, minimal intervention, and per-step computational load.

math.OC↗

arXiv2026

Staged Laser Wakefield Acceleration for Saturated Lasing of Bandwidth-Tunable Free-Electron Lasers from EUV to X-ray

Free-electron lasers (FELs) provide a revolutionary tool for capturing the structure and dynamics of matter in real time at the atomic scale. The size and cost of FELs can be substantially reduced by using laser wakefield acceleration (LWFA), which offers acceleration gradients orders of magnitude beyond radiofrequency technology, producing multi-GeV electron beams within tens of centimeters. This compactness opens the possibility of integrating multiple operating modes - from the EUV to X-rays including broadband operation - into one facility. Realizing this vision, however, faces key challenges: current LWFA bunches are too short to sustain sufficient radiation slippage, limiting FEL pulse energy at EUV wavelengths, while the large energy spread and emittance make X-ray lasing even more demanding. Here we present a LWFA-driven FEL scheme that addresses these challenges, enabling multi-mode operation spanning different wavelengths and bandwidths within a single facility. The scheme employs staged acceleration to reach multi-GeV energies while preserving beam quality, combined with a dual-chicane beamline that stretches the bunch to mitigate the radiation slippage for EUV FEL and tailors the energy chirp for diverse FEL bandwidth modes. Simulations demonstrate that the scheme can generate high-quality electron beams with energies up to 7 GeV and tunable energy chirp, enabling both FEL saturation from the EUV to X-ray wavelengths and large bandwidth operation with a bandwidth of up to 11%. This work provides a roadmap for compact, multi-mode FELs based on plasma acceleration, and the high-energy, high-quality beams achieved also point toward compact injectors for next-generation storage-ring light sources.

physics.acc-ph↗

arXiv2026

MDL: A Unified Multi-Distribution Learner in Large-scale Industrial Recommendation through Tokenization

Industrial recommender systems increasingly adopt multi-scenario learning (MSL) and multi-task learning (MTL) to handle diverse user interactions and contexts, but existing approaches suffer from two critical drawbacks: (1) underutilization of large-scale model parameters due to limited interaction with complex feature modules, and (2) difficulty in jointly modeling scenario and task information in a unified framework. To address these challenges, we propose a unified \textbf{M}ulti-\textbf{D}istribution \textbf{L}earning (MDL) framework, inspired by the "prompting" paradigm in large language models (LLMs). MDL treats scenario and task information as specialized tokens rather than auxiliary inputs or gating signals. Specifically, we introduce a unified information tokenization module that transforms features, scenarios, and tasks into a unified tokenized format. To facilitate deep interaction, we design three synergistic mechanisms: (1) feature token self-attention for rich feature interactions, (2) domain-feature attention for scenario/task-adaptive feature activation, and (3) domain-fused aggregation for joint distribution prediction. By stacking these interactions, MDL enables scenario and task information to "prompt" and activate the model's vast parameter space in a bottom-up, layer-wise manner. Extensive experiments on real-world industrial datasets demonstrate that MDL significantly outperforms state-of-the-art MSL and MTL baselines. Online A/B testing on Douyin Search platform over one month yields +0.0626\% improvement in LT30 and -0.3267\% reduction in change query rate. MDL has been fully deployed in production, serving hundreds of millions of users daily.

cs.IR↗

arXiv2026

TokenMixer-Large: Scaling Up Large Ranking Models in Industrial Recommenders

While scaling laws for recommendation models have gained significant traction, existing architectures such as Wukong, HiFormer and DHEN, often struggle with sub-optimal designs and hardware under-utilization, limiting their practical scalability. Our previous TokenMixer architecture (introduced in RankMixer paper) addressed effectiveness and efficiency by replacing self-attention with a ightweight token-mixing operator; however, it faced critical bottlenecks in deeper configurations, including sub-optimal residual paths, vanishing gradients, incomplete MoE sparsification and constrained scalability. In this paper, we propose TokenMixer-Large, a systematically evolved architecture designed for extreme-scale recommendation. By introducing a mixing-and-reverting operation, inter-layer residuals and the auxiliary loss, we ensure stable gradient propagation even as model depth increases. Furthermore, we incorporate a Sparse Per-token MoE to enable efficient parameter expansion. TokenMixer-Large successfully scales its parameters to 7-billion and 15-billion on online traffic and offline experiments, respectively. Currently deployed in multiple scenarios at ByteDance, TokenMixer-Large has achieved significant offline and online performance gains, delivering an increase of +1.66\% in orders and +2.98\% in per-capita preview payment GMV for e-commerce, improving ADSS by +2.0\% in advertising and achieving a +1.4\% revenue growth for live streaming.

cs.IR↗

arXiv2026

Zenith: Scaling up Ranking Models for Billion-scale Livestreaming Recommendation

Accurately capturing feature interactions is essential in recommender systems, and recent trends show that scaling up model capacity could be a key driver for next-level predictive performance. While prior work has explored various model architectures to capture multi-granularity feature interactions, relatively little attention has been paid to efficient feature handling and scaling model capacity without incurring excessive inference latency. In this paper, we address this by presenting Zenith, a scalable and efficient ranking architecture that learns complex feature interactions with minimal runtime overhead. Zenith is designed to handle a few high-dimensional Prime Tokens with Token Fusion and Token Boost modules, which exhibits superior scaling laws compared to other state-of-the-art ranking methods, thanks to its improved token heterogeneity. Its real-world effectiveness is demonstrated by deploying the architecture to TikTok Live, a leading online livestreaming platform that attracts billions of users globally. Our A/B test shows that Zenith achieves +1.05%/-1.10% in online CTR AUC and Logloss, and realizes +9.93% gains in Quality Watch Session / User and +8.11% in Quality Watch Duration / User.

cs.LG↗

arXiv2025

RankMixer: Scaling Up Ranking Models in Industrial Recommenders

Recent progress on large language models (LLMs) has spurred interest in scaling up recommendation systems, yet two practical obstacles remain. First, training and serving cost on industrial Recommenders must respect strict latency bounds and high QPS demands. Second, most human-designed feature-crossing modules in ranking models were inherited from the CPU era and fail to exploit modern GPUs, resulting in low Model Flops Utilization (MFU) and poor scalability. We introduce RankMixer, a hardware-aware model design tailored towards a unified and scalable feature-interaction architecture. RankMixer retains the transformer's high parallelism while replacing quadratic self-attention with multi-head token mixing module for higher efficiency. Besides, RankMixer maintains both the modeling for distinct feature subspaces and cross-feature-space interactions with Per-token FFNs. We further extend it to one billion parameters with a Sparse-MoE variant for higher ROI. A dynamic routing strategy is adapted to address the inadequacy and imbalance of experts training. Experiments show RankMixer's superior scaling abilities on a trillion-scale production dataset. By replacing previously diverse handcrafted low-MFU modules with RankMixer, we boost the model MFU from 4.5\% to 45\%, and scale our ranking model parameters by 100x while maintaining roughly the same inference latency. We verify RankMixer's universality with online A/B tests across two core application scenarios (Recommendation and Advertisement). Finally, we launch 1B Dense-Parameters RankMixer for full traffic serving without increasing the serving cost, which improves user active days by 0.3\% and total in-app usage duration by 1.08\%.

cs.IR↗

arXiv2025

New Determination of the $^{14}$C(n, $γ$)$^{15}$C Reaction Rate and Its Astrophysical Implications

We present a novel experiment to investigate the spectroscopic factor of the $^{15}$C ground state for the first time using single-neutron $removal$ transfer reactions on $^{15}$C. Two consistent spectroscopic factors were derived from the (p, d) and (d, t) reactions, which were subsequently used to deduce the $^{14}$C(n, $γ$)$^{15}$C reaction cross section and the corresponding stellar reaction rate. A typical cross section of (3.89 $\pm$ 0.76) $μ$b is determined at $E_\mathrm{_{c.m.}}$ = 23.3 keV. At the temperature range of 0.01-4 GK, our new reaction rate is 2.4-3.7 times higher than that of the first direct measurement and 20\%-25\% lower than that of the most recent direct measurement, respectively. Moreover, it is interesting that we can associate a long-standing nuclear structure issue, i.e., the so-called ``quenching'' effect, with this astrophysically relevant reaction. Finally, motivated by astrophysical interests of this reaction decades ago, implications of our new rate on several astrophysical problems are evaluated using state-of-the-art theoretical models. Our calculations demonstrate that the abundances of $^{14}$N and $^{15}$N can be enhanced in the inner regions of asymptotic giant branch (AGB) stars, though with minimal impact on the chemical compositions of the interstellar medium. In the inhomogeneous Big Bang nucleosynthesis, the updated reaction rate can lead to a $\sim 20\%$ variation in the final yields of $^{15}$N in neutron rich regions. For the $r$-process in the core-collapse supernovae, a slight difference of $\sim 0.2\%$ in the final abundances of heavy elements with $A > 90$ can be found by using our new rate.

nucl-ex↗

arXiv2025

COIG-P: A High-Quality and Large-Scale Chinese Preference Dataset for Alignment with Human Values

Aligning large language models (LLMs) with human preferences has achieved remarkable success. However, existing Chinese preference datasets are limited by small scale, narrow domain coverage, and lack of rigorous data validation. Additionally, the reliance on human annotators for instruction and response labeling significantly constrains the scalability of human preference datasets. To address these challenges, we design an LLM-based Chinese preference dataset annotation pipeline with no human intervention. Specifically, we crawled and carefully filtered 92k high-quality Chinese queries and employed 15 mainstream LLMs to generate and score chosen-rejected response pairs. Based on it, we introduce COIG-P (Chinese Open Instruction Generalist - Preference), a high-quality, large-scale Chinese preference dataset, comprises 1,009k Chinese preference pairs spanning 6 diverse domains: Chat, Code, Math, Logic, Novel, and Role. Building upon COIG-P, to reduce the overhead of using LLMs for scoring, we trained a 8B-sized Chinese Reward Model (CRM) and meticulously constructed a Chinese Reward Benchmark (CRBench). Evaluation results based on AlignBench \citep{liu2024alignbenchbenchmarkingchinesealignment} show that that COIG-P significantly outperforms other Chinese preference datasets, and it brings significant performance improvements ranging from 2% to 12% for the Qwen2/2.5 and Infinity-Instruct-3M-0625 model series, respectively. The results on CRBench demonstrate that our CRM has a strong and robust scoring ability. We apply it to filter chosen-rejected response pairs in a test split of COIG-P, and our experiments show that it is comparable to GPT-4o in identifying low-quality samples while maintaining efficiency and cost-effectiveness. Our codes and data are released in https://github.com/multimodal-art-projection/COIG-P.

cs.CL↗

arXiv2025

Notions of Stack-manipulating Computation and Relative Monads (Extended Version)

Monads provide a simple and concise interface to user-defined computational effects in functional programming languages. This enables equational reasoning about effects, abstraction over monadic interfaces and the development of monad transformer stacks to combine different effects. Compiler implementors and assembly code programmers similarly virtualize effects, and would benefit from similar abstractions if possible. However, the implementation details of effects seem disconnected from the high-level monad interface: at this lower level much of the design is in the layout of the runtime stack, which is not accessible in a high-level programming language. We demonstrate that the monadic interface can be faithfully adapted from high-level functional programming to a lower level setting with explicit stack manipulation. We use a polymorphic call-by-push-value (CBPV) calculus as a setting that captures the essence of stack-manipulation, with a type system that allows programs to define domain-specific stack structures. Within this setting, we show that the existing category-theoretic notion of a relative monad can be used to model the stack-based implementation of computational effects. To demonstrate generality, we adapt a variety of standard monads to relative monads. Additionally, we show that stack-manipulating programs can benefit from a generalization of do-notation we call "monadic blocks" that allow all CBPV code to be reinterpreted to work with an arbitrary relative monad. As an application, we show that all relative monads extend automatically to relative monad transformers, a process which is not automatic for monads in pure languages.

cs.PL↗

arXiv2025

A Knowledge Distillation-Based Approach to Enhance Transparency of Classifier Models

With the rapid development of artificial intelligence (AI), especially in the medical field, the need for its explainability has grown. In medical image analysis, a high degree of transparency and model interpretability can help clinicians better understand and trust the decision-making process of AI models. In this study, we propose a Knowledge Distillation (KD)-based approach that aims to enhance the transparency of the AI model in medical image analysis. The initial step is to use traditional CNN to obtain a teacher model and then use KD to simplify the CNN architecture, retain most of the features of the data set, and reduce the number of network layers. It also uses the feature map of the student model to perform hierarchical analysis to identify key features and decision-making processes. This leads to intuitive visual explanations. We selected three public medical data sets (brain tumor, eye disease, and Alzheimer's disease) to test our method. It shows that even when the number of layers is reduced, our model provides a remarkable result in the test set and reduces the time required for the interpretability analysis.

cs.AI↗

arXiv2025

Simulations of Common Unsupervised Domain Adaptation Algorithms for Image Classification

Traditional machine learning assumes that training and test sets are derived from the same distribution; however, this assumption does not always hold in practical applications. This distribution disparity can lead to severe performance drops when the trained model is used in new data sets. Domain adaptation (DA) is a machine learning technique that aims to address this problem by reducing the differences between domains. This paper presents simulation-based algorithms of recent DA techniques, mainly related to unsupervised domain adaptation (UDA), where labels are available only in the source domain. Our study compares these techniques with public data sets and diverse characteristics, highlighting their respective strengths and drawbacks. For example, Safe Self-Refinement for Transformer-based DA (SSRT) achieved the highest accuracy (91.6\%) in the office-31 data set during our simulations, however, the accuracy dropped to 72.4\% in the Office-Home data set when using limited batch sizes. In addition to improving the reader's comprehension of recent techniques in DA, our study also highlights challenges and upcoming directions for research in this domain. The codes are available at https://github.com/AIPMLab/Domain_Adaptation.

cs.LG↗

arXiv2024

Siamese Multiple Attention Temporal Convolution Networks for Human Mobility Signature Identification

The Human Mobility Signature Identification (HuMID) problem stands as a fundamental task within the realm of driving style representation, dedicated to discerning latent driving behaviors and preferences from diverse driver trajectories for driver identification. Its solutions hold significant implications across various domains (e.g., ride-hailing, insurance), wherein their application serves to safeguard users and mitigate potential fraudulent activities. Present HuMID solutions often exhibit limitations in adaptability when confronted with lengthy trajectories, consequently incurring substantial computational overhead. Furthermore, their inability to effectively extract crucial local information further impedes their performance. To address this problem, we propose a Siamese Multiple Attention Temporal Convolutional Network (Siamese MA-TCN) to capitalize on the strengths of both TCN architecture and multi-head self-attention, enabling the proficient extraction of both local and long-term dependencies. Additionally, we devise a novel attention mechanism tailored for the efficient aggregation of multi-scale representations derived from our model. Experimental evaluations conducted on two real-world taxi trajectory datasets reveal that our proposed model effectively extracts both local key information and long-term dependencies. These findings highlight the model's outstanding generalization capabilities, demonstrating its robustness and adaptability across datasets of varying sizes.

cs.AI↗

arXiv2024

FedVAE: Trajectory privacy preserving based on Federated Variational AutoEncoder

The use of trajectory data with abundant spatial-temporal information is pivotal in Intelligent Transport Systems (ITS) and various traffic system tasks. Location-Based Services (LBS) capitalize on this trajectory data to offer users personalized services tailored to their location information. However, this trajectory data contains sensitive information about users' movement patterns and habits, necessitating confidentiality and protection from unknown collectors. To address this challenge, privacy-preserving methods like K-anonymity and Differential Privacy have been proposed to safeguard private information in the dataset. Despite their effectiveness, these methods can impact the original features by introducing perturbations or generating unrealistic trajectory data, leading to suboptimal performance in downstream tasks. To overcome these limitations, we propose a Federated Variational AutoEncoder (FedVAE) approach, which effectively generates a new trajectory dataset while preserving the confidentiality of private information and retaining the structure of the original features. In addition, FedVAE leverages Variational AutoEncoder (VAE) to maintain the original feature space and generate new trajectory data, and incorporates Federated Learning (FL) during the training stage, ensuring that users' data remains locally stored to protect their personal information. The results demonstrate its superior performance compared to other existing methods, affirming FedVAE as a promising solution for enhancing data privacy and utility in location-based applications.

cs.AI↗