arXiv subjects

Yue Cheng

Publications and source records attributed to Yue Cheng.

At least 19 recordsLinked to original sources

arXiv2026

Learning Dynamics of Logits Debiasing for Long-Tailed Semi-Supervised Learning

Long-tailed distributions are prevalent in real-world semi-supervised learning (SSL), where pseudo-labels tend to favor majority classes, leading to degraded generalization. While many long-tailed semi-supervised learning (LTSSL) methods have been proposed, the mechanisms by which they implicitly debias logits remain poorly understood. In this work, we revisit LTSSL through the lens of learning dynamics and provide a theoretical characterization of logits debiasing. Specifically, we derive a step-wise decomposition of the logits updates, showing that predictions are dominated by class-imbalance bias that reliably reflects label priors. To expose this effect, we use the logits of a task-irrelevant baseline image as an indicator of accumulated bias and prove that they converge to the class prior. This provides a unified view where LTSSL remedies such as logit adjustment, reweighting, and resampling correspond to reshaping gradient dynamics. Based on this insight, we propose DyTrim, a principle-based dynamic pruning framework that reallocates gradient budget through class-aware pruning on labeled data and confidence-based soft pruning on unlabeled data. We provide theoretical guarantees that DyTrim reduces class bias and improves generalization. Extensive experiments on standard LTSSL benchmarks show consistent gains across architectures and methods. Code available at: https://jiajun0425.github.io/DyTrim

cs.LG↗

arXiv2026

DSETA: A Dual-Stage Continual Learning Framework for Travel Time Prediction in Dynamic Traffic Environments

Estimated Time of Arrival (ETA) prediction is a core component of intelligent transportation systems. As traffic congestion patterns become increasingly dynamic in large cities, maintaining high prediction accuracy poses a major challenge for ride-hailing platforms. Existing methods either fail to adapt to irregular traffic patterns and sudden congestion, or suffer from new distributions without disentangling long-term trends from short-term fluctuations, thereby degrading model performance in real-world scenarios. To address this challenge, we propose DSETA, an incrementally updated Dual-Stage ETA prediction framework. Specifically, the continual learning process is divided into \textit{inter-day} and \textit{intra-day} stages. We first design the \textit{intra-day} learning stage, which relies entirely on real-time data to enable dynamic adaptation to short-term traffic patterns caused by events like holidays or accidents. Next, we develop the \textit{inter-day} learning stage, which leverages aggregated historical data from a short time window to capture knowledge of long-term distribution shifts, such as seasonal trends and traffic network evolution. Subsequently, to prevent catastrophic forgetting and preserve knowledge of regular patterns, we explore a \textit{Historical Traffic Knowledge Consolidation} module. Finally, we validate DSETA's effectiveness and robustness through extensive offline and online experiments conducted on real-world datasets from DiDi's platform. Online A/B tests across three major cities including Beijing, Wuhan, and Xi'an consistently demonstrated performance gains, achieving MAE reductions of 6.62\%, 0.73\%, and 2.40\% respectively. This framework has been successfully deployed in DiDi's production environment, processing hundreds of millions of daily requests and validating its strong performance in industrial applications.

cs.LG↗

arXiv2026

[AAFLOW+] Stateful Operator Abstraction with Zero-Copy Distributed KV Cache Orchestration for Multi-Agent Workflows

Multi-agent LLM systems increasingly integrate retrieval, planning, and reasoning, but remain fundamentally text-centric, requiring agents to repeatedly recompute shared context through expensive prefill. Although single-request inference is known to be accelerated by KV-cache management, it is usually restricted to local serving scopes. We introduce AAFLOW+, a stateful extension of agentic workflow operators that makes KV cache a first-class distributed systems object. AAFLOW+ builds processes into communication-aware graphs that concurrently optimize data, prompts, and reusable model state. It also provides operators for KV materialization, transfer, fork, composition, and eviction. Its runtime enables zero-copy, transfer-aware execution, allowing agents to reuse long context without recomputation. AAFLOW+ reduces TTFT by up to 50.2x, achieves up to 7.63x reduced multi-agent compute cost at 16-agent scale, reduces KV memory by 1.72-6.10x, and increases throughput by more than 7.74x, based on an analytical cost model parameterized by empirical hardware microbenchmarks. The results demonstrate that KV transmission outperforms recomputation on networks with moderate to high bandwidth, making sure KV-state sharing greatly increases efficiency in multi-agent LLM systems by replacing text passing.

cs.DC↗

arXiv2026

Mechanistically Interpreting the Role of Sample Difficulty in RLVR for LLMs

Reinforcement Learning with Verifiable Reward (RLVR) is empirically shown to notably enhance the reasoning performance of large language models (LLMs), particularly in mathematics and programming. However, the mechanistic role of Sample Difficulty in RLVR remains poorly understood. In this paper, we investigate RLVR through the lens of difficulty-wise and one-sample analysis. We find that sample difficulty has a non-monotonic effect on RLVR: easy and medium-difficulty problems yield the strongest and most stable reasoning improvements, whereas overly hard problems often provide weak learning signals, induce degenerate behaviors such as answer repetition or skipping necessary computation, and can ultimately degrade the model's pre-existing capabilities. Beyond the obverse of response, we further analyze the model's internal feature dynamics using Temporal Sparse Autoencoders (T-SAE). Easy problems mainly reinforce direct-answer and basic-computation features while suppressing deliberative-reasoning features; hard problems activate reasoning-related features but become useful only when successful trajectories are sampled; medium-difficulty problems provide a more balanced signal, strengthening both computation and multi-step reasoning features. Motivated by these findings, we propose difficulty-adaptive strategies for hard-sample utilization, using backward-reasoning reformulation and T-SAE-guided training signals to improve reward density and credit assignment during RLVR. Overall, our results identify sample difficulty as a key factor governing both the optimization dynamics and representation evolution of RLVR.

cs.AI↗

arXiv2026

LatentBox: Storing AI-Generated Images at Scale via a Latent-First Design

The explosive growth of AI-generated images has created a sustainability challenge for storage infrastructure. Platforms like Midjourney and Adobe Firefly already host billions of generative images, yet conventional object stores persist them as blobs with full-resolution pixels, consuming huge amounts of storage capacity and bandwidth. Unlike natural photos, however, AI-generated images can be deterministically reconstructed from compact, model-native latent tensors, making persistent image storage fundamentally redundant. This paper presents LatentBox, a latent-first storage system for AI-generated images. LatentBox treats compressed latents as durable storage objects and uses on-demand GPU reconstruction on the read path to trade inexpensive compute for large persistent storage savings. Our design is guided by the first large-scale analysis of AI-generated image access we are aware of, based on a 35-month, 2-billion-request production trace from a major generative-content platform. Motivated by the trace analysis, LatentBox keeps frequently accessed images in decoded pixel format for fast hits, stores less-active objects as compressed latents to expand effective cache capacity, and continuously adjusts the splits between the image and latent cache to optimize user-perceived access latency.We build a LatentBox prototype and evaluate it with the production trace. LatentBox reduces persistent storage by 78.7% with competitive or even lower mean and tail latency over a pure image-based storage.

cs.DC↗

arXiv2026

C2CServe: Leveraging NVLink-C2C for Elastic Serverless LLM Serving on MIG

Modern LLM serving is increasingly serverless in shape: large model catalogs, long-tail invocations, and multi-tenant demand. Existing GPU serving systems face a tradeoff: dedicated-GPU allocation wastes scarce HBM under sparse traffic, while GPU time sharing places model initialization and weight loading on the cold-start path. Spatial GPU sharing such as multi-instance GPU (MIG) provides isolation and accounting, but each slice has too little HBM for modern LLM weights. We observe that high-bandwidth CPU--GPU interconnects, such as NVLink-C2C (C2C) in NVIDIA GH200 and GB200 Superchips, change the memory constraint: model weights can reside in CPU memory and be streamed on demand to MIG instances, shifting model residency from scarce HBM to abundant host memory. Leveraging this capability, we present C2CServe, a request-granularity serverless LLM serving system that allows MIG instances to switch models across requests without reloading weights into HBM. C2CServe introduces HybridGEMM, a heterogeneous-memory-aware GEMM kernel that adapts data access patterns to balance HBM and C2C bandwidth across MIG partitions using a single tuning knob. To mitigate shared-C2C contention, C2CServe further uses a hierarchical scheduler that coordinates model placement, input chunking, and kernel selection with online feedback control. On GH200, C2CServe reduces cold-start latency by up to 7.1x for dense models and 4.6x for MoE models compared with state-of-the-art serverless LLM serving systems, while maintaining over 95\% TTFT and TPOT attainment under C2C contention.

cs.OS↗

arXiv2026

MoE-Prefill: Zero Redundancy Overheads in MoE Prefill Serving

Production LLM workloads increasingly serve discriminative tasks, such as classification, recommendation, and verification, whose answers are read from the logits of a single prefill pass with no autoregressive decoding. Serving these prefill-only workloads on mixture-of-experts (MoE) models is bottlenecked not by compute but by the distributed execution required to fit the model: existing parallel strategies (tensor, expert, and pipeline parallelism) trade memory pressure for redundant computation, communication, and synchronization, severely degrading MoE prefill serving efficiency. We observe that these overheads stem from coupling expert placement with synchronous activation routing -- a design inherited from the decoding era. The long, compute-bound forward passes of large-batch prefill open a per-layer window wide enough to stream expert weights in the background, replacing per-layer activation AllToAll with asynchronous weight AllGather fully overlapped with computation. We propose MoE-Prefill, a prefill-only serving system whose backend, AsyncEP (Asynchronous Expert Parallelism), gathers experts by weight rather than routing them by activation, and whose frontend co-enforces a physically-derived saturation threshold through prefix-aware routing and true-FLOPs load tracking. On Qwen3-235B-A22B across four hardware/precision configurations, MoE-Prefill delivers 1.35-1.37x throughput over the strongest distributed baseline on real-world workloads and up to 1.59x on long-context synthetic workloads, sustaining 29.8-36.2% per-GPU model FLOPs utilization.

cs.LG↗

arXiv2026

TStore: Rethinking AI Model Hub with Tensor-Centric Compression

Modern AI models are growing rapidly in size and redundancy, leading to significant storage and distribution challenges in model hubs. We present TStore, a tensor-centric system for reducing storage overhead through fine-grained deduplication and compression. TStore leverages tensor-level fingerprinting and clustering to identify redundancy across models without requiring annotations. Our design enables efficient storage reduction while preserving model usability and performance. Experiments on real-world model repositories demonstrate substantial storage savings with minimal overhead.

cs.DC↗

arXiv2026

M3D-Net: Multi-Modal 3D Facial Feature Reconstruction Network for Deepfake Detection

With the rapid advancement of deep learning in image generation, facial forgery techniques have achieved unprecedented realism, posing serious threats to cybersecurity and information authenticity. Most existing deepfake detection approaches rely on the reconstruction of isolated facial attributes without fully exploiting the complementary nature of multi-modal feature representations. To address these challenges, this paper proposes a novel Multi-Modal 3D Facial Feature Reconstruction Network (M3D-Net) for deepfake detection. Our method leverages an end-to-end dual-stream architecture that reconstructs fine-grained facial geometry and reflectance properties from single-view RGB images via a self-supervised 3D facial reconstruction module. The network further enhances detection performance through a 3D Feature Pre-fusion Module (PFM), which adaptively adjusts multi-scale features, and a Multi-modal Fusion Module (MFM) that effectively integrates RGB and 3D-reconstructed features using attention mechanisms. Extensive experiments on multiple public datasets demonstrate that our approach achieves state-of-the-art performance in terms of detection accuracy and robustness, significantly outperforming existing methods while exhibiting strong generalization across diverse scenarios.

cs.CV↗

arXiv2026

Reducing Hallucination in Enterprise AI Workflows via Hybrid Utility Minimum Bayes Risk (HUMBR)

Although LLMs drive automation, it is critical to ensure immense consideration for high-stakes enterprise workflows such as those involving legal matters, risk management, and privacy compliance. For Meta, and other organizations like ours, a single hallucinated clause in such high stakes workflows risks material consequences. We show that by framing hallucination mitigation as a Minimum Bayes Risk (MBR) problem, we can dramatically reduce this risk. Specifically, we introduce a Hybrid Utility MBR (HUMBR) framework that synthesizes semantic embedding similarity with lexical precision to identify consensus without ground-truth references, for which we derive rigorous error bounds. We complement this theoretical analysis with a comprehensive empirical evaluation on widely-used public benchmark suites (TruthfulQA and LegalBench) and also real world data from Meta production deployment. The results from our empirical study show that MBR significantly outperforms standard Universal Self-Consistency. Notably, 81% of the pipeline's suggestions were preferred over human-crafted ground truth, and critical recall failures were virtually eliminated.

cs.LG↗

arXiv2026

ViviDoc: Generating Interactive Documents through Human-Agent Collaboration

Interactive documents help readers engage with complex ideas through dynamic visualization, interactive animations, and exploratory interfaces. However, creating such documents remains costly, as it requires both domain expertise and web development skills. Recent Large Language Model (LLM)-based agents can automate content creation, but directly applying them to interactive document generation often produces outputs that are difficult to control. To address this, we present ViviDoc, to the best of our knowledge the first work to systematically address interactive document generation. ViviDoc introduces a multi-agent pipeline (Planner, Styler, Executor, Evaluator). To make the generation process controllable, we provide three levels of human control: (1) the Document Specification (DocSpec) with SRTC Interaction Specifications (State, Render, Transition, Constraint) for structured planning, (2) a content-aware Style Palette for customizing writing and interaction styles, and (3) chat-based editing for iterative refinement. We also construct ViviBench, a benchmark of 101 topics derived from real-world interactive documents across 11 domains, along with a taxonomy of 8 interaction types and a 4-dimensional automated evaluation framework validated against human ratings (Pearson r > 0.84). Experiments show that ViviDoc achieves the highest content richness and interaction quality in both automated and human evaluation. A 12-person user study confirms that the system is easy to use, provides effective control over the generation process, and produces documents that satisfy users.

cs.HC↗

arXiv2025

Combining Serverless and High-Performance Computing Paradigms to support ML Data-Intensive Applications

Data is found everywhere, from health and human infrastructure to the surge of sensors and the proliferation of internet-connected devices. To meet this challenge, the data engineering field has expanded significantly in recent years in both research and industry. Traditionally, data engineering, Machine Learning, and AI workloads have been run on large clusters within data center environments, requiring substantial investment in hardware and maintenance. With the rise of the public cloud, it is now possible to run large applications across nodes without owning or maintaining hardware. Serverless functions such as AWS Lambda provide horizontal scaling and precise billing without the hassle of managing traditional cloud infrastructure. However, when processing large datasets, users often rely on external storage options that are significantly slower than direct communication typical of HPC clusters. We introduce Cylon, a high-performance distributed data frame solution that has shown promising results for data processing using Python. We describe how we took inspiration from the FMI library and designed a serverless communicator to tackle communication and performance issues associated with serverless functions. With our design, we demonstrate that the scaling efficiency of AWS Lambda achieves within 6.5% of serverful AWS (EC2) at 64 nodes, based on implementing direct communication via NAT Traversal TCP Hole Punching.

cs.DC↗

arXiv2025

Jacobian-Based Interpretation of Nonlinear Neural Encoding Model

In recent years, the alignment between artificial neural network (ANN) embeddings and blood oxygenation level dependent (BOLD) responses in functional magnetic resonance imaging (fMRI) via neural encoding models has significantly advanced research on neural representation mechanisms and interpretability in the brain. However, these approaches remain limited in characterizing the brain's inherently nonlinear response properties. To address this, we propose the Jacobian-based Nonlinearity Evaluation (JNE), an interpretability metric for nonlinear neural encoding models. JNE quantifies nonlinearity by statistically measuring the dispersion of local linear mappings (Jacobians) from model representations to predicted BOLD responses, thereby approximating the nonlinearity of BOLD signals. Centered on proposing JNE as a novel interpretability metric, we validated its effectiveness through controlled simulation experiments on various activation functions and network architectures, and further verified it on real fMRI data, demonstrating a hierarchical progression of nonlinear characteristics from primary to higher-order visual cortices, consistent with established cortical organization. We further extended JNE with Sample-Specificity (JNE-SS), revealing stimulus-selective nonlinear response patterns in functionally specialized brain regions. As the first interpretability metric for quantifying nonlinear responses, JNE provides new insights into brain information processing. Code available at https://github.com/Gaitxh/JNE.

q-bio.NC↗

arXiv2025

HBSplat: Robust Sparse-View Gaussian Reconstruction with Hybrid-Loss Guided Depth and Bidirectional Warping

Novel View Synthesis (NVS) from sparse views presents a formidable challenge in 3D reconstruction, where limited multi-view constraints lead to severe overfitting, geometric distortion, and fragmented scenes. While 3D Gaussian Splatting (3DGS) delivers real-time, high-fidelity rendering, its performance drastically deteriorates under sparse inputs, plagued by floating artifacts and structural failures. To address these challenges, we introduce HBSplat, a unified framework that elevates 3DGS by seamlessly integrating robust structural cues, virtual view constraints, and occluded region completion. Our core contributions are threefold: a Hybrid-Loss Depth Estimation module that ensures multi-view consistency by leveraging dense matching priors and integrating reprojection, point propagation, and smoothness constraints; a Bidirectional Warping Virtual View Synthesis method that enforces substantially stronger constraints by creating high-fidelity virtual views through bidirectional depth-image warping and multi-view fusion; and an Occlusion-Aware Reconstruction component that recovers occluded areas using a depth-difference mask and a learning-based inpainting model. Extensive evaluations on LLFF, Blender, and DTU benchmarks validate that HBSplat sets a new state-of-the-art, achieving up to 21.13 dB PSNR and 0.189 LPIPS, while maintaining real-time inference. Code is available at: https://github.com/eternalland/HBSplat.

cs.CV↗

arXiv2025

WiFiSim: Simulating WiFi Probe Requests via AOSP Analysis and Device Behavior Modeling

WiFi probe request (PR) frames encode fine-grained device interactions and serve as a critical basis for mobility and crowd analytics. However, pervasive MAC address randomization and the scarcity of labeled datasets hinder progress in PR-based studies. We introduce WiFiSim, a simulation framework that reconstructs PR generation through Android Open Source Project (AOSP) protocol analysis and finite-state device behavior modeling. WiFiSim identifies the key determinants of PR structure and timing while capturing realistic user-driven state transitions. Experiments show that WiFiSim achieves less than 5% deviation from real measurements in both distributional and temporal dynamics, scales to large-scale dataset synthesis, and enables reliable evaluation of downstream applications. Source code and sample datasets are publicly released to foster reproducible research.

cs.NI↗

arXiv2025

DTGen: Generative Diffusion-Based Few-Shot Data Augmentation for Fine-Grained Dirty Tableware Recognition

Intelligent tableware cleaning is a critical application in food safety and smart homes, but existing methods are limited by coarse-grained classification and scarcity of few-shot data, making it difficult to meet industrialization requirements. We propose DTGen, a few-shot data augmentation scheme based on generative diffusion models, specifically designed for fine-grained dirty tableware recognition. DTGen achieves efficient domain specialization through LoRA, generates diverse dirty images via structured prompts, and ensures data quality through CLIP-based cross-modal filtering. Under extremely limited real few-shot conditions, DTGen can synthesize virtually unlimited high-quality samples, significantly improving classifier performance and supporting fine-grained dirty tableware recognition. We further elaborate on lightweight deployment strategies, promising to transfer DTGen's benefits to embedded dishwashers and integrate with cleaning programs to intelligently regulate energy consumption and detergent usage. Research results demonstrate that DTGen not only validates the value of generative AI in few-shot industrial vision but also provides a feasible deployment path for automated tableware cleaning and food safety monitoring.

cs.CV↗

arXiv2025

UltraEar: a multicentric, large-scale database combining ultra-high-resolution computed tomography and clinical data for ear diseases

Ear diseases affect billions of people worldwide, leading to substantial health and socioeconomic burdens. Computed tomography (CT) plays a pivotal role in accurate diagnosis, treatment planning, and outcome evaluation. The objective of this study is to present the establishment and design of UltraEar Database, a large-scale, multicentric repository of isotropic 0.1 mm ultra-high-resolution CT (U-HRCT) images and associated clinical data dedicated to ear diseases. UltraEar recruits patients from 11 tertiary hospitals between October 2020 and October 2035, integrating U-HRCT images, structured CT reports, and comprehensive clinical information, including demographics, audiometric profiles, surgical records, and pathological findings. A broad spectrum of otologic disorders is covered, such as otitis media, cholesteatoma, ossicular chain malformation, temporal bone fracture, inner ear malformation, cochlear aperture stenosis, enlarged vestibular aqueduct, and sigmoid sinus bony deficiency. Standardized preprocessing pipelines have been developed for geometric calibration, image annotation, and multi-structure segmentation. All personal identifiers in DICOM headers and metadata are removed or anonymized to ensure compliance with data privacy regulation. Data collection and curation are coordinated through monthly expert panel meetings, with secure storage on an offline cloud system. UltraEar provides an unprecedented ultra-high-resolution reference atlas with both technical fidelity and clinical relevance. This resource has significant potential to advance radiological research, enable development and validation of AI algorithms, serve as an educational tool for training in otologic imaging, and support multi-institutional collaborative studies. UltraEar will be continuously updated and expanded, ensuring long-term accessibility and usability for the global otologic research community.

eess.IV↗

arXiv2025

MorphServe: Efficient and Workload-Aware LLM Serving via Runtime Quantized Layer Swapping and KV Cache Resizing

Efficiently serving large language models (LLMs) under dynamic and bursty workloads remains a key challenge for real-world deployment. Existing serving frameworks and static model compression techniques fail to adapt to workload fluctuations, leading to either service-level objective (SLO) violations under full-precision serving or persistent accuracy degradation with static quantization. We present MorphServe, a dynamic, workload-aware LLM serving framework based on morphological adaptation. MorphServe introduces two asynchronous, token-level runtime mechanisms: quantized layer swapping, which selectively replaces less impactful layers with quantized alternatives during high-load periods, and pressure-aware KV cache resizing, which dynamically adjusts KV cache capacity in response to memory pressure. These mechanisms enable state-preserving transitions with minimum runtime overhead and are fully compatible with modern scheduling and attention techniques. Extensive experiments on Vicuna and Llama family models with real-world workloads demonstrate that MorphServe reduces average SLO violations by 92.45 percent and improves the P95 TTFT latency by 2.2x-3.9x compared to full-precision serving, without compromising generation quality. These results establish MorphServe as a practical and elastic solution for LLM deployment in dynamic environments.

cs.DC↗