arXiv subjects

Yifei Wang

Publications and source records attributed to Yifei Wang.

At least 37 records · Page 2Linked to original sources

arXiv2026

Reliability Testing of Medical Model Performance under Distributed Deployment

Distributed inference has become an indispensable part of deploying medical models under practical latency, memory, and throughput constraints. Although modern frameworks improve serving efficiency through tensor parallelism, mixed precision, kernel fusion, and multi-device communication, they are generally assumed to preserve the behavior observed during centralized HuggingFace evaluation. This assumption creates an evaluation-deployment mismatch: a model may pass offline evaluation but produce a different output after the execution stack changes. To address this mismatch, we propose a testing framework and an improved, distributed-execution-sensitive medical-model benchmark that evaluates the same checkpoint and input under a centralized HuggingFace reference and matched distributed deployments. Extensive experiments across language, vision, and multimodal medical models show that execution changes can produce measurable output disagreements. Across supported visual settings, the test success rate ranges from 0.21 to 0.43 for single-modality models and from 0.32 to 0.98 for multimodal models. The benchmark is aimed at extending medical-model evaluation from capability and security to evaluation-deployment consistency.

eess.IV↗

arXiv2026

Scaling Video Generation for Reasoning: At What Cost?

We study whether scaling video generation enables models to reason about hidden information from the past frames, and at what computational cost. Our controlled benchmark requires predicting nine prescribed moves of an initially solved 2x2x2 Rubik's Cube from a fixed view of three faces. Correct predictions require inferring how actions change hidden states, and the simulator provides exact ground truth for evaluation. Models learn plausible cube geometry early, while correct sticker configurations require substantially more training. Although validation MSE follows approximate power-law scaling, lower MSE loss does not reliably indicate downstream reasoning capabilities. Smaller autoregressive models achieve higher state accuracy with limited compute, while larger models reach higher accuracy after more training. At roughly 0.1 PF-days, the 70M-parameter model correctly predicts the visible sticker configuration in 44.6% of post-action frames, compared with 0.3% for the 1B model, which reaches 83.7% at 3.14 PF-days. Symbolic state supervision raises the 20M model's frame accuracy from 31.1% to 67.3% at the same training-data budget, suggesting that learning representations of state changes can complement scaling.

cs.CV↗

arXiv2026

No More K-means: Single-Stage Sparse Coding for Efficient Multi-Vector Retrieval

Multi-vector retrieval (MVR) models, exemplified by ColBERT, have established new benchmarks in retrieval accuracy by preserving fine-grained token-level interactions. However, this granularity imposes prohibitive storage and retrieval efficiency bottlenecks: to manage the immense memory footprint and computational overhead of billion-scale token vectors, state-of-the-art systems are forced to rely on aggressive dimension reduction and complex clustering (e.g., K-means). This compromise introduces two critical limitations: excessive indexing latency of clustering large-scale corpora and semantic information loss inherent to compression. In this paper, we propose Single-stage Sparse Retrieval (SSR}, a paradigm shift that replaces expensive clustering with efficient sparse coding. Instead of compressing features into low-dimensional dense vectors, we utilize Sparse Autoencoder (SAE) to project token embeddings into a high-dimensional but highly sparse representation. This transformation enables us to bypass vector clustering entirely and leverage inverted indexing for precise, high-throughput retrieval. Extensive experiments on the BEIR benchmark demonstrate that SSR achieves a "trifecta" of improvements: it reduces indexing time by 15x compared to ColBERTv2, halves retrieval latency, and simultaneously improves retrieval performance over leading baselines.

cs.IR↗

arXiv2026

What Paired Evaluations Reveal under Visual Perturbations

Robustness evaluation must examine diverse visual perturbations, while benchmarks cover only some real-world conditions and physical testing is costly. Paired evaluations link clean and perturbed predictions for the same image, capturing changes in correctness, confidence, and acceptance beyond aggregate accuracy. We investigate how this image correspondence supports two needs in robustness evaluation: interpreting paired evaluation results and prioritizing samples for physical testing. To interpret paired evaluation results, we fix both sets of prediction records and vary their correspondence within each class. We prove that classwise correct-correct counts give the same sharp bounds on lost acceptance and mean true-class probability decrease among retained-correct inputs as any feasible five-state refinement. Distinguishing persistent from changed wrong answers can further constrain accepted-error transitions, while shared correspondence can establish policy orderings left unresolved by separate cost intervals. To prioritize samples for physical testing, we retain each image's synthetic responses and rank clean-correct images by their mean true-class probability under corruption. Across 44 classifiers, testing the highest-risk 20% finds 67% and 45% of failures under mild screen and print recaptures, versus 58% and 36% for clean confidence and 60% and 37% for an equal-size natural-transformation average. With both probability averaging and an A3Rank scoring adaptation, the tested corruption set yields higher mean failure recall than the natural-transform set; differences between scores depend on the source and budget. Together, these findings show that the value of correspondence depends on the evaluation objective: classwise counts suffice for specified reliability bounds, while image-specific synthetic responses improve the allocation of physical tests within the evaluated pool.

cs.CV↗

arXiv2026

Representation by Design in Generation: Cross-View Class-Token Alignment in Diffusion Transformers

Generative and representation learning remain asymmetrically connected: semantic representations are used to improve diffusion generation, whereas the models' own representations are often treated as a by-product of synthesis. We ask whether diffusion models can instead be trained to learn substantially stronger semantic representations without sacrificing generation quality. SelfFlow takes a step in this direction by introducing self-supervised patch alignment into flow matching, but its main gains remain in faster convergence and improved generation. Inspired by DINO and iBOT, we extend this framework with cross-view class-token alignment to further strengthen semantic representations. Specifically, we form two independently noised, dual-timestep observations of each image and align each student class-token representation with the stop-gradient EMA-teacher target from the other observation. This objective is optimized jointly with the inherited flow-matching and local patch objectives. Notably, although the additional objective acts only on the class token, it strengthens both class-token and patch representations. Compared with a matched two-view baseline, ImageNet linear-probing accuracy improves by 9.4\% using the class token and 10.1\% using mean-pooled patch tokens, while frozen-backbone VOC2012 segmentation improves by 3.6 mIoU. These representation gains are achieved while maintaining comparable ImageNet generation FID. In text-to-image training, the same objective also improves generation FID, reducing it from 2.52 to 2.37 at matched checkpoints. Our results show that representation need not remain a by-product of generation or merely a tool for improving it: it can be directly optimized as a first-class capability of diffusion pretraining alongside generation.

cs.CV↗

arXiv2026

Compress to Remember: Learning Compact Memory via On-Policy Distillation for Long Video Generation

Standard video generators do not natively compact historical context into reusable memory tokens. As generation continues, the growing history makes it increasingly difficult to retain information from earlier frames due to long-context degradation. Key-frame-based approaches address this challenge by retaining selected past frames, but can discard information needed for future generation. Rather than relying on frame selection alone, we study whether a frozen video generator can supply the supervision needed to learn a compact representation of the history. We propose Prediction-Aligned Context Compaction (PACC), which uses a learned compressor to aggregate information across past frames into compact memory tokens. We train the compressor through on-policy distillation, using the same frozen generator both as a student when conditioned on compressed memory and as a teacher when conditioned on the full history. The student generates continuations, while the teacher provides targets for the same noisy inputs at each denoising step. Only the compressor is updated to align the student's predictions with these targets. We evaluate PACC on MBench, which jointly measures memory-event coverage and consistency. PACC outperforms the strongest baseline by 6.63 points on Causal-rCM and 3.19 points on Causal Forcing. Evaluation on VBench-Long using MovieGen prompts further shows that PACC produces minute-long videos with generation quality competitive with baselines. Together, these results show that learning to compact historical context can improve long-video memory without modifying the underlying generator.

cs.CV↗

arXiv2026

Beyond Memory Construction: Rethinking Memory Access for LLM-based Conversational Agents

Memory is a core component of conversational agents, enabling coherent and context-aware behavior over long interactions. Recent approaches commonly rely on LLM-based memory construction, where raw interactions are rewritten into structured memory units and later retrieved via a RAG pipeline. While effective in controlled settings, we show that this paradigm breaks down in long-horizon, high-entropy conversations: memory construction becomes increasingly lossy and unstable as context length and information complexity grow, and incurs prohibitive cost due to repeated LLM invocation. To address these limitations, we propose Threader, a memory system that shifts the focus from memory construction to efficient, structure-aware access over raw interactions. Instead of rewriting interactions, Threader preserves them as first-class memory, organizes them into topic-coherent segments via lightweight incremental segmentation, and enables accurate retrieval through multi-view representation. At query time, it performs multi-signal retrieval that combines segment-level access with localized evidence matching, ensuring both completeness and coherence. Extensive experiments demonstrate that Threader consistently improves answer accuracy and evidence recall, while significantly reducing the memory construction overhead.

cs.CL↗

arXiv2026

Using LLMs in Software Design: An Empirical Study of GitHub and A Practitioner Survey

Recent advancements in Large Language Models (LLMs) have demonstrated significant potential across software engineering tasks, including software design, an area traditionally regarded as highly dependent on human expertise and judgment. However, limited research has examined how LLMs are used in software design, aswell as the associated benefits and drawbacks. This paper addresses this gap by empirically investigating how software developers use LLMs in software design. We conducted a mixed-methods study, combining a mining study of 291 developer-ChatGPT conversations shared on GitHub with a survey of 65 software practitioners. From the mined conversations, we identified nine categories of design tasks supported by ChatGPT, including architecture design, data model design, and the use of design patterns.We further characterize developer-ChatGPT interactions, showing that developers in the mined conversations primarily use ChatGPT for knowledge acquisition and designrelated code generation, with most tasks situated at the detailed design level. The survey participants reported seven key benefits, such as better technology selection and early detection of design flaws, and six limitations, including lengthy outputs, inexecutable or incorrect code, and dependence on project context. These findings provide an evidence-based characterization of current LLM use in software design from both open-source and practitioner perspectives.

cs.SE↗

arXiv2026

Compositional Reasoning in Language Models under Reinforcement Learning Post-Training

Compositional reasoning is critical for real-world problem solving: since training data is necessarily limited, models must generalize by composing learned skills in new ways. While post-training methods such as reinforcement learning (RL) have substantially improved the reasoning abilities of language models (LMs), their effects on compositional reasoning remain less well understood. We propose a dependency-graph framework to formalize compositional reasoning, yielding three levels of compositionality with increasing complexity. Empirically, we instantiate this framework with data-structure tasks, which provide deterministic reward computation and clear compositional structure. We find a consistent decomposed-to-composed asymmetry: decomposed-skill training does not reliably transfer to composed tasks, whereas composed-task training transfers more readily back to decomposed tasks. We provide theoretical explanation for this asymmetry, and further evaluate compositional generalization under length extrapolation, structural distribution shift, and transfer to tasks requiring unseen skills. Finally, we present a pilot study on real-world tool-calling benchmarks, showing preliminary evidence that the decomposed-to-composed asymmetry can extend to practical settings.

cs.AI↗

arXiv2026

Towards Requirements Engineering for GenAI-Enabled Software: Bridging Responsibility Gaps through Human Oversight Requirements

Context: Responsibility gaps, long-recognized challenges in socio-technical systems where accountability becomes diffuse or ambiguous, have become increasingly pronounced in GenAI-enabled software. This study seeks to establish a coherent perspective for a systematic analysis of responsibility gaps from a human oversight requirements standpoint, encompassing how these responsibility gaps should be conceptualized, identified, and represented throughout the requirements engineering process. Methods: The proposed design methodology is structured across three analytical layers. At the conceptualization layer, it establishes a conceptual framing that defines the key elements of responsibility across the human and system dimensions and explains how potential responsibility gaps emerge from their interactions. At the methodological layer, it introduces a deductive pipeline for identifying responsibility gaps by analyzing interactions between these dimensions and deriving corresponding oversight requirements within established requirements engineering frameworks. At the artifact layer, it formalizes the results in a Deductive Backbone Table, a reusable representation that traces the reasoning path from responsibility gaps identification to human oversight requirements derivation. Results: A user study compared the proposed methodology with a baseline goal-oriented requirements engineering approach. Participant-perception results indicated generally favorable views of the enhanced approach across the evaluated dimensions, while expert ratings of the generated requirement artifacts showed higher mean scores for six of the seven artifact-quality items, with the strongest descriptive improvements observed for consistency and representational traceability.

cs.SE↗

arXiv2026

A Theoretical Framework for Masked Pretraining (MPT)

Recently, Masked Pretraining (MPT) based on reconstruction pretraining tasks has risen to a promising self-supervised learning paradigm across various domains and achieves remarkable performance in multiple downstream tasks. However, the theoretical understanding of the working mechanism behind MPT is still limited. In this paper, we introduce a new theoretical framework to analyze MPT and understand the crucial role of masking in extracting meaningful representations. We establish theoretical connections between MPT and another popular self-supervised paradigm: contrastive learning. We prove that the masking technique implicitly creates positive pairs that are semantically similar and the reconstruction loss pulls them together in the feature space. Besides, as a result of the implicit alignment, we point out the dimensional collapse issue of MPT and propose a Uniformity-enhanced MPT (U-MPT) loss that can effectively address this issue and bring significant improvements in downstream tasks including linear evaluation, cross-dataset fine-tuning and out-of-distribution generalization on real-world data sets. Furthermore, we establish downstream guarantees of U-MPT and theoretically analyze the influence of masking strategies. Based on the theoretical analysis, we propose a new masking strategy which enhances the downstream performance of MPT and explains current improvements of masking strategies with our theoretical perspective.

cs.LG↗

arXiv2026

Elite-Weighted Supervised Fine-tuning for Goal-Directed Molecular Optimization

Goal-directed optimization is essential for steering molecular generators to propose candidates with desired properties. However, it is often implemented with policy-gradient reinforcement learning, which requires a generation-trajectory log-probability whose form depends on the model architecture and generation procedure. This makes an optimizer difficult to reuse across architectures and conditional generative designs. Supervised fine-tuning needs none of that machinery, but its update is driven by a fixed dataset, so the reward never enters the update. We introduce Elite-Weighted Supervised Fine-tuning (EW-SFT), which uses reward to guide elite selection of high-scoring molecules, and updates the model by its own pretraining loss on that set. Ablations show that reward information is passed primarily through elite selection, rather than through continuous weighting within the selected set. Because the update consumes only scored molecules and the model's native loss, the same rule applies across autoregressive, masked-diffusion, and discrete-flow generators, and across de novo, motif-extension, and linker-design tasks. Under a fixed budget of 3D shape alignment oracle calls on two kinase reference compounds, EW-SFT consistently outperforms the corresponding native optimizers. It further improves goal-directed optimization under a 2D similarity oracle on four held-out references and achieves comparable performance on a sample-efficiency benchmark without a trajectory-level RL formulation. These results demonstrate that EW-SFT is a unified and effective optimizer across molecular generators, design constraints, references, and oracles.

cs.LG↗

arXiv2026

Decoupling 2D translation-invariant topological CSS codes

Two-dimensional translation-invariant topological CSS codes on qubits are known to be locally equivalent, after coarse-graining, to stacks of toric codes. However, existing constructions generally break more translation symmetry than is required to remove anyon-permuting translations, leaving open whether any further obstruction exists. We prove that no such obstruction occurs: after passing to the maximal anyon-preserving superlattice, every such code admits a local unitary decoupling into toric codes and product states. We further provide an efficient algorithm for explicitly constructing the decoupling map, together with bounds on the required supercell size and operator spreading. The decoupling requires no additional ancillas in generic cases and extends to finite systems with suitable boundary conditions.

quant-ph↗

arXiv2026

NFR-to-Code Traceability in a Blockchain-IoT System: An Empirical Study

Requirement-to-Code traceability has been widely studied, yet existing research and public benchmarks remain largely centered on functional requirements (FRs). In contrast, traceability for non-functional requirements (NFRs) remains more difficult and underexplored, which hinders the verification of critical quality concerns such as security and reliability.This paper studies NFR-to-Code traceability based on a real-world blockchain-IoT project. We design an annotation protocol for constructing trace links across heterogeneous artifacts and build a manually curated subset containing both FR and NFR links. Using this subset, we examine the characteristics of NFR traceability and further evaluate four representative retrieval baselines: TF-IDF, BM25, LSI, and WMD. The results show that FR-to-Code tracing is consistently easier than NFR-to-Code tracing, while security-related NFRs are the most difficult subset. They further indicate that the main challenge of NFR traceability lies not in requirement availability, but in implementation evidence that is distributed and not clearly localized in code.

cs.SE↗

arXiv2026

TransBiolab: A Real-World Multi-View Dataset of Cluttered Transparent Biomedical Objects

Autonomous biomedical laboratories increasingly rely on visual perception to recognize, localize, and manipulate transparent plasticware, yet high-quality real-world datasets for this setting remain limited. The scarcity of domain-relevant data is particularly restrictive in cluttered multi-object scenes, where mutual occlusion and view-dependent appearance changes remain challenging even for contemporary visual foundation models. Existing transparent-object datasets have advanced segmentation, depth, and pose estimation, but they usually do not evaluate the combined setting of multi-object clutter, occlusion, and calibrated multi-view capture that characterizes real laboratory manipulation scenes. To address this gap, we present TrainsBiolab, a real-world RGB-D dataset of cluttered transparent biomedical objects captured as calibrated multi-view sequences. TrainsBiolab contains 161,315 frames from 98 scenes and 1.03M instance annotations over 15 laboratory object types, including 6D poses, full and visible masks, depth, and per-frame camera calibration. The dataset is organized along three axes that reflect operational difficulty: object category, the total number of objects in a frame, and camera viewpoint. We further define dataset-centric benchmarks for segmentation, depth estimation and completion, and 6D pose estimation, and report a system-level robot manipulation evaluation enabled by the released annotations and calibrations. By focusing on repeated transparent instances, clutter, and multi-view laboratory capture, TrainsBiolab provides a resource for segmentation, depth estimation, 6D pose estimation, and multi-view reasoning in autonomous laboratory manipulation. Project page: https://dualtransparency.github.io/TransBiolab/.

cs.CV↗

arXiv2026

Joint Energy Management and Coordinated AIGC Workload Scheduling for Distributed Data Centers: A Diffusion-Aided Reward Shaping Approach

Artificial intelligence-generated content (AIGC) has emerged as a transformative paradigm for automating the creation of diverse and customized content, giving rise to rapidly growing computational workloads in cloud data centers. It is imperative for AIGC service providers (ASPs) to strategically schedule AIGC workloads to reduce data center energy costs while guaranteeing high-quality content generation. However, the distinctive characteristics of AIGC services pose critical challenges, including model heterogeneity across ASPs, implicit service quality evaluation, and complex inference process control. To tackle these challenges, we propose a joint energy management and coordinated AIGC workload scheduling framework, which introduces an explicit mathematical characterization of service quality to promote both job transfer among ASPs and fine-grained inference process configuration. Moreover, various energy resources within data centers are jointly considered to enhance power usage flexibility. Subsequently, a system utility maximization problem is formulated to balance AIGC service revenue with operational penalties and costs. Nevertheless, the strong coupling among job scheduling decisions induces severe reward sparsity, which limits the effectiveness of existing deep reinforcement learning (DRL) algorithms. To address this issue, we develop a diffusion model-aided reward shaping approach to synthesize complementary reward signals through a multi-step denoising process. This approach is seamlessly integrated with DRL to enable efficient learning of scheduling policies under sparse environmental feedback. Experiments based on real-world models and datasets demonstrate that our scheme effectively accommodates electricity price fluctuations and AIGC model heterogeneity, while achieving superior learning convergence and system utility compared with benchmark methods.

cs.LG↗

arXiv2026

Towards Automated Identification of Violation Symptoms of Architecture Erosion

Architecture erosion harms software maintenance when implementation deviates from intended architecture. Early detection of violation symptoms is essential, and code review comments provide a promising textual source for automated identification. This paper investigates automatic detection of architecture violation symptoms from code reviews using traditional machine learning (ML), deep learning (DL), and large language models (LLMs). We built and evaluated 15 ML and 4 DL classifiers with word embeddings (word2vec, fastText, GloVe) on code reviews from four open-source projects (OpenStack Nova, Neutron, Qt Base, Creator). We also constructed LLM-based classifiers using GPT-4o, Qwen-2.5, and DeepSeek-R1. Results show that SVM with word2vec performs best among ML/DL methods (F1-score: 0.808), with 200-dimensional embeddings generally superior. Majority-vote ensembles further improve the performance. On an imbalanced test set, SVM remains strongest in Macro-F1. Statistical tests confirm significant performance differences. The best LLM-based classifiers outperformed traditional ML/DL approaches, with GPT-4o achieving the highest F1-score (0.851). Practitioner surveys and interviews indicate practical value for early warning of architectural issues. A controlled experiment shows that providing detected violation symptoms improves developers' detection rate from 25.9% to 64.7%. The study contributes an automated detection approach, comprehensive ML/DL vs LLM comparison, and practitioner-validated evidence for improving architectural conformance and sustainability.

cs.SE↗

arXiv2026

Improvement of a focused ion beam fabricated diamond pillar for scanning ensemble nitrogen-vacancy magnetometry probe using an ultrapure diamond

Scanning diamond nitrogen-vacancy probe microscopy (SNVM) is an important tool for studying nanoscale condensed-matter phenomena. Ga$^+$-ion-focused-ion-beam (FIB) milling has been introduced as a method for fabricating SNVM, while the probe diameter is limited to a few micrometers due to Ga$^+$-induced damage. We report a method for improving the quality of FIB-fabricated SNVM probes by polyvinyl alcohol and Pt/Pd capping, followed by post-fabrication UV/ozone exposure. The effectiveness of the method is confirmed by the fabrication of an 800 nm-diameter probe with shallow NV centers, demonstrating an imaging of maze-like magnetic domain structure with a resolution of a few hundred nanometers and an improved sensitivity of 6.7 $μ$T/Hz${1/2}$, while preserving spin coherence properties.

cond-mat.mtrl-sci↗