arXiv subjects

Li Chen

Publications and source records attributed to Li Chen.

At least 145 records · Page 8Linked to original sources

arXiv2025

Communication-Efficient Serving for Video Diffusion Models with Latent Parallelism

Video diffusion models (VDMs) perform attention computation over the 3D spatio-temporal domain. Compared to large language models (LLMs) processing 1D sequences, their memory consumption scales cubically, necessitating parallel serving across multiple GPUs. Traditional parallelism strategies partition the computational graph, requiring frequent high-dimensional activation transfers that create severe communication bottlenecks. To tackle this issue, we exploit the local spatio-temporal dependencies inherent in the diffusion denoising process and propose Latent Parallelism (LP), the first parallelism strategy tailored for VDM serving. \textcolor{black}{LP decomposes the global denoising problem into parallelizable sub-problems by dynamically rotating the partitioning dimensions (temporal, height, and width) within the compact latent space across diffusion timesteps, substantially reducing the communication overhead compared to prevailing parallelism strategies.} To ensure generation quality, we design a patch-aligned overlapping partition strategy that matches partition boundaries with visual patches and a position-aware latent reconstruction mechanism for smooth stitching. Experiments on three benchmarks demonstrate that LP reduces communication overhead by up to 97\% over baseline methods while maintaining comparable generation quality. As a non-intrusive plug-in paradigm, LP can be seamlessly integrated with existing parallelism strategies, enabling efficient and scalable video generation services.

cs.DC↗

arXiv2025

Model Whisper: Steering Vectors Unlock Large Language Models' Potential in Test-time

It is a critical challenge to efficiently unlock the powerful reasoning potential of Large Language Models (LLMs) for specific tasks or new distributions. Existing test-time adaptation methods often require tuning model parameters, which is not only computationally expensive but also risks degrading the model's pre-existing abilities.To address this, we introduce a lightweight component, Test-Time Steering Vectors (TTSV), which is prepended to the input while keeping the LLM's parameters entirely frozen. By optimizing the TTSV on test data to minimize the model's output entropy, we steer the model towards an internal state of higher confidence, activating its inherent abilities most relevant to the current task. TTSV is both lightweight and highly efficient to optimize, making it a true plug-and-play enhancement. Extensive experiments validate our approach's effectiveness on both base models and reasoning-enhanced models. For instance, on the MATH500 task, TTSV achieves a 45.88% relative performance gain on the Qwen2.5-Math-7B model and a 16.22% relative gain on the Qwen3-4B model. Furthermore, our approach exhibits robust generalization, with its steering vectors proving highly transferable across diverse tasks.

cs.CL↗

arXiv2025

Test-time Correction: An Online 3D Detection System via Visual Prompting

This paper introduces Test-time Correction (TTC), an online 3D detection system designed to rectify test-time errors using various auxiliary feedback, aiming to enhance the safety of deployed autonomous driving systems. Unlike conventional offline 3D detectors that remain fixed during inference, TTC enables immediate online error correction without retraining, allowing autonomous vehicles to adapt to new scenarios and reduce deployment risks. To achieve this, we equip existing 3D detectors with an Online Adapter (OA) module -- a prompt-driven query generator for real-time correction. At the core of OA module are visual prompts: image-based descriptions of objects of interest derived from auxiliary feedback such as mismatches with 2D detections, road descriptions, or user clicks. These visual prompts, collected from risky objects during inference, are maintained in a visual prompt buffer to enable continuous correction in future frames. By leveraging this mechanism, TTC consistently detects risky objects, achieving reliable, adaptive, and versatile driving autonomy. Extensive experiments show that TTC significantly improves instant error rectification over frozen 3D detectors, even under limited labels, zero-shot settings, and adverse conditions. We hope this work inspires future research on post-deployment online rectification systems for autonomous driving.

cs.CV↗

arXiv2025

Evaluating Spatio-Temporal Forecasting Trade-offs Between Graph Neural Networks and Foundation Models

Modern IoT deployments for environmental sensing produce high volume spatiotemporal data to support downstream tasks such as forecasting, typically powered by machine learning models. While existing filtering and strategic deployment techniques optimize collected data volume at the edge, they overlook how variations in sampling frequencies and spatial coverage affect downstream model performance. In many forecasting models, incorporating data from additional sensors denoise predictions by providing broader spatial contexts. This interplay between sampling frequency, spatial coverage and different forecasting model architectures remain underexplored. This work presents a systematic study of forecasting models - classical models (VAR), neural networks (GRU, Transformer), spatio-temporal graph neural networks (STGNNs), and time series foundation models (TSFMs: Chronos Moirai, TimesFM) under varying spatial sensor nodes density and sampling intervals using real-world temperature data in a wireless sensor network. Our results show that STGNNs are effective when sensor deployments are sparse and sampling rate is moderate, leveraging spatial correlations via encoded graph structure to compensate for limited coverage. In contrast, TSFMs perform competitively at high frequencies but degrade when spatial coverage from neighboring sensors is reduced. Crucially, the multivariate TSFM Moirai outperforms all models by natively learning cross-sensor dependencies. These findings offer actionable insights for building efficient forecasting pipelines in spatio-temporal systems. All code for model configurations, training, dataset, and logs are open-sourced for reproducibility: https://github.com/UIUC-MONET-Projects/Benchmarking-Spatiotemporal-Forecast-Models

cs.LG↗

arXiv2025

A Multi-Stage Optimization Framework for Deploying Learned Image Compression on FPGAs

Deep learning-based image compression (LIC) has achieved state-of-the-art rate-distortion (RD) performance, yet deploying these models on resource-constrained FPGAs remains a major challenge. This work presents a complete, multi-stage optimization framework to bridge the gap between high-performance floating-point models and efficient, hardware-friendly integer-based implementations. First, we address the fundamental problem of quantization-induced performance degradation. We propose a Dynamic Range-Aware Quantization (DRAQ) method that uses statistically-calibrated activation clipping and a novel weight regularization scheme to counteract the effects of extreme data outliers and large dynamic ranges, successfully creating a high-fidelity 8-bit integer model. Second, building on this robust foundation, we introduce two hardware-aware optimization techniques tailored for FPGAs. A progressive mixed-precision search algorithm exploits FPGA flexibility to assign optimal, non-uniform bit-widths to each layer, minimizing complexity while preserving performance. Concurrently, a channel pruning method, adapted to work with the Generalized Divisive Normalization (GDN) layers common in LIC, removes model redundancy by eliminating inactive channels. Our comprehensive experiments show that the foundational DRAQ method reduces the BD-rate overhead of a GDN-based model from $30\%$ to $6.3\%$. The subsequent hardware-aware optimizations further reduce computational complexity by over $20\%$ with negligible impact on RD performance, yielding a final model that is both state-of-the-art in efficiency and superior in quality to existing FPGA-based LIC implementations.

cs.CV↗

arXiv2025

A catalog of new blue stragglers in open clusters with Gaia DR3

The high-precision {\it Gaia} data release 3 (DR3) enables the discovery of numerous open clusters in the Milky Way, providing an excellent opportunity to search for blue straggler stars in open clusters and investigate their formation and evolution in these environments. Using the member stars from literature open cluster catalogs, we visually inspected the color-magnitude diagram (CMD) of each cluster and selected cluster candidates that potentially host blue stragglers. We then reassessed cluster memberships using the {\tt pyUPMASK} algorithm with {\it Gaia} DR3 and performed isochrone fitting to derive physical parameters for each cluster, including age, distance modulus, mean reddening, and metallicity. Finally, we empirically identified straggler stars based on their positions relative to the best-fitting isochrone, zero-age main sequence (ZAMS), and equal-mass binary sequence on the CMD. In total, we identified 272 new straggler stars in 99 open clusters, comprising 153 blue stragglers, 98 probable blue stragglers, and 21 yellow stragglers. Compared to the reported blue straggler catalogs based on earlier {\it Gaia} data, our results increase the number of open clusters with stragglers in the Milky Way by 22.2\%, and the total number of blue stragglers by 11.2\%.

astro-ph.SR↗

arXiv2025

Scaffolding Metacognition in Programming Education: Understanding Student-AI Interactions and Design Implications

Generative AI tools such as ChatGPT now provide novice programmers with unprecedented access to instant, personalized support. While this holds clear promise, their influence on students' metacognitive processes remains underexplored. Existing work has largely focused on correctness and usability, with limited attention to whether and how students' use of AI assistants supports or bypasses key metacognitive processes. This study addresses that gap by analyzing student-AI interactions through a metacognitive lens in university-level programming courses. We examined more than 10,000 dialogue logs collected over three years, complemented by surveys of students and educators. Our analysis focused on how prompts and responses aligned with metacognitive phases and strategies. Synthesizing these findings across data sources, we distill design considerations for AI-powered coding assistants that aim to support rather than supplant metacognitive engagement. Our findings provide guidance for developing educational AI tools that strengthen students' learning processes in programming education.

cs.HC↗

arXiv2025

The ALMA-ATOMS-QUARKS survey: Resolving a chemically rich massive protostellar outflow

We present a comprehensive study on the physical and chemical structures of a chemically rich bipolar outflow in a high-mass star forming region IRAS 16272$-$4837 (SDC335), utilizing high-resolution spectral line data at 1.3 mm and 3 mm dual-bands from the ALMA ATOMS and QUARKS surveys. The high-velocity jet is enveloped by a lower-velocity outflow cavity, containing bright knots that show enhanced molecular intensities and elevated excitation temperatures. Along the outflow, we have identified 35 transitions from 22 molecular species. By analyzing the spatial distribution and kinematics of these molecular lines, we find that the molecular inventory in the outflow is regulated by three processes: (i) direct entrainment from the natal molecular core by the outflow; (ii) shock-induced release of molecules or atoms from dust grains; and (iii) thermal desorption and gas-phase reactions driven by shock heating. These results confirm that outflows are not only dynamical structures but also active chemical factories, where entrainment, shocks, and thermal processing jointly enrich the molecular content. Our findings confirmed that outflow chemistry has multi-origin nature, and provide critical insights into chemical evolution during high-mass star formation.

astro-ph.GA↗

arXiv2025

DMA: Online RAG Alignment with Human Feedback

Retrieval-augmented generation (RAG) systems often rely on static retrieval, limiting adaptation to evolving intent and content drift. We introduce Dynamic Memory Alignment (DMA), an online learning framework that systematically incorporates multi-granularity human feedback to align ranking in interactive settings. DMA organizes document-, list-, and response-level signals into a coherent learning pipeline: supervised training for pointwise and listwise rankers, policy optimization driven by response-level preferences, and knowledge distillation into a lightweight scorer for low-latency serving. Throughout this paper, memory refers to the model's working memory, which is the entire context visible to the LLM for In-Context Learning. We adopt a dual-track evaluation protocol mirroring deployment: (i) large-scale online A/B ablations to isolate the utility of each feedback source, and (ii) few-shot offline tests on knowledge-intensive benchmarks. Online, a multi-month industrial deployment further shows substantial improvements in human engagement. Offline, DMA preserves competitive foundational retrieval while yielding notable gains on conversational QA (TriviaQA, HotpotQA). Taken together, these results position DMA as a principled approach to feedback-driven, real-time adaptation in RAG without sacrificing baseline capability.

cs.AI↗

arXiv2025

Statistical Rounding Error Analysis for Random Matrix Computations

The conventional rounding error analysis provides worst-case bounds with an associated failure probability and ignores the statistical property of the rounding errors. In this paper, we develop a new statistical rounding error analysis for random matrix computations. Such computations have numerous applications in the field of wireless communications, signal processing, and machine learning. By assuming the relative errors are independent random variables, we derive the approximate closed-form expressions for the expectation and variance of the rounding errors in various key computations for random matrices. Numerical experiments validate the accuracy of our derivations and demonstrate that our analytical expressions are generally at least two orders of magnitude tighter than alternative worst-case bounds, exemplified through the inner products.

math.NA↗

arXiv2025

Sim-to-Real Gentle Manipulation of Deformable and Fragile Objects with Stress-Guided Reinforcement Learning

Robotic manipulation of deformable and fragile objects presents significant challenges, as excessive stress can lead to irreversible damage to the object. While existing solutions rely on accurate object models or specialized sensors and grippers, this adds complexity and often lacks generalization. To address this problem, we present a vision-based reinforcement learning approach that incorporates a stress-penalized reward to discourage damage to the object explicitly. In addition, to bootstrap learning, we incorporate offline demonstrations as well as a designed curriculum progressing from rigid proxies to deformables. We evaluate the proposed method in both simulated and real-world scenarios, showing that the policy learned in simulation can be transferred to the real world in a zero-shot manner, performing tasks such as picking up and pushing tofu. Our results show that the learned policies exhibit a damage-aware, gentle manipulation behavior, demonstrating their effectiveness by decreasing the stress applied to fragile objects by 36.5% while achieving the task goals, compared to vanilla RL policies.

cs.RO↗

arXiv2025

Curvilinear Structure-preserving Unpaired Cross-domain Medical Image Translation

Unpaired image-to-image translation has emerged as a crucial technique in medical imaging, enabling cross-modality synthesis, domain adaptation, and data augmentation without costly paired datasets. Yet, existing approaches often distort fine curvilinear structures, such as microvasculature, undermining both diagnostic reliability and quantitative analysis. This limitation is consequential in ophthalmic and vascular imaging, where subtle morphological changes carry significant clinical meaning. We propose Curvilinear Structure-preserving Translation (CST), a general framework that explicitly preserves fine curvilinear structures during unpaired translation by integrating structure consistency into the training. Specifically, CST augments baseline models with a curvilinear extraction module for topological supervision. It can be seamlessly incorporated into existing methods. We integrate it into CycleGAN and UNSB as two representative backbones. Comprehensive evaluation across three imaging modalities: optical coherence tomography angiography, color fundus and X-ray coronary angiography demonstrates that CST improves translation fidelity and achieves state-of-the-art performance. By reinforcing geometric integrity in learned mappings, CST establishes a principled pathway toward curvilinear structure-aware cross-domain translation in medical imaging.

cs.CV↗

arXiv2025

GADA: Graph Attention-based Detection Aggregation for Ultrasound Video Classification

Medical ultrasound video analysis is challenging due to variable sequence lengths, subtle spatial cues, and the need for interpretable video-level assessment. We introduce GADA, a Graph Attention-based Detection Aggregation framework that reformulates video classification as a graph reasoning problem over spatially localized regions of interest. Rather than relying on 3D CNNs or full-frame analysis, GADA detects pathology-relevant regions across frames and represents them as nodes in a spatiotemporal graph, with edges encoding spatial and temporal dependencies. A graph attention network aggregates these node-level predictions through edge-aware attention to generate a compact, discriminative video-level output. Evaluated on a large-scale, multi-center clinical lung ultrasound dataset, GADA outperforms conventional baselines on two pathology video classification tasks while providing interpretable region- and frame-level attention.

eess.IV↗

arXiv2025

From Generic to Specialized: A Subspecialty Diagnostic System Powered by Self-Supervised Learning for Cervical Histopathology

Cervical cancer remains a major malignancy, necessitating extensive and complex histopathological assessments and comprehensive support tools. Although deep learning shows promise, these models still lack accuracy and generalizability. General foundation models offer a broader reach but remain limited in capturing subspecialty-specific features and task adaptability. We introduce the Cervical Subspecialty Pathology (CerS-Path) diagnostic system, developed through two synergistic pretraining stages: self-supervised learning on approximately 190 million tissue patches from 140,000 slides to build a cervical-specific feature extractor, and multimodal enhancement with 2.5 million image-text pairs, followed by integration with multiple downstream diagnostic functions. Supporting eight diagnostic functions, including rare cancer classification and multimodal Q&A, CerS-Path surpasses prior foundation models in scope and clinical applicability. Comprehensive evaluations demonstrate a significant advance in cervical pathology, with prospective testing on 3,173 cases across five centers maintaining 99.38% screening sensitivity and excellent generalizability, highlighting its potential for subspecialty diagnostic translation and cervical cancer screening.

cs.CV↗

arXiv2025

A Statistical Framework for Co-Mediators of Zero-Inflated Single-Cell RNA-Seq Data

Single-cell RNA sequencing (scRNA-seq) has revolutionized the study of cellular heterogeneity, enabling detailed molecular profiling at the individual cell level. However, integrating high-dimensional single-cell data into causal mediation analysis remains challenging due to zero inflation and complex mediator structures. We propose a novel mediation framework leveraging zero-inflated negative binomial models to characterize cell-level mediator distributions and beta regression for zero-inflation proportions. The model can identify expression level as well as expressed proportion that could mediate disease-leading causal pathway. Extensive simulation studies demonstrate improved power and controlled false discovery rates. We further illustrate the utility of this approach through application to ROSMAP single-cell transcriptomic data, uncovering biologically meaningful mediation effects that enhance understanding of disease mechanisms.

stat.ME↗

arXiv2025

Robust Majorana Platform Driven by a Meissner-Induced Anisotropic Doppler Shift

The realization of robust Majorana zero modes (MZMs), a cornerstone for fault-tolerant quantum computing, is hindered by the challenge of creating a platform that simultaneously offers a large topological gap, high tunability, and resilience to disorder. A system unifying these properties has remained elusive. Here, we propose and validate a novel platform that harnesses the Meissner effect in a topological insulator (TI) nanowire partially covered by a superconducting (SC) layer. Under an external magnetic field, Meissner screening currents in the SC induce a spatially varying Doppler shift on the TI surface. This effect generates a highly anisotropic effective g-factor, which selectively drives a topological phase transition localized on the nanowire's bottom surface. This mechanism is crucial as it spatially separates the topological phase from the SC/TI interface, permitting strong proximity-induced superconductivity while preventing detrimental band renormalization at the interface from closing the topological gap. Furthermore, by confining the topological superconducting phase to the gate-tunable bottom surface, our platform fully leverages the intrinsic disorder resilience of the TI's topologically protected surface states. Through a combination of supercurrent simulations, self-consistent Schrödinger-Poisson calculations, and large-scale tight-binding computations, we validate the platform's robustness. Our work establishes a practical pathway toward Meissner-mediated topological superconductivity for realizing robust MZMs in SC/TI hybrid systems.

cond-mat.mes-hall↗

arXiv2025

Dual Reinforcement Learning Synergy in Resource Allocation: Emergence of Self-Organized Momentum Strategy

In natural ecosystems and human societies, self-organized resource allocation and policy synergy are ubiquitous and significant. This work focuses on the synergy between Dual Reinforcement Learning Policies in the Minority Game (DRLP-MG) to optimize resource allocation. Our study examines a mixed-structured population with two sub-populations: a Q-subpopulation using Q-learning policy and a C-subpopulation adopting the classical policy. We first identify a synergy effect between these subpopulations. A first-order phase transition occurs as the mixing ratio of the subpopulations changes. Further analysis reveals that the Q-subpopulation consists of two internal synergy clusters (IS-clusters) and a single external synergy cluster (ES-cluster). The former contribute to the internal synergy within the Q-subpopulation through synchronization and anti-synchronization, whereas the latter engages in the inter-subpopulation synergy. Within the ES-cluster, the classical momentum strategy in the financial market manifests and assumes a crucial role in the inter-subpopulation synergy. This particular strategy serves to prevent long-term under-utilization of resources. However, it also triggers trend reversals and leads to a decrease in rewards for those who adopt it. Our research reveals that the frozen effect, in either the C- or Q-subpopulation, is a crucial prerequisite for synergy, consistent with previous studies. We also conduct mathematical analyses on subpopulation synergy effects and the synchronization and anti-synchronization forms of IS-clusters in the Q-subpopulation. Overall, our work comprehensively explores the complex resource-allocation dynamics in DRLP-MG, uncovers multiple synergy mechanisms and their conditions, enriching the theoretical understanding of reinforcement-learning-based resource allocation and offering valuable practical insights

nlin.AO↗

arXiv2025

A Data-Driven Adaptive Impedance Matching Method Robust to Parasitic Effects

Adaptive impedance matching between antennas and radio frequency front-end (RFFE) power modules is essential for mobile communication systems. To address the matching performance degradation caused by parasitic effects in practical tunable matching networks (TMNs), this paper proposes a data-driven adaptive impedance matching method that avoids physical adjustment. First, we propose the residual enhanced circuit behavior modeling network (RECBM-Net), a deep learning model that maps TMN operating states to their scattering parameters (S-parameters). Then, we formulate the matching process based on the trained surrogate model as a mathematical optimization problem. We employ two classic numerical methods with different online computational overhead, namely simulated annealing particle swarm optimization (SAPSO) and adaptive moment estimation with automatic differentiation (AD-Adam), to search for the matching solution. To further reduce the online inference overhead caused by repeated forward propagation through RECBM-Net, we train an inverse mapping solver network (IMS-Net) to directly predict the optimal solution. Simulation results show that RECBM-Net accurately predicts S-parameters, achieving a mean absolute error of $6.98 \times 10^{-5}$. Across 9000 mismatched scenarios, the compliance rate after tuning increases from 0.97% with the analytical solution of the ideal L-network to 95.92% with SAPSO, 93.42% with AD-Adam, and 95% with IMS-Net. While AD-Adam significantly reduces computational overhead, lowering the average number of RECBM-Net inferences from 2097 with SAPSO to 285, it sacrifices some accuracy. IMS-Net requires only a single inference to obtain the matching solution, resulting in minimal online overhead while maintaining excellent matching accuracy.

eess.SP↗