About

Dr. Yubei Chen is an Assistant Professor in the Department of Electrical and Computer Engineering at UC Davis. His research is at the intersection of computational neuroscience and deep unsupervised/self-supervised learning, sensorimotor representation, NeuroAI, and computational neuroscience. His work focuses on world models and efficient deep learning, contributing to the understanding and development of AI systems that integrate biological principles and computational techniques.

Research topics

• Computer Science
• Artificial Intelligence
• Machine Learning
• Algorithm
• Mathematics

Selected publications

• Latent profile analysis of nurses’ moral resilience and its influencing factors

  Nursing Ethics · 2025-08-16 · 1 citations

  articleCorresponding

  Background Nurses often face complex moral distress in clinical practice, where maintaining emotional stability and making moral decisions under pressure is crucial for patient care and safety. Moral resilience helps nurses stay composed, think rationally, and navigate moral challenges effectively. Exploring the characteristics and influencing factors of different moral resilience subgroups can provide insights into nurses’ resilience levels and inform targeted interventions to enhance their occupational well-being. Aim To identify the latent profile model of moral resilience in nurses and analyze the characteristic differences and influencing factors across different profiles. Design A multicenter cross-sectional study. Participants and research context The study involved nurse practitioners from nine different hospitals. A total of 1098 nurses from different specialties participated in the study. Ethical considerations The study was approved by the ethical board of the 900 Hospital of the Joint Service Support Force. Findings Nurses’ moral resilience was classified into three profiles: Vulnerable (46.45%), Discrepant (12.02%), and Robust (41.53%). Divorced or widowed nurses ( p = .017, Z = 2.395) and those working in infectious disease departments ( p < .001, Z = 3.796) were more likely to belong to the Robust profile. Higher moral distress increased the likelihood of being in the Vulnerable profile ( p = .017/<.001, Z = −2.379/−7.001). A stronger ethical climate ( p < .001, Z = 4.540) and greater workplace trust ( p = .010, Z = 2.586) were more likely to belong to the Robust profile. Nurses with greater moral courage were more likely to have robust moral resilience and less likely to be in the Discrepant profile ( p = .007/<.001, Z = 2.687/−5.225). Conclusion Nurses’ moral resilience is heterogeneous, highlighting the need for managers to consider profile characteristics and specialty-specific factors. Strengthening the hospital ethical climate can reduce moral distress, enhance workplace trust, and foster moral courage, ultimately improving moral resilience in nurses.

  PublisherDOI

• Workplace trust and moral resilience in nurses: The parallel mediating roles of moral courage, ethical climate and social support

  Nurse Education Today · 2025-12-05

  article
  PublisherDOI

• Task-Driven Semantic Quantization and Imitation Learning for Goal-Oriented Communications

  2025-06-08

  article

  Semantic communication marks a new paradigm shift from bit-wise data transmission to semantic information delivery for the purpose of bandwidth reduction. To more effectively carry out specialized downstream tasks at the receiver end, it is crucial to define the most critical semantic message in the data based on the task or goal-oriented features. In this work, we propose a novel goal-oriented communication (GO-COM) framework, namely Goal-Oriented Semantic Variational Autoencoder (GOS-VAE), by focusing on the extraction of the semantics vital to the downstream tasks. Specifically, we adopt a Vector Quantized Variational Autoencoder (VQ-VAE) to compress media data at the transmitter side. Instead of targeting the pixel-wise image data reconstruction, we measure the quality-of-service at the receiver end based on a pre-defined task-incentivized model. Moreover, to capture the relevant semantic features in the data reconstruction, imitation learning is adopted to measure the data regeneration quality in terms of goal-oriented semantics. Our experimental results demonstrate the power of imitation learning in characterizing goal-oriented semantics and bandwidth efficiency of our proposed GOS-VAE.

  PublisherDOI

• BOARD #129: AI as a Teaching Assistant: Aiding Engineering Students Beyond Office Hours

  2025-08-21

  article
  PublisherDOI

• Semantic Description Framework for Institutional Authority Files Based on Knowledge Organization Systems

  Communications in computer and information science · 2025-11-20

  book-chapter
  PublisherDOI

• Neural Motion Simulator: Pushing the Limit of World Models in Reinforcement Learning

  ArXiv.org · 2025-04-09

  preprintOpen accessSenior author

  An embodied system must not only model the patterns of the external world but also understand its own motion dynamics. A motion dynamic model is essential for efficient skill acquisition and effective planning. In this work, we introduce the neural motion simulator (MoSim), a world model that predicts the future physical state of an embodied system based on current observations and actions. MoSim achieves state-of-the-art performance in physical state prediction and provides competitive performance across a range of downstream tasks. This works shows that when a world model is accurate enough and performs precise long-horizon predictions, it can facilitate efficient skill acquisition in imagined worlds and even enable zero-shot reinforcement learning. Furthermore, MoSim can transform any model-free reinforcement learning (RL) algorithm into a model-based approach, effectively decoupling physical environment modeling from RL algorithm development. This separation allows for independent advancements in RL algorithms and world modeling, significantly improving sample efficiency and enhancing generalization capabilities. Our findings highlight that world models for motion dynamics is a promising direction for developing more versatile and capable embodied systems.

  PublisherOA PDFDOI

• Structural and functional analysis of soy protein-skipjack tuna protein gels prepared by two-screw extrusion and their effects on metabolites and gut microbiota in vitro

  Food Chemistry X · 2025-04-01 · 2 citations

  articleOpen access

  This study examined the impact of varying skipjack tuna fish protein (FP) concentrations on the structural and functional properties of soy protein (SP)-based gels produced via two-screw extrusion. Scanning electron microscopy showed that gels with ≤10 % FP had compact structures, while ≥20 % FP caused voids, cracks, and phase separation. Fluorescence spectroscopy indicated disrupted protein conformation with increasing FP. X-ray diffraction revealed enhanced semi-crystalline structures, while textural analysis showed reduced hardness and chewiness but consistent springiness with higher FP. Gels with FP improved gelation behavior and thermal stability, with SP@20 %FP demonstrating optimal thermal resistance. Rheological analysis showed higher FP improved gel strength but weakened structural integrity beyond 20 %. Short-chain fatty acids analysis indicated increased acetate and propionate levels with FP, enhancing fermentation potential. Microbial diversity decreased with increasing FP, favoring fermentative bacteria. Moderate FP addition optimized gel properties, but excessive FP weakened structure and altered microbial dynamics, emphasizing balanced FP usage. • Moderate fish protein improved gel structure, thermal stability, and gelation behavior. • High fish protein caused voids, phase separation, and reduced gel textural properties. • Fish protein addition increased acetate and propionate, enhancing fermentation potential. • Higher fish protein reduced microbial diversity, favoring fermentative bacterial growth. • Soy protein with 20 % fish protein achieved optimal structural and microbial stability.

  PublisherDOI

• Neural Motion Simulator Pushing the Limit of World Models in Reinforcement Learning

  2025-06-10

  articleSenior author

  An embodied system must not only model the patterns of the external world but also understand its own motion dynamics. A motion dynamic model is essential for efficient skill acquisition and effective planning. In this work, we introduce the neural motion simulator (MoSim), a world model that predicts the future physical state of an embodied system based on current observations and actions. MoSim achieves state-of-the-art performance in physical state prediction and provides competitive performance across a range of downstream tasks. This works shows that when a world model is accurate enough and performs precise long-horizon predictions, it can facilitate efficient skill acquisition in imagined worlds and even enable zero-shot reinforcement learning. Furthermore, MoSim can transform any model-free reinforcement learning (RL) algorithm into a model-based approach, effectively decoupling physical environment modeling from RL algorithm development. This separation allows for independent advancements in RL algorithms and world modeling, significantly improving sample efficiency and enhancing generalization capabilities. Our findings highlight that world models for motion dynamics is a promising direction for developing more versatile and capable embodied systems.

  PublisherDOI

• Insight: A Multi-Modal Diagnostic Pipeline using LLMs for Ocular Surface Disease Diagnosis

  arXiv (Cornell University) · 2024-10-01

  preprintOpen access

  Accurate diagnosis of ocular surface diseases is critical in optometry and ophthalmology, which hinge on integrating clinical data sources (e.g., meibography imaging and clinical metadata). Traditional human assessments lack precision in quantifying clinical observations, while current machine-based methods often treat diagnoses as multi-class classification problems, limiting the diagnoses to a predefined closed-set of curated answers without reasoning the clinical relevance of each variable to the diagnosis. To tackle these challenges, we introduce an innovative multi-modal diagnostic pipeline (MDPipe) by employing large language models (LLMs) for ocular surface disease diagnosis. We first employ a visual translator to interpret meibography images by converting them into quantifiable morphology data, facilitating their integration with clinical metadata and enabling the communication of nuanced medical insight to LLMs. To further advance this communication, we introduce a LLM-based summarizer to contextualize the insight from the combined morphology and clinical metadata, and generate clinical report summaries. Finally, we refine the LLMs' reasoning ability with domain-specific insight from real-life clinician diagnoses. Our evaluation across diverse ocular surface disease diagnosis benchmarks demonstrates that MDPipe outperforms existing standards, including GPT-4, and provides clinically sound rationales for diagnoses.

  PublisherOA PDFDOI

• Pose-Aware Self-Supervised Learning with Viewpoint Trajectory Regularization

  arXiv (Cornell University) · 2024-03-22

  preprintOpen access

  Learning visual features from unlabeled images has proven successful for semantic categorization, often by mapping different $views$ of the same object to the same feature to achieve recognition invariance. However, visual recognition involves not only identifying $what$ an object is but also understanding $how$ it is presented. For example, seeing a car from the side versus head-on is crucial for deciding whether to stay put or jump out of the way. While unsupervised feature learning for downstream viewpoint reasoning is important, it remains under-explored, partly due to the lack of a standardized evaluation method and benchmarks. We introduce a new dataset of adjacent image triplets obtained from a viewpoint trajectory, without any semantic or pose labels. We benchmark both semantic classification and pose estimation accuracies on the same visual feature. Additionally, we propose a viewpoint trajectory regularization loss for learning features from unlabeled image triplets. Our experiments demonstrate that this approach helps develop a visual representation that encodes object identity and organizes objects by their poses, retaining semantic classification accuracy while achieving emergent global pose awareness and better generalization to novel objects. Our dataset and code are available at http://pwang.pw/trajSSL/.

  PublisherOA PDFDOI

Frequent coauthors

• Stella X. Yu

  University of California, Berkeley

  24 shared

• Yann LeCun

  New York University

  22 shared

• Friedrich T. Sommer

  Center for Theoretical Biological Physics

  19 shared

• Bruno A. Olshausen

  17 shared

• Jiayun Wang

  16 shared

• Zengyi Li

  13 shared

• Qiqi Fan

  Shanghai Stomatological Hospital

  9 shared

• Zeyu Yun

  9 shared