About

Chen Sun is an assistant professor of computer science at Brown University and a part-time staff research scientist at Google DeepMind. His research focuses on computer vision, machine learning, and AI, with current interests in learning temporal dynamics from unlabeled videos and applying these video-centric world models in robotics. He directs the PALM research lab at Brown, which concentrates on learning generalizable temporal dynamics from unlabeled videos, including multimodal concepts, human behaviors, and raw pixels, as well as exploring applications of these models in robotics. Chen Sun has received several awards including an NSF CAREER Award, a Richard B. Salomon Faculty Research Award, a University Research Seed Award, and a Samsung Global Research Outreach Award. His work on behavior prediction in videos was a CVPR 2019 best paper finalist. His research has been supported by industry and research organizations such as Adobe, Honda, Meta, NASA, NSF, NVIDIA, and Samsung. He is a member of the NSF AI Research Institute on Interaction for AI Assistants (ARIA). In addition to his research, Chen Sun teaches courses on deep learning and advanced topics in deep learning, and has delivered short courses on multimodal transformers at ICASSP and AAAI. He has mentored numerous PhD students and undergraduates, many of whom have gone on to positions at leading research institutions and industry. His research projects include goal-conditioned video models, multimodal video understanding, physics-conditioned video generation, and long-term action anticipation, among others. Chen Sun actively participates in the academic community as a workshop chair, area chair for major conferences such as CVPR, ICCV, ECCV, NeurIPS, and ACL, and as an action editor for TMLR.

Research topics

• Immunology
• Medicine
• Internal medicine
• Biology
• Chemistry
• Cancer research
• Anatomy
• Cell biology
• Molecular biology
• Biochemistry
• Pathology

Selected publications

• Analysis of the Economic Impacts of the China-U.S. Trade War on China and US Economy

  Lecture Notes in Education Psychology and Public Media · 2025-03-12

  articleOpen access1st authorCorresponding

  The current trade war between China and the United States is the largest trade war on the global market in nearly half a century. The trade war refers to a series of economic and policy clashes between China and the United States in the field of trade since 2018. This trade war was mainly initiated by the United States to reduce the trade deficit between China and United States, protect the U.S. manufacturing industry, and limit the rise of Chinese technology, while involving issues such as intellectual property rights and technology transfer. As the world’s two largest economies, the trade conflict between China and the United States directly affects the trend and stability of the global economy. At the same time, the trade conflict between China and the United States has also had a profound impact on technological innovation in both countries. Through a literature review, this paper comprehensively analyzes the impact of the China-U.S. trade war on the bilateral economy and employment of China and the United States, and discusses the impact of the China-U.S. trade war on the development trend of the global economy. This paper argues that governments should repeatedly study the causes and impacts of the China-U.S. trade war, and realize that the trade war is not desirable, and win-win cooperation is the direction of future development.

  PublisherOA PDFDOI

• Word Misinterpretation and Its Impact on Philosophy: An Etymological and Cultural Perspective

  SSRN Electronic Journal · 2025-01-01

  preprintOpen access1st authorCorresponding
  PublisherDOI

• AS‐XAI: Self‐Supervised Automatic Semantic Interpretation for CNN

  Advanced Intelligent Systems · 2024-12-01 · 1 citations

  articleOpen access1st author

  Interpretable Machine Learning Interpretable machine learning is essential for building trustworthy AI systems. Automated Semantically Interpretable AI (AS-XAI) extracts the common semantic feature space of diverse data samples and combines this feature space with a sensitivity analysis of neural networks in each semantic space to understand the networks’ decision-making processes. AS-XAI leverages the model’s understanding of common semantics in existing data to enable a wide range of fine-grained and scalable real-world applications. This approach allows for comprehensive semantic conceptual interpretations of out-of-distribution hybrids as well as species that are difficult for humans to recognize. See article number 2400359 by Changqi Sun, Hao Xu, Yuntian Chen, and Dongxiao Zhang.

  PublisherDOI

• Alpha-2-Macroglobulin Attenuates Posttraumatic Osteoarthritis Cartilage Damage by Inhibiting Inflammatory Pathways With Modified Intra-articular Drilling in a Yucatan Minipig Model

  The American Journal of Sports Medicine · 2024-08-30 · 5 citations

  articleOpen access1st author

  Background: Posttraumatic osteoarthritis (PTOA) arises secondarily to joint trauma and is driven by catabolic inflammatory pathways. Alpha-2-macroglobulin (α 2 M) is a naturally occurring proteinase inhibitor found in human serum and synovial fluid that binds proteases as well as proinflammatory cytokines involved in the pathogenesis of PTOA. Purpose: (1) To investigate the therapeutic potential of intra-articular α 2 M injections during the acute stages of PTOA by inhibiting inflammatory pathways driven by the cytokines expressed by the synovium in a large preclinical Yucatan minipig model and (2) to determine if 3 intra-articular α 2 M injections have greater chondroprotective effects compared with 1 intra-articular injection. Study Design: Controlled laboratory study. Methods: A total of 48 Yucatan minipigs were randomized into 4 groups (n = 12 each): (1) modified intra-articular drilling (mIAD) and saline (mIAD + saline), (2) mIAD and 1 intra-articular α 2 M injection (mIAD +α 2 M-1), (3) mIAD and 3 α 2 M injections (mIAD +α 2 M-3), and (4) sham control. Surgical hindlimbs were harvested at 15 weeks after surgery. Cartilage degeneration, synovial changes, inflammatory gene expression, and matrix metalloproteinase levels were evaluated. Gait asymmetry was measured before and after surgery using a pressure-sensing walkway system. Results: Macroscopic lesion areas and microscopic cartilage degeneration scores were lower in the mIAD +α 2 M-1 and mIAD +α 2 M-3 groups compared with the mIAD + saline group ( P < .05) and similar to those in the sham group ( P > .05). Synovial membrane scores of the mIAD +α 2 M-1 and mIAD +α 2 M-3 groups were lower than that of the mIAD + saline group ( P < .05) and higher than that of the sham group ( P < .05). Interleukin-1 beta, nuclear factor kappa B, and tumor necrosis factor alpha mRNA expression in the synovium and matrix metalloproteinase-1 levels in synovial fluid were significantly lower in the mIAD +α 2 M-1 and mIAD +α 2 M-3 groups compared with the mIAD + saline group ( P < .05). No significant differences were observed between the mIAD +α 2 M-1 and mIAD +α 2 M-3 groups for all measured outcomes. There were early changes in gait ( P < .05) between preoperative and postoperative time points for the mIAD + saline, mIAD +α 2 M-1, and mIAD +α 2 M-3 groups that normalized by 15 weeks. Conclusion: Animals receiving early α 2 M treatment exhibited less cartilage damage, milder synovitis, and lower inflammation compared with animals with no α 2 M treatment. These results exemplify the early anti-inflammatory effects of α 2 M and provide evidence that intra-articular α 2 M injections may slow the progression of PTOA. Clinical Relevance: In patients presenting with an acute joint injury, an early intervention with α 2 M may have the potential to reduce cartilage degeneration from catabolic pathways and delay the development of PTOA.

  PublisherOA PDFDOI

• AS‐XAI: Self‐Supervised Automatic Semantic Interpretation for CNN

  Advanced Intelligent Systems · 2024-09-30 · 3 citations

  articleOpen access1st author

  Explainable artificial intelligence (XAI) aims to develop transparent explanatory approaches for “black‐box” deep learning models. However, it remains difficult for existing methods to achieve the trade‐off of the three key criteria in interpretability, namely, reliability, understandability, and usability, which hinder their practical applications. In this article, we propose a self‐supervised automatic semantic interpretable explainable artificial intelligence (AS‐XAI) framework, which utilizes transparent orthogonal embedding semantic extraction spaces and row‐centered principal component analysis (PCA) for global semantic interpretation of model decisions in the absence of human interference, without additional computational costs. In addition, the invariance of filter feature high‐rank decomposition is used to evaluate model sensitivity to different semantic concepts. Extensive experiments demonstrate that robust and orthogonal semantic spaces can be automatically extracted by AS‐XAI, providing more effective global interpretability for convolutional neural networks (CNNs) and generating human‐comprehensible explanations. The proposed approach offers broad fine‐grained extensible practical applications, including shared semantic interpretation under out‐of‐distribution (OOD) categories, auxiliary explanations for species that are challenging to distinguish, and classification explanations from various perspectives. In a systematic evaluation by users with varying levels of AI knowledge, AS‐XAI demonstrated superior “glass box” characteristics.

  PublisherDOI

• A knowledge-based data-driven (KBDD) framework for all-day identification of cloud types using satellite remote sensing

  arXiv (Cornell University) · 2023-12-01

  preprintOpen access

  Cloud types, as a type of meteorological data, are of particular significance for evaluating changes in rainfall, heatwaves, water resources, floods and droughts, food security and vegetation cover, as well as land use. In order to effectively utilize high-resolution geostationary observations, a knowledge-based data-driven (KBDD) framework for all-day identification of cloud types based on spectral information from Himawari-8/9 satellite sensors is designed. And a novel, simple and efficient network, named CldNet, is proposed. Compared with widely used semantic segmentation networks, including SegNet, PSPNet, DeepLabV3+, UNet, and ResUnet, our proposed model CldNet with an accuracy of 80.89+-2.18% is state-of-the-art in identifying cloud types and has increased by 32%, 46%, 22%, 2%, and 39%, respectively. With the assistance of auxiliary information (e.g., satellite zenith/azimuth angle, solar zenith/azimuth angle), the accuracy of CldNet-W using visible and near-infrared bands and CldNet-O not using visible and near-infrared bands on the test dataset is 82.23+-2.14% and 73.21+-2.02%, respectively. Meanwhile, the total parameters of CldNet are only 0.46M, making it easy for edge deployment. More importantly, the trained CldNet without any fine-tuning can predict cloud types with higher spatial resolution using satellite spectral data with spatial resolution 0.02°*0.02°, which indicates that CldNet possesses a strong generalization ability. In aggregate, the KBDD framework using CldNet is a highly effective cloud-type identification system capable of providing a high-fidelity, all-day, spatiotemporal cloud-type database for many climate assessment fields.

  PublisherOA PDFDOI

• AS-XAI: Self-supervised Automatic Semantic Interpretation for CNN

  arXiv (Cornell University) · 2023-12-02

  preprintOpen access1st authorCorresponding

  Explainable artificial intelligence (XAI) aims to develop transparent explanatory approaches for "black-box" deep learning models. However,it remains difficult for existing methods to achieve the trade-off of the three key criteria in interpretability, namely, reliability, causality, and usability, which hinder their practical applications. In this paper, we propose a self-supervised automatic semantic interpretable explainable artificial intelligence (AS-XAI) framework, which utilizes transparent orthogonal embedding semantic extraction spaces and row-centered principal component analysis (PCA) for global semantic interpretation of model decisions in the absence of human interference, without additional computational costs. In addition, the invariance of filter feature high-rank decomposition is used to evaluate model sensitivity to different semantic concepts. Extensive experiments demonstrate that robust and orthogonal semantic spaces can be automatically extracted by AS-XAI, providing more effective global interpretability for convolutional neural networks (CNNs) and generating human-comprehensible explanations. The proposed approach offers broad fine-grained extensible practical applications, including shared semantic interpretation under out-of-distribution (OOD) categories, auxiliary explanations for species that are challenging to distinguish, and classification explanations from various perspectives.

  PublisherOA PDFDOI

• A novel large animal model of posttraumatic osteoarthritis induced by inflammation with mechanical stability.

  PubMed · 2023-01-01 · 2 citations

  articleOpen access1st authorCorresponding

  OBJECTIVES: Animal models are needed to reliably separate the effects of mechanical joint instability and inflammation on posttraumatic osteoarthritis (PTOA) pathogenesis. We hypothesized that our modified intra-articular drilling (mIAD) procedure induces cartilage damage and synovial changes through increased inflammation without causing changes in gait. METHODS: Twenty-four Yucatan minipigs were randomized into the mIAD (n=12) or sham control group (n=12). mIAD animals had two osseous tunnels drilled into each of the tibia and femur adjacent to the anterior cruciate ligament (ACL) attachment sites on the left hind knee. Surgical and contralateral limbs were harvested 15 weeks post-surgery. Cartilage degeneration was evaluated macroscopically and histologically. Synovial changes were evaluated histologically. Interleukin-1 beta (IL-1β), nuclear factor kappa B (NF-κB), and tumor necrosis factor alpha (TNF-α) mRNA expression levels in the synovial membrane were measured using quantitative real-time polymerase chain reaction. IL-1β and NF-κB levels in chondrocytes were assessed using immunohistochemistry. Load asymmetry during gait was recorded by a pressure-sensing walkway system before and after surgery. RESULTS: The mIAD surgical knees demonstrated greater gross and histological cartilage damage than contralateral (P<.01) and sham knees (P<.05). Synovitis was present only in the mIAD surgical knee. Synovial inflammatory marker (IL-1β, NF-κB, and TNF-α) expression was three times higher in the mIAD surgical knee than the contralateral (P<.05). Chondrocyte IL-1β and NF-κB levels were highest in the mIAD surgical knee. In general, there were no significant changes in gait. CONCLUSIONS: The mIAD model induced PTOA through inflammation without affecting gait mechanics. This large animal model has significant applications for evaluating the role of inflammation in PTOA and for developing therapies aimed at reducing inflammation following joint injury.

  PublisherOA PDF

• The Effects of an Osteoarthritic Joint Environment on ACL Damage and Degeneration: A Yucatan Miniature Pig Model

  Biomolecules · 2023-09-20

  articleOpen access

  Posttraumatic osteoarthritis (PTOA) arises secondary to joint injuries and is characteristically driven by inflammatory mediators. PTOA is often studied in the setting of ACL tears. However, a wide range of other injuries also lead to PTOA pathogenesis. The purpose of this study was to characterize the morphological changes in the uninjured ACL in a PTOA inflammatory environment. We retrospectively reviewed 14 ACLs from 13 Yucatan minipigs, 7 of which had undergone our modified intra-articular drilling (mIAD) procedure, which induced PTOA through inflammatory mediators. Seven ACLs were harvested from mIAD minipigs (PTOA) and seven ACLs from control minipigs with no cartilage degeneration (non-PTOA). ACL degeneration was evaluated using histological scoring systems. IL-1β, NF-κB, and TNF-α mRNA expression in the synovium was measured using qRT-PCR. PTOA minipigs demonstrated significant ACL degeneration, marked by a disorganized extracellular matrix, increased vascularity, and changes in cellular shape, density, and alignment. Furthermore, IL-1β, NF-κB, and TNF-α expression was elevated in the synovium of PTOA minipigs. These findings demonstrate the potential for ACL degeneration in a PTOA environment and emphasize the need for anti-inflammatory disease-modifying therapies following joint injury.

  PublisherOA PDFDOI

• A2M inhibits inflammatory mediators of chondrocytes by blocking IL‐1β/NF‐κB pathway

  Journal of Orthopaedic Research® · 2022 · 35 citations

  1st authorCorresponding
  • Chemistry
  • Molecular biology
  • Cell biology

  A hallmark of osteoarthritis (OA) is cartilage degeneration, which has been previously correlated with dramatic increases in inflammatory enzymes. Specifically, interleukin-1β (IL-1β) and subsequent upregulation of nuclear factor kappa B (NF-κB) is implicated as an important player in the development of posttraumatic osteoarthritis (PTOA). Alpha 2-macroglobulin (A2M) can inhibit this inflammatory pathway, making it a promising therapy for PTOA. Herein, we demonstrate that A2M binds and neutralizes IL-1β, blocking downstream NF-κB-induced catabolism seen in in vitro. Human chondrocytes (cell line C28) were incubated with A2M protein and then treated with IL-1β. A2M was labeled with VivoTag™ 680 to localize the protein postincubation. The degree of binding between A2M and IL-1β was evaluated through immunoprecipitation (IP). Catabolic proteins, including IL-1β and NF-kB, were detected by Western blot. Pro-inflammatory and chondrocyte-related gene expression was examined by qRT-PCR. VivoTag™ 680-labeled A2M was observed in the cytoplasm of C28 human chondrocytes by fluorescence microscopy. IP experiments demonstrated that A2M could bind IL-1β. Additionally, western blot analysis revealed that A2M neutralized IL-1β and NF-κB in a dose-dependent manner. Moreover, A2M decreased levels of MMPs and TNF-α and increased the expression of cartilage protective genes Col2, Type2, Smad4, and aggrecan. Mostly importantly, A2M was shown to directly neutralize IL-1β to downregulate the pro-inflammatory responses mediated by the NF-kB pathway. These results demonstrate a mechanism by which A2M reduces inflammatory catabolic activity and protects cartilage after joint injury. Further in vivo studies are needed to fully understand the potential of A2M as a novel PTOA therapy.

  PublisherDOI

Frequent coauthors

• Wei Lei

  Weifang People's Hospital

  27 shared

• Brett D. Owens

  Brown University

  26 shared

• Anthony M. Reginato

  Providence College

  22 shared

• Kenny Chang

  Brown University

  22 shared

• Lei Wei

  Manchester University

  21 shared

• Braden C. Fleming

  Rhode Island Hospital

  17 shared

• Olin D. Liang

  Brown University

  14 shared

• Guoxuan Peng

  Brown University

  13 shared

Labs

• Socially Responsible Computing @ Brown CSPI