Google Scholar /
LinkedIn /
Github /
CV
|
Chenshu Liu (Chen)
Currently, I am a Researcher at Terasaki Institute, working on wearable biomedical devices and interfaces to integrate AI components in medicine to foster a smart-healthcare ecosystem, supervised by Professor Yangzhi Zhu, as well as on AI in immunology, supervised by Professor Chongming Jiang.
I hold a joint appointment at Laboratory for Smart and Additive Manufacturing at CSUN, working on AI applications in additive manufacturing (AM) and deploying knowledge-graph driven methods for efficient human machine interaction in AM process, supervised by Professor Bingbing Li.
I also serve as a technical consultant for the Smart Textile group led by Professor Wei Cao at ARCS on evaluating robustness of wearable smart textiles from the lens of consumer aspects.
I did my Masters in Bioengineering and a dual bachelor’s in Neuroscience and Statistics at University of California, Los Angeles.
|
I am a neuroscientist 🧠 and statistician 📊 in mind, an engineer ⚙️ at hand, and an artist/photographer 📷 at heart. I am a curious innovator exploring AI integration in Health Science & Humanities, merging computer science, medical engineering, heritage science, neuroscience, etc.
|
Research Interests
My research interests focus on creating a synergy between human and AI, applying interdisciplinary technology to enhance well-being both physically and spiritually.
I come up with engineering innovations to increase the accessibility and intuitiveness of human-AI interaction in both healthcare applications and societal enrichments.
I propose to incorporate knowlege-enhanced multidomain AI as decision support agents. I am interested in constructing reliable knowledge bases (KB) and knowledge graphs (KG) for decision supports with high interpretability and explainability in complex decision processes.
I am interested in leveraging advanced sensing technologies to create a comprehensive profile of the subject of interest. By integrating these descriptive inputs and combining them with domain-specific knowledge, I aim to enable more informed and context-aware reasoning, whether for enhance patient well-being or restoring an artifact.
Through the unification of sensor data and expert knowledge, I seek to enhance the overall decision-making process.
I envision working with people from different fields of expertise that can potentially benefitted from AI-aided decision supports and integrate with reliable and efficient pipelines in their workflow.
I’m also interested in rethinking the role of AI in emotion support via affective computing and multi-sensor fusion,
to enable creativity and self-reflection through the most natural form of personal expression and the most realistic day-to-day behaviors, which gave rise to the on-going Posture2Melody project.
In the project, I propose to use generative AI, alongside with different modalities of physiology and psychology tracking, as the mirror onto our own understanding of music, lowering the bar for expression.
This framework invites us to consider how AI can perceive and make-sense of human emotion overtime, enabling AI agents to deliver responses appropriately aligned with the user's emotional state. .
|
Research Publications / Projects / Talks
I have cultivated a broad range domain experiences, spanning both natural sciences and humanities.
I am proficient in developing AI algorithms diverse application scenarios. The wordcloud below shows the span of my projects, fields of expertise, and techniques.
Projects are arranged in chronological order below, including published works and on-going ones.
|
On-going Projects
Human Motion Tracking
Generative AI
Transformer
GAN
Emotion Therapy
|
Posture2Melody: Transforming Movement into Musical Melody with AI Harmony
Chenshu Liu, Yiran Wang, Haolin Fan.
TLDR: Posture2Melody uses GAN-Transformer-based architecture to generate melodies from human postures. Inspired by the idea that the expansiveness of human posture reflects emotional states, this project seeks to create a seamless interaction between bodily movement and music.
Whether it’s through dance or everyday postures, Posture2Melody transforms these movements into musical melodies, potentially acting as an emotional therapy tool. By synchronizing bodily movement and music, Posture2Melody seeks to develop a creative technique that could be used in emotional therapy.
|
Electromyography
Transcutaneous Electrical Stimulation
Gesture Recognition
Neurodegenerative Disease
Theranostics System
|
NeuroMT: Neurodegenerative Monitoring and Therapy Device Using EMG and TENS
Chenshu Liu, Yiran Wang, Haolin Fan, Ziyuan Che, Yangzhi Zhu, Bingbing Li.
TLDR: The NeuroMT project introduces an innovative theranostic device combining electromyography (EMG) and Transcutaneous Electrical Nerve Stimulation (TENS) unit to monitor and treat abnormal neuromuscular activity in real-time.
By leveraging single-channel EMG, the system detects irregularities in muscle activation associated with neurodegenerative disorders. Subsequently, it employs TENS unit to modulate neuromuscular pathways and restore stable motor function.
This closed-loop system integrates both diagnostics and therapy, enabling real-time feedback and personalized treatment.
Our approach offers a novel solution for neurodegenerative patients, potentially improving mobility, reducing symptoms, and enhancing overall quality of life.
|
Electromyography
Transcutaneous Electrical Stimulation
Gesture Recognition
Neurodegenerative Disease
Theranostics System
|
MetalMind: A Knowledge Graph-Based Retrieval Framework for Enhanced Human-Machine Interaction in Metal Additive Manufacturing
Haolin Fan, Xinyu Liu, Zhen Fan, Chenshu Liu, Jerry Ying Hsi Fuh, Wen Feng Lu, Bingbing Li. NPJ Advanced Manufacturing, 2025
TLDR: This study presents a multi-modal Knowledge Graph (KG) combined with Retrieval-Augmented Generation (RAG) to enhance HMI in Digital Twin (DT) systems, facilitating structured, semantically enriched access to metal AM knowledge.
Our system supports training and decision-making through an automated KG construction pipeline encompassing preprocessing, data extraction using Large Language Models (LLMs), and post-processing with collaborative verification to ensure quality.
We implemented three retrieval modes: vector-based for detailed queries, graph-based for contextual insights, and a hybrid method for balanced information retrieval.
Additionally, a KG-based image retrieval feature connects entity descriptions to relevant visual data.
Experimental results indicate that the KG-driven hybrid retrieval mode enhances both global and granular understanding.
This publicly accessible system establishes a foundation for advanced HMI in smart manufacturing.
|
2024
Wearable
Computer Vision
RNN
Smart Contact Lens
Color Calibration
Diagnostic Platform
|
OPTMISE: Ocular Platform with Telemetric Mechano-Electro-Chromic Intelligent Sensing Ecosystem
Chenshu Liu, Hyo-Jeong Choi, Chengguang Zhang, Pengrui Dang, Wangjie Chen, Yongju Lee, Bingbing Li, Meyer Dawn, Pete Kollbaum, Hyeok Kim, Ali Khademhosseini, Yangzhi Zhu. Nature Biomedical Engineering (under review), 2024
TLDR: The OPTMISE lens offers a self-powered minimal-invasive wearable measuring alternative for measuring eyelid pressure based on Triboelectric Nanogenerator (TENG).
In constrast to traditional eyelid measurement methods that involve setting non-conformable foreign measuring apparatus between the eyeball and the eyelid, OPTMISE lens is a stand alone apparatus, making the device self-contained and less invasive.
User can directly wear the OPTMISE lens in the same way as a regular contact lens. The colorimetric mechano-chomic display (composed of PEDOT:PSS material) provides visual qualification of different eyelid pressures.
A customized software based on computer vision and time series processing AI algorithm creates an interface to allow users to dynamically quantify the eye pressure captured by the mechano-chomic display.
|
Electronic Skin
Multidomain AI
Decentralized Healthcare
Electronic Health Record
Decision Support System
|
Democratizing Healthcare: The Synergy of Electronic Skin and Multidomain AI
Chenshu Liu, Haolin Fan, Tong Zhou, Pinyi Yang, Lingdi Zhao, Yiran Wang, Yangzhi Zhu, Bingbing Li. ACS Chemical Reviews (submitted), 2024
TLDR: Electronic skin (E-skin) is revolutionizing the monitoring of physiological states.
With artificial intelligence (AI), sensory data from E-skin can now be decoded to provide valuable health insights.
However, current E-skin and AI integrations are largely unimodal, which limits the ability comprehensively represent subject's overall physiological state.
Furthermore, most AI algorithms did not take advantage of the large body of empirical data and domain knowledge to offer deeper clinical decision support.
Thus, this review proposes a framework to accelerate the shift toward decentralized healthcare by fostering synergy between multidomain AI and multimodal E-skin.
We highlight various E-skin modalities and using multi-sensor fusion AI technologies to create comprehensive patient profiles.
Additionally, we discuss how multimodal AI can improve medical reasoning and patient outcomes by integrating EHR data and domain-specific knowledge.
We explore the future potential and challenges of the multidomain AI and E-skin partnership, emphasizing the personalized healthcare benefits this collaboration can deliver.
|
Smart Manufacturing
Vision Language Model
Decision Support System
|
MaViLa: Unlocking New Potentials in Smart Manufacturing through Vision Language Models
Haolin Fan, Chenshu Liu, Neville Elieh Janvisloo ,Shijie Bian ,Jerry Ying Hsi Fuh ,Wen Feng Lu ,Bingbing Li. Journal of Manufacturing Systems (under review), 2024
TLDR: This paper presents MaViLa, a novel Vision Language Model (VLM) designed to unlock new potentials in smart manufacturing.
Through rigorous methodology and extensive experiments, MaViLa demonstrates superior performance across various benchmarks compared to other general-purpose VLMs.
This enhanced grasping of domain knowledge is attributed to the innovative use of an external vector store during the dataset construction process.
Practical experiments, including lab tests and the application of the CAXTON dataset, reveal that MaViLa excels in manufacturing tasks.
|
Smart Manufacturing
Multimodal AI
Knowledge Graph
Decision Support System
|
New Era Towards Autonomous Additive Manufacturing: A Review of Recent Trends and Future Perspectives
Haolin Fan, Chenshu Liu, Shijie Bian, Changyu Ma, Xuan Liu, Marshall Doyle, Thomas Lu, Lianyi Chen, Jerry Ying Hsi Fuh, Wen Feng Lu, Bingbing Li. International Journal of Extreme Manufacturing (accepted for publication), 2024
TLDR: The Additive Manufacturing (AM) landscape has significantly transformed in alignment with Industry 4.0 principles, primarily driven by the integration of Artificial Intelligence (AI) and Digital Twin (DT).
This review paper examines current solutions in Intelligent Additive Manufacturing, emphasizing control, monitoring and process autonomy, and end-to-end process integration.
This paper addresses the lifelong learning and self-optimization capabilities of AI agents. As manufacturing evolves, this paper posits that the future of AM will be characterized by a symbiotic relationship between advanced autonomy and human expertise, fostering a more adaptive and autonomous future manufacturing ecosystem.
|
Computer-aided Diagnostics
Cultural Heritage Conservation
Computer Vision
Knowledge Base
|
Web-based diagnostic platform for microorganism-induced deterioration on paper-based cultural relics with iterative training from human feedback
Chenshu Liu*, Songbin Ben, Chongwen Liu, Xianchao Li, Qingxia Meng, Yilin Hao, Qian Jiao, Pinyi Yang. Heritage Science, 2024
TLDR: Paper-based artifacts hold significant cultural and social values.
However, paper is intrinsically fragile to microorganisms, such as mold, due to its cellulose composition. Mold not only can damage papers’ structural integrity and pose significant challenges to conservation works
but also may subject individuals attending the contaminated artifacts to health risks. Current conservation practices with mold-contaminated artifacts have
little to no pre-screening, and the cleaning techniques are usually broad-spectrum rather than strain-specific.
This study investigated the feasibility of using a convolutional neural network (CNN) for fast in-situ recognition and classification of mold species on paper.
A webtool deployed via Streamlit was developed to provide public access to the mold classification tool, alongside with decision support system that provide detailed description about the strain of mold detected.
|
Classification Model
Audio Processing
Machine Learning
|
Recognition of bird species with birdsong records using machine learning methods
Yi Tang, Chenshu Liu, Xiang Yuan. Plos one, 2024
TLDR: The recognition of bird species through the analysis of their vocalizations is a crucial aspect of wildlife conservation and biodiversity monitoring. In this study, the acoustic features of Certhia americana, Certhia brachydactyla, and Certhia familiaris were calculated to train three machine learning models, Random Forest (RF), Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost). The XGBoost model had the best performance among the three models, with the highest accuracy and the highest AUC. The study provides a new approach to bird species recognition that utilizes sound data and acoustic characteristics.
|
Natural Langauge Processing
Semantics Classification
Online Pedagogy
COVID-19 Pandemic
|
A practical evaluation of online self-assisted previewing architecture on rain classroom for biochemistry lab courses
Chenshu Liu*, Songbin Ben*, Pinyi Yang, Jiayi Gong, Yin He. Frontiers in Education, 2024
TLDR: This study investigated the design of the optimal structure of online self-assisting coursework for laboratory courses that can assist students to better prepare for hands-on experiments.
Survey was conducted among undergraduate students who took Biochemistry during and post-pandemic. Textmining and semantics classification were performed on students' responses to analyze their emotions towards established online pedagogy frameworks and gain insights in their suggestion for a more effective online learning platform design.
We offer a few strategic suggestions that may guide the design of future online resources for laboratory classes such as involving multi-modality media to improve engagement and perfecting the interactive feature to increase its usage by students.
|
Computer-aided Diagnostics
Cultural Heritage Conservation
Paper Artifact Conservation
Computer Vision
|
Ai-Assisted Classification of Microorganism Strains on Paper-Based Cultural Relics (Workshop Presentation)
Chenshu Liu*, Chongwen Liu, Allison Wall. Lucas Museum Paper Conservation Workshop
TLDR: Invited to showcase integration of computer vision algorithms in the practice of conservation at the paper conservation workshop held at the Lucas Museum of Narrative Art hosted by Erin Jue.
Our method focused on using miscroscopic images of mold stains on paper-based cultural relics using computer vision algorithm.
|
2023
Cultural Heritage Conservation
Mold Removal
Surface Cleaning Agent
Surfactant
Bioenzyme
|
A biological cleaning agent for removing mold stains from paper artifacts
Qingxia Meng, Xianchao Li, Junqiang Geng, Chenshu Liu*, Songbin Ben*. Heritage Science, 2023
TLDR: Efficient removal of mold stains becomes an important research topic for paper conservation.
In this study, a cleaning scheme based on the combination of bioenzymes and biosurfactants was explored.
A cleaning agent composed of Sophorolipid and Betaine offer superior deacidification, anti-acidification, anti-aging, and reinforcement capabilities, which can provide extra support to the fibrous structure in addition to cleaning the paper materials.
The microbial contamination cleaning agent proposed in this study shows promising application prospects in conserving mold-contaminated paper artifacts.
|
Cultural Heritage Conservation
Mold Identification
Computer Vision
Convolutional Neural Network
|
Ai-Assisted Classification of Microorganism Strains on Paper-Based Cultural Relics (Conference Presentation)
Chenshu Liu, Chongwen Liu, Allison Wall. Art Bio Matters (ABM) Conference, 2023
TLDR: Our project that focused on using miscroscopic images of mold stains on paper-based cultural relics using computer vision algorithm was selected to present at the Art Bio Matters conference.
The project gained recognition in the cultural heritage conservation community and hold promises in assisting diagnostic procedure in paper-based cultural relic biodeterioration conservation.
|
|
Talks and Presentations
I have been invited and attending conferences, here is the list of the events I have attended and corresponding recorded/pre-recorded videos:
|
|
Teaching
I am devoted to education of all education levels. I have been teaching for over 9 years since the start of senior high. I have taught a broad range of courses, covering subjects: biology (AP, college level), chemistry (AP), calculus (AP, college level), statistics (AP, college level), programming in R, programming in Python, etc. I was the TA for LS23L: Laboratory and Scientific Methods at UCLA in the 2022-2023 academic year.
I am also devoted to enhance public understanding of machine learning and artificial intelligence. I have a personal education channel over RED to share bite-size knowledge of ML algorithms and AI-related techniques. I also curate a suite of ML-education repositories on Github:
|
This website was built thanks to the help of this source code.
|
|
