Download CV

Neuron Scientist engaged in visual system end general medical Ai, imagging analysis design, accessibility + usability, and digital twins. Keen to support emerging and non-traditional publishing models and technologies that advance scholarly publishing's transformation.

Degrees

MSc, Design Informatics (Artificial Intelligence); University of Edinburgh, Edinburgh, UK, 2023

BEng, Data Science & Big Data; University of Liverpool, Liverpool, UK, 2019

Data Engineer, UCAP, China, 06/2022-09/2022

• Conducted statistical analysis using Excel on the operational performance of Traditional Chinese Medicine (TCM) therapy institutions in Chengdu, identifying market opportunities.
• Utilized the PANDAS library for data preprocessing to ensure the integrity and accuracy of Traditional Chinese Medicine (TCM) metadata.
• Developed anomaly detection algorithms using machine learning techniques to automate data quality checks, significantly improving operational efficiency.

Work

Algorithm Engineer, Liverpool University Library System (Suzhou), University of Liverpool, 2021-08/2021

• Project manager for University of Liverpool and activities; provided the strategic planning, day-to-day management, staff/department education, marketing and promotion for WSU’s institutional repository.
• Collaborated with WSU Press to plan, implement, and manage online subscription based journals program; managed and developed back-end work flow to migrate 6 peer-reviewed journals to Digital Commons; created workshops and training for editorial staff; developed digital publishing internship.
• Created and led the WSU Scholarly Communications Team, managing projects, outreach, marketing, and other scholarly communications initiatives; coordinated and led first WSU Open Access Week.
• Researched, planned, implemented, and manage faculty self-archiving program; developed workflows and services to handle permissions and rights clearance, document reformatting, and publication deposits to Digital Commons for faculty and staff; hired, trained, and supervised two undergraduate and two graduate interns; developed workshops on self-archiving and author’s rights for librarians, faculty, and graduate students.

Robotics Algorithm Engineer, iFLYTEK Co., Ltd., 2021-06/2021

• Simulated robot movements using ROS (Robot Operating System) on Ubuntu platform.
• Optimized the shortest path module of the food delivery robot navigation system using Dijkstra’s algorithm, resulting in a 10% improvement in delivery efficiency.
• Developed an accurate speech recognition system using the K-means clustering algorithm to enhance the accuracy of dish name identification during delivery, improving overall service experience.
• Project manager for Wayne State University Institutional Repository and activities; provided the strategic planning, day-to-day management, staff/department education, marketing and promotion for institutional repository.

Freelance Deep Learning Developer, 2020-present

Providing small-scale deep learning prohect design, development, and consulting services for publishers, academic digital humanities projects, and other digital publishing ventures.

Research

Deep Learning Model for Neuron Response, ANC EDINBURGH, 2024-10/2024

• Proposed a new ViT-based architecture to predict the mouse V1 neural responses. The architecture integrates visual and behavioral input across animals, surpassing the previous prediction performance of CNN models.
• Investigated how the “attention weights” learned by the self-attention blocks change with behaviors to gain insight into how the model predicts neural activity.
• Developed a deep neural network trained on large amounts of neuronal responses to ecological videos from multiple visual cortical areas and mice. The model accurately predicted neuronal responses to natural videos and various new stimulus domains, such as coherent moving dots and noise patterns, demonstrating its strong generalization abilities.

Plug-and-Play Self-Distillation for Multimodal Sentiment Analysis with Incomplete Modalities, Monash Medical Ai Group, 2024-10/2024

• Estimating a Mixture of Student’s t-distributions for each modality combination to capture heavy-tailed data characteristics and handle anomalies.
• Using a confidence score based on degrees of freedom, optimizing the fusion of multiple St distributions into a unified distribution.
• Incorporating a Confidence-Aware Knowledge Distillation process to express prediction confidence by minimizing the discrepancy between uncertainty scores of the teacher and student networks, enhancing robustness with incomplete data.

Predicting IPO initial returns using machine learning algorithm, UNSW Fintech, 2023-07/2024

• Developed a sophisticated machine learning method using a hybrid random forest model to predict the initial returns of IPOs issued on the Hong Kong Stock Exchange (HKEX).
• Incorporated Bayesian optimization (BO) and Cross-Validation for feature selection into the hybrid random forest model, enabling it to effectively capture and utilize important features for more accurate predictions.
• Evaluated the performance of the proposed method against mainstream models such as Linear Regression, Boosting, and MLP. The experimental results significantly demonstrated the effectiveness and superiority of our method, underscoring its potential for practical applications.

Smoking Detection based on Attention Mechanism, MMAI LIVERPOOL, 2022-11/2023

• Implemented data augmentation techniques to evaluate the performance of three SOTA deep learning models (YOLOv5s, SSD, Faster-RCNN) for real-time object recognition and classification in smoking images.
• Constructed a Transformer block (C3TR) integrated into the backbone of YOLOv5s. This significantly reduced the com- plexity of the model and improved learning convergence, enhancing detection accuracy.
• Developed an innovative algorithm named “YOLOv5s-Transformer” that prioritizes both computational efficiency and mini- mal resource utilization, making it ideal for resource-constrained environments.
• Benchmarked the proposed YOLOv5s-Transformer model against Faster RCNN, SSD, and the baseline YOLOv5s model. The new model demonstrated superior performance by achieving a 37.8 percent higher average precision (AP) while maintaining a processing speed of 1235 frames per second (FPS).

Skills

Technology Applications and Expertise

Specialized Skills

Visual Reasoning Design, User Testing, Accessibility,Data Design, Data Analytics, Digital Health Workflows, Digital Heritage

Markup/Programming

Pytorch, Python, C++, JavaScript, Ruby/Rails, PHP, Perl, MySQL

Applications

Adobe Creative Suite, Oxygen, Balsamiq, Atom, Microsoft Office, Google Drive

OS/Environments

Mac, Linux, Unix, Ubuntu, Windows

Presentations and Publications

Edinburgh, UK. Invited Workshop Speaker. Creative Informatics Annual Meeting 2023, Production and Design Workshop: Accessibility is Accessible, "Getting to Accessible Publishing at the University of Michigan Press," Sep 2023. Slides available at https://goo.gl/ZaEkgC.

Melbourne, AUS. Conference Panelist. Library Publishing Forum 2016, University of North Texas, "Mellon Project Updates: Fulcrum: Building a Hosted Platform for Managing Monographic Source Materials and Born Digital Publications," May 2016.

Nanjing, China. Conference Presenter. University of Michigan StaffWorks Best Practices and Technology Conference, "Dear Editor: Cooperation for Accessible Publishing at the University of Michigan Press," May 2016.

Ann Arbor, MI. Invited Co-Presenter. Rackham Graduate School, "Building a Professional Presence Online," March 2015. Presented with Meredith Kahn.