Curriculum Vitae

Publications

A Parallel Gannet Optimization Algorithm with Communication Strategies (PGOA)

Su, Jing-Bo, et al. "A Parallel Gannet Optimization Algorithm with Communication Strategies (PGOA)." International Conference on Intelligent Information Hiding and Multimedia Signal Processing. Singapore: Springer Nature Singapore, 2022.

Enhancing Adversarial Attacks through Chain of Thought

Research Experience

UC, Riverside Parallel Algorithm Lab with Prof. Yan Gu and Prof. Yihan Sun

Project: ANNlib: Quickly Build-up Your Approximate Nearest Neighbor System

Reproduced existing graph-based filtered ANNS algorithms on ANNlib system by using interfaces provided by ANNlib to demonstrate the ability of ANNlib system to support various graph-based ANNS algorithms
Refactored experimental code, encapsulated each ANNS algorithm into modules for easier evaluating. Developers can implement their own graph-based ANNS algorithms with a separate codebase, which reflects the engineering and user-friendliness of ANNlib system
Designed ablation experiments and evaluated multiple ANNS algorithms on various vector dataset benchmarks. Recorded and analyzed the index construction delay, search accuracy (recall) and throughput (QPS) as the performance of graph-based ANNS algorithms
Experimental results showed that these algorithms can perform as efficiently or better on ANNlib system than their well-optimized counterparts

NCUT Evolutionary Computing Lab with Prof. Ruobin Wang and Dr. Lin Xu

Project: A Parallel Gannet Optimization Algorithm with Communication Strategies (PGOA)

Presented final research findings at the 18th International Conference on Intelligent Information Hiding and Multimedia Signal Processing, substantially enhancing my communication skills and academic English proficiency
Carried out experiments with CEC-2013 to benchmark against the baseline Gannet Optimization Algorithm
Evaluated on both compact and parallel techniques. Experiential results demonstrated that the parallel strategy significantly surpassed the original GOA algorithm in computational efficiency and precision
Contributed to weekly lab meetings by conducting literature reviews on 10+ papers about Evolutionary Computing and presenting emerging concepts at graduate-level group discussions

Work Experience

China National Investment & Guaranty Corporation

Full-Stack Software Developer Intern

Applied full-stack development techniques by leveraging various frameworks (Django, React, SpringBoot) and refining programming skills in multiple languages (Python, TypeScript, Java)
Developed a project management system, using Django for backend and React for frontend, that significantly optimized project management processes and enhanced product managers’ efficiency
Independently built a script manager tool that empowered 20+ engineers to manage and execute Python scripts, contributing to streamlined script management and enhanced productivity across the team

Selected Projects

UC, Riverside CS222 - Natural Language Processing

Project: Enhancing Adversarial Attacks through Chain of Thought

Explored a novel attack method to raise success rates of adversarial attacks on large language models (LLMs)
Integrated chain-of-thought (CoT) prompting into a gradient-based adversarial attack on aligned LLMs and conducted experiments comparing performance on both baseline and Amazon’s auto-cot
Evaluated the ablation studies on various harmful types categorized by Llama-Guard, demonstrating that the novel method successfully triggered the CoT ability on multiple aligned LLMs during attacks and outperformed prior approaches.

UC, Riverside CS178 - Project Sequence

Project: MedShare: Healthcare Made Easy

Built an easily-use medical management system with key features like messaging, forums, and appointment scheduling, integrating an API-callable chatbot to effectively enhance communication between doctors and patients and reduce healthcare delays.
Designed and implemented the frontend and backend of the system using ReactJS, NodeJS, and MongoDB. Employed GitHub to do version control for the project. Ensured efficient data management and a user-friendly interface.
Designing a chatbot system based on the LangChain framework and deploying it in a distributed manner with the Node system. Utilizing the Retrieval Augmented Generation (RAG) technique enables doctors to upload patient-related documents to flesh out the system knowledge base.

Jingbo Su