Yi-Chun Liao

‣ Advisor: Prof. Sharon Hu ‣ Conduct independent research on realizing in-memory retrieval for RAG, leveraging a quantization scheme to meet hardware constraints and proposing an algorithm to introduce new capability for content-addressable memory (CAM). ‣ Conduct independent research on alleviating the scalability issue of in-memory hyperdimensional computing, redesigning model to optimize performance under hardware constraints and introducing an algorithm to enhance CAM scalability. ‣ Contribute to the extension of open-source CAM simulator CAMASim, focusing on data placement optimization with the approaches proposed in the first-authored paper RETENTION.

‣ Advisor: Prof. Tei-Wei Kuo and Prof. Yuan-Hao Chang ‣ Conduct independent research on minimizing resource consumption of in-memory inference for tree-based model, introducing an end-to-end framework RETENTION to enhance resource efficiency for CAM-based acceleration. ‣ Explore and quantify the relationship between capacity redundancy and processing overhead, enabling tradeoffs between controllable accuracy loss, energy consumption, and capacity requirement. ‣ Propose a pruning algorithm to minimize model complexity within controllable accuracy loss, and two data placement strategies to prioritize space and energy efficiency respectively. ‣ Improve space efficiency by 4.35×to 207.12×with less than 3% accuracy loss and no degradation on throughput and energy efficiency. The first-authored paper is currently under review by TCAD. ‣ Collaborate on minimizing retrieved tokens to accelerate RAG, exploring the tradeoffs between accuracy, chunk size, and capacity requirement.

‣ Advisor: Prof. Yennun Huang ‣ Collaborated on developing intrusion detection tool IPMES to detect incremental tactics, techniques, and procedures (TTPs) in system audit event streams. The co-authored paper was published in DSN 2024. ‣ Constructed an automated system independently to record hundreds of system attacks, utilizing MITRE Caldera to inject malware and TTPs into virtual machines. The system was subsequently used for synthesizing system attack datasets.

‣ Supported the instruction of the junior compulsory course Computer Architecture at NTU, enrolled by over 100 students. ‣ Led weekly discussion sessions, designed and/or graded assignments, quizzes, and exams. ‣ Provided voluntary office hours, exceeding 10 hours per week during exam periods.

‣ Directed the student council of the CSIE department with over 50 members. ‣ Promoted student welfare and represented over 600 students in faculty meetings, contributing to departmental decisions.

‣ Represented over 1500 students in college faculty meetings and helped shape student-related policies.

‣ Led a 50-member service club, managing budget allocation and arranging educational and recreational activities. ‣ Worked with Childhood Cancer Foundation R.O.C., organizing voluntary services to reduce the burden of parents.

‣ Provided TA hours for over 300 C++ beginners during two-week crash courses. ‣ Adapted teaching strategies including programming problem solving and customized course reviews to meet the diverse needs and skill levels of students.

‣ Provided TA hours for over 300 C++ beginners during two-week crash courses. ‣ Adapted teaching strategies including programming problem solving and customized course reviews to meet the diverse needs and skill levels of students.