CV

Leyan Pan

Phone: +1 (470) 439-8014 Email: leyanpan@gatech.edu Address: 6217 Providence Club Drive, Atlanta GA GitHub: github.com/leyanpan

ACADEMIC INTEREST

Machine Learning and Deep Learning
Large Language Models
Theoretical Computer Science
Computer Security

EDUCATION

Ph.D. Candidate in Computer Science, School of Cybersecurity and Privacy Georgia Institute of Technology | Atlanta, GA August 2023 - Now Advisor: Prof. Wenke Lee
B.S/M.S in Computer Science (Concentrations: Artificial Intelligence & Theoretical Computer Science) Georgia Institute of Technology | Atlanta, GA Aug 2018 - May 2023 GPA: 4.0/4.0 Master’s Thesis advisor: Prof. Santosh Vempala Relevant Coursework: Machine Learning, Deep Learning, Robotics & Perception, Computer Vision, Natural Language Processing, Data Structure & Algorithms, Automata & Complexity, Design & Analysis of Algorithms, Advanced Algorithms & Randomness, Computational Social Science

RESEARCH

Towards Understanding Neural Collapse: The Effects of Batch Normalization and Weight Decay Jun 2022 - May 2023 Supervised by Prof. Santosh Vempala at Algorithm and Randomness Center, Georgia Tech
- Investigated a deep learning mathematical structure called Neural Collapse.
- Introduced geometrically intuitive intra-class and inter-class cosine similarity measure for the phenomenon.
- Established theoretical guarantees for the emergence of NC under the influence of last-layer BN and weight decay.
- Performed experiments revealing a pronounced occurrence of NC in models incorporating BN and appropriate weight-decay values. Preprint
Malware Analysis Using Graph Neural Networks January 2021 - Present Supervised by Prof. Wenke Lee and Dr. Yisroel Mirsky at Institute for Information Security & Privacy, Georgia Tech
- Proposed to apply function structural information in malicious behavioral analysis.
- Conducted experiments to compare the effectiveness of different graph autoencoder architectures, including GCN, GAT, and structure2vec, in identifying malicious functionalities using PyTorch.
- Improved malicious functionality detection AUC from 0.84 to 0.89.
- Currently analyzing clustering performance of extracted features for malicious functionality classification.
Lattice-Based Cryptography in Homomorphic Encryption Fall 2019 - Spring 2021 Supervised by Assoc. Prof. Vincent Mooney at Hardware/Software Codesign for Security Group, Georgia Tech
- Applied homomorphic encryption to analyze camera image data in the encrypted domain to detect motion.
- Designed homomorphic arithmetic operations using TFHE library in C programming language.
- Novel attempt to design simple application-specific homomorphic encryption schemes for computationally limited hardware. Code & Paper

INTERNSHIPS & TEACHING EXPERIENCE

Graduate Teaching Assistant for CS 4510 - Georgia Institute of Technology January 2022 - Now
- Assisted Professor and Lecturer Prof. Zvi Galil in instructing the course Automata and Complexity.
- Held office hours once per week for student inquiries.
- Graded homework, exams, and projects and provided lecture notes for different materials in the theory of computation.
Facebook Reality Labs - Silicon Security Engineering Intern June 2021 - August 2021
- Assisted several teams to extract and organize security configurations and detect vulnerabilities.
- Wrote C-based unauthorized access attacks to test firewall effectiveness.
- Developed a firewall configuration analysis system using Python.
- Improved configuration extraction efficiency from 30 min to 2-5 seconds.
IBM X-force Command Center Cybersecurity Intern June 2020 - August 2020
- Built a demonstrative industrial heat exchange system controlled by PLCs.
- Demonstrated potential cyber-attacks on the system.
- Analyzed Wireshark Packets and built Python scripts to attack the system using the Ethernet/IP Protocol.
Intel Mobile Communications China - Technical Intern May 2019 - August 2019
- Ran LTE testing on modems for iPhone.
- Fixed connection and hardware problems and reported software bugs to developers.
- Developed Python scripts to improve result collection efficiency.

PROJECTS

Theoretical Analysis for Buy-Many Mechanisms for Multiple Buyers Spring 2022-Current
- Surveyed recent work on theoretical bounds of buy-many mechanisms.
- Extended results for multiple buyers from unit-demand to additive buyers.
- Currently extending theoretical results to sub-additive and XOS buyers.
Analyzing the Changes in Mental Health of Twitter Users Before and After the COVID-19 Pandemic Spring 2022
- Utilized a word weight model to analyze language expressions on Twitter.
- Examined the impact of coronavirus on different genders and regions.
- Compared levels of stress and loneliness between different tweet datasets.
Comparison of Transposed Convolutional Neural Network Architectures on Art Creation using GAN Fall 2021
- Compared transposed versions of various convolutional neural network architectures as GAN decoders.
- Developed evaluation criteria and discovered limitations of certain network designs.
Joke Identification Fall 2019
- Accomplished classification models to identify jokes using sound analysis and text.
- Trained and compared performance of various classification algorithms. Code

SKILLS

Languages: C/C++, Python
Library & Frameworks: Pytorch
Tools & System: LaTeX, Git, vim, IDA Pro

Publications

Can Transformers Reason Logically? A Study in SAT Solving

Leyan Pan, Vijay Ganesh, Jacob Abernethy, Chris Esposo, and Wenke Lee. Can transformers reason logically? A study in SAT solving. Proceedings of the 42nd International Conference on Machine Learning, 2025.