CV
Education
- BEng in Computer Science and Technology, Big Data and Intelligence Track, Xidian University, 2022-2026
- GPA: 4.0/4.0
- Rank: 2/1500+ (first year), 1/476 (second year), 1/106 (third year)
Academic Research
- Chemical Reasoning with Reasoning-Oriented Large Language Models (Jun. 2025 - May 2026; Lead Researcher; initial research supervised by Prof. Shengchao Liu; undergraduate thesis advised by Prof. Yunan Li)
- Engineered training pipelines for LLMs of various scales across diverse HPC environments, including Compute Canada (SLURM), Shanghai AI Lab (MegCompute), and AutoDL (On-Demand Instance).
- Achieved about 80% of the predictive performance of trillion-parameter LLMs with less than 0.5% of their model parameters and less than 5% of their token consumption.
- Explored explicit and latent chain-of-thought paradigms for chemical reasoning by reviewing LLM-based chemical reasoning, proposing latent-space mechanism dynamics, and completing the thesis with an Excellent grade.
- LES-Talker: Fine-Grained Emotion Editing in Linear Emotion Space (Apr. 2024 - May 2025; Co-first Author, Accepted at IEEE Transactions on Affective Computing)
- Led the full research pipeline, from problem formulation to model design, experimentation, and paper writing.
- Proposed the Linear Emotion Space (LES), a novel interpretable framework enabling fine-grained emotion editing across types, intensities, and facial units.
- Designed LES-Talker with a universal Cross-Dimension Attention Network to align 3D model deformation with emotional control signals, achieving high-quality and controllable talking head generation.
- Multimodal-MOF: Metal-Organic Framework Design (Oct. 2025 - Feb. 2026; Lead Researcher, under the supervision of Prof. Shengchao Liu and Prof. Zhiling (Zach) Zheng)
- Built a pretraining pipeline for multi-modal MOF-related data and ran initial experiments.
- Conducted exploratory trials toward downstream transfer, identifying key bottlenecks and refining the research direction based on preliminary findings.
- KAN-Face: Efficient Resource Usage and Precision Lip-Sync (Mar. 2024 - Nov. 2024; Co-author, Published at ICASSP 2025)
- Contributed to the design of a lightweight framework utilizing KANs for efficient talking head generation.
- Actively contributed to rebuttal, focusing on highlighting contributions and addressing reviewer feedback.
- EmoSpeaker: Emotion-Controlled Talking Face Generation (Nov. 2023 - Feb. 2025; Co-author, Accepted at IEEE Transactions on Multimedia)
- Assisted in a one-shot framework capable of fine-grained emotional control and precise lip synchronization.
- Contributed to revision and rebuttal writing, focusing on technical clarity and supplementary experiments.
- Multi-modal Learning for Audio-driven Talking Head Generation (Nov. 2023 - Jan. 2024; Champion, Kaggle Competition, BDIV Lab)
- As 1 of 2 core developers, achieved 4 #1 and 1 #2 ranks among 9 metrics in both video- and audio-based tracks.
- Modified SadTalker by removing PoseNet and redesigning ExpNet for direct 70D 3DMM prediction, enabling strategic trade-offs that improved key directional metrics.
- Enhanced realism via innovative blinking and lip-sync strategies using OpenFace and Hubert, improving non-English alignment and eye movement synthesis.
Course Projects
- Public Transportation Management System - Android App Development (Team Leader)
- Led a team in developing a full-stack Android application for public transportation management, utilizing tools such as MySQL, Python, C++, and Qt Creator
- Designed a low-redundancy relational database tailored to project needs, and implemented a user-friendly visual interface with strong human-computer interaction capabilities
- Proposed a technical framework for lightweight management systems with similar structures, enabling easy expansion and portability across different use cases
- Exploration of Direct Preference Optimization
- Exploring key conceptual challenges in Direct Preference Optimization through a question-and-answer format
- Conducting a brief survey of Direct Preference Optimization from multiple perspectives
Skills
- Developer Tools: Python, Pytorch, C, Assembly Language, MySQL, C++, Java
- Languages: English (TOEFL iBT 96), Chinese (native)
