Research Interests

Reinforcement learning in non-stationary environments, robot learning, spatial memory, world models, online adaptation, and human–robot collaboration.

Education

Liaoning Normal University (LNNU), Dalian, China
School of Computer Science and Artificial Intelligence
Undergraduate Student in Artificial Intelligence

Sep. 2024 - Expected Jun. 2028

• Ranked 1/60 students in the Artificial Intelligence major.
• Recipient of the First-Class Scholarship in 2025.
• Relevant coursework: Data Structures and Algorithms, Machine Learning, Deep Learning, Operating Systems, Computer Networks, and Database Systems.

Research Experience

Deep Learning for Human-Robot Collaboration

Undergraduate Researcher

Nov. 2024 - Present

• Conduct research under faculty supervision on predicting a human partner’s intended target object from hand-motion information for proactive human–robot collaboration.
• Served as the sole implementation contributor, independently developing and maintaining the project’s full research codebase, including PyTorch models, data-processing pipelines, training procedures, evaluation scripts, and experimental configurations.
• Designed and conducted comparative experiments, ablation studies, and robustness analyses, and interpreted the resulting performance differences and failure cases.
• A manuscript based on this project was previously submitted to IEEE Robotics and Automation Letters (RA-L). Reviewer feedback identified insufficient real-robot validation as a key limitation; we are currently conducting physical robot experiments and revising the system accordingly.
• Developing a real-robot validation pipeline using an RGB-D camera, hand-pose estimation, target-object prediction, and coordinate transformation between camera and robot frames.

Instructional Sequencing over Prerequisite DAGs

Experiments Contributor

Jun. 2026 – Jul. 2026

• Participated in an international, cross-institutional student research collaboration led by a student researcher at the University of California, Davis.
• Configured the experimental environment and reproduced existing baseline methods under the project’s evaluation settings.
• Ran comparative experiments, collected and checked experimental results, and organized the reported results into tables for the manuscript.
• The resulting manuscript was submitted to AAAI 2027, with authors 2–4 designated as equal contributors and listed alphabetically.

Ongoing Research

Recurring Dynamics Adaptation

Independent Research Project

Jul. 2026 – Present

• Investigating whether retrieving and restoring previously learned policies can reduce re-adaptation time compared with continual single-policy updating when environment dynamics recur in an unobserved A→B→A sequence.
• Designed a hidden-switch benchmark based on MiniGrid-LavaGapS7-v0 and implemented a PPO continual-adaptation baseline.
• Structured the evaluation around change-detection delay, interactions required to recover task performance, and the relative speed of relearning previously encountered dynamics.
• Current experiments reveal substantial exploration difficulty after the first dynamics shift and slow recovery when the original dynamics reappears, motivating further investigation into memory-based model retrieval and interference reduction.

Selected Projects

inside-transformer

Independent Learning Project

A minimal, educational implementation of modern Transformer architectures in PyTorch. https://github.com/Antrairs/inside-transformer

• Implemented self-attention, multi-head attention, rotary positional embeddings, feed-forward networks, normalization variants, and Transformer encoder/decoder blocks from scratch.
• Developed controlled experiments to examine attention behavior, positional information, and sequence-processing capabilities.
• Studied practical implementation topics including variable-length sequences, sequence packing, and pre-norm versus post-norm architectures.
• Wrote explanatory documentation connecting mathematical formulations with executable code.

Additional Technical Experience

Laboratory Deep Learning Server Maintenance

Nov. 2024 – Present

• Assisted in maintaining a shared Ubuntu 22.04 deep-learning server for laboratory research.
• Configured multi-user workspaces, Python and CUDA/PyTorch environments, SSH access, and remote networking.
• Diagnosed dependency, GPU-memory, and development-environment issues encountered during research experiments.

Technical Skills

• Programming: Python, TypeScript/JavaScript; working knowledge of C/C++, Java/Kotlin, and SQL.
• Machine Learning: PyTorch, NumPy, Pandas, Seaborn, Matplotlib, scikit-learn, and Hugging Face Transformers.
• Research Engineering: Model implementation, training and evaluation pipelines, data preprocessing, controlled experiments, result analysis, and reproducible environment setup.
• Systems and Tools: Linux, Git, Docker, SSH, CUDA/PyTorch environment configuration, and GPU server maintenance.
• Additional Development: Bun, Vite, React, TypeScript, Tailwind CSS, and Astro.
• English: Passed CET-6 (557); able to read technical documentation and academic papers in English.