I am seeking a Ph.D. position!

RA @ Singapore Management University

Prev. Research Intern @ THUNLP

Excited for the future of AI, want to be a part of it!

Email: [email protected]
Hometown: Chengdu, China

GitHub | Google Scholar

M.Eng. in Computer Science and Technology
Beijing Language and Culture University
September 2021 - July 2024

B.Eng. in Computer Science and Technology
Sichuan University
September 2017 - July 2021

Singapore Management University
Sept 2024 - present

• Conducting LLM research under the supervision of Professor Yang Deng.
• Exploring LLM as educational agents for data science tasks.

THUNLP, Tsinghua University
April 2023 - July 2024

• Conducting NLP research under the supervision of Shuo Wang, Ph.D.
• Distilled table reasoning skills from LLMs to small PLMs.
• Developed LLM agents for scientific data visualization.
• Curated multilingual SFT data.
• Participated in devising an agent framework for processing long context inputs.

• Authors: Zhiyu Yang, Zihan Zhou, Shuo Wang, Xin Cong, Xu Han, Yukun Yan, Zhenghao Liu, Zhixing Tan, Pengyuan Liu, Dong Yu, Zhiyuan Liu, Xiaodong Shi, Maosong Sun.
• Summary: Introduced MatPlotBench for automatic evaluation of AI methods for scientific data visualization. Proposed MatPlotAgent, a framework using visual feedback to enhance LLM performance.
• Contribution: Designed the agent framework and evaluation method, conducted experiments, and curated data for MatPlotBench.

• Authors: Haoyu Wang, Shuo Wang, Yukun Yan, Xujia Wang, Zhiyu Yang, Yuzhuang Xu, Zhenghao Liu, Liner Yang, Ning Ding, Xu Han, Zhiyuan Liu, Maosong Sun.
• Summary: Developed a multilingual SFT dataset with language-specific and language-agnostic subsets using knowledge-enhanced data augmentation methods with Wikipedia as knowledge source.
• Contribution: Concretized the paper's idea, designed initial prompt templates for data synthesis, and revised the paper.

• Authors: Zhiyu Yang, Shuo Wang, Yukun Yan, Pengyuan Liu, Dong Yu.
• Summary: Proposed a knowledge distillation method for table QA tasks using relevant-cell-based rationales, achieving SOTA results on the FeTaQA benchmark.
• Contribution: Developed the distillation method, conducted experiments, and authored the paper.

• Converted MatPlotAgent into an interactive online demo.
• Demonstrated its workflow and performance to scholars attending CCL 2024.

• Explored various pre-trained language models for understanding the plausibility of implicit and underspecified texts.
• Fine-tuned Facebook AI’s MUPPET model for optimal performance.

• Proposed a novel garbage classification deep neural network architecture.
• Achieved superior performance compared to mainstream models on a Huawei Cloud Garbage Classification Competition dataset.

• Mandarin: Native
• English: Fluent (IELTS Overall Band 8.0, Reading: 9, Listening: 9, Writing: 7.5, Speaking: 7)

• Python, C++, Java

• PyTorch, Huggingface, PyG, Keras, TensorFlow 2, Linux, Android Studio, vLLM, Matplotlib, Numpy, Pandas

• Convolutional Neural Networks, Pre-trained NLU and NLG models, LLMs