Yi Pan (潘逸)

The Chinese University of Hong Kong, Shenzhen Ph.D. Student, The University of Georgia (2024)

Hi! I am Yi Pan and I am a second-year Ph.D. student at The University of Georgia, supervised by Distinguished Research Prof. Tianming Liu. My current research areas foucs on Large Language Models (LLMs), Quantum AI, and Multimodal Learning.

Google Scholar GitHub LinkedIn ORCID
Education
  • The University of Georgia

    The University of Georgia

    Ph.D. in Computer Science Aug. 2024 - Now

  • University of Glasgow

    University of Glasgow

    B.S. in Electrical and Electronic Engineering Sep. 2020 - Jul. 2024

  • University of Electronic Science and Technology of China

    University of Electronic Science and Technology of China

    B.S. in Electrical Information Engineering Sep. 2020 - Jul. 2024

Honors & Awards
  • NSF Student Travel Award, AAAI FSS25 (QIML) 2025
  • Outstanding Graduate, Glasgow College, UESTC 2024
  • First Prize Scholarship for Academic Excellence, UESTC 2022
  • Scholarship for English Proficiency, Glasgow College, UESTC 2022
  • First Prize Scholarship for Academic Excellence, UESTC 2021
  • Academic Scholarship, Glasgow College, UESTC 2021
Experience
  • School of Computing, UGA

    School of Computing, UGA

    Research Assistant Aug. 2024 - Now

  • School of Computing, UGA

    School of Computing, UGA

    Teaching Assistant Aug. 2025 - Now

  • Center for Advanced Medical Computing and Analysis (CAMCA), MGB/Harvard Medical School

    Center for Advanced Medical Computing and Analysis (CAMCA), MGB/Harvard Medical School

    Graduate Research Intern May 2025 - Aug. 2025

  • Glasgow College, UESTC

    Glasgow College, UESTC

    Teaching Assistant Sep. 2023 - Jun. 2024

Service
News
2025
🎉 Received NSF Student Travel Award for AAAI FSS25 (QIML)!
Sep 15
Aug 25
Aug 20
Aug 11
Aug 07
Selected Publications (view all )
Bridging Classical and Quantum Computing for Next-Generation Language Models
Bridging Classical and Quantum Computing for Next-Generation Language Models

Yi Pan, Hanqi Jiang, Junhao Chen, Yiwei Li, Huaqin Zhao, Lin Zhao, Yohannes Abate, Yingfeng Wang†, Tianming Liu†(† corresponding author)

AAAI Symposium on Quantum Information & Machine Learning (QIML) 2025 Conference

Integrating Large Language Models (LLMs) with quantum computing is a critical challenge, hindered by the severe constraints of Noisy Intermediate-Scale Quantum (NISQ) devices, including barren plateaus and limited coherence. Current approaches often fail due to static quantum-classical partitioning. We introduce… Integrating Large Language Models (LLMs) with quantum computing is a critical challenge, hindered by the severe constraints of Noisy Intermediate-Scale Quantum (NISQ) devices, including barren plateaus and limited coherence. Current approaches often fail due to static quantum-classical partitioning. We introduce Adaptive Quantum-Classical Fusion (AQCF), the first framework to bridge this gap through dynamic, quantum-classical co-design. AQCF's core principle is real-time adaptation: it analyzes input complexity to orchestrate seamless transitions between classical and quantum processing. The framework features three key innovations: (1) entropy-driven adaptive circuits that circumvent barren plateaus; (2) quantum memory banks that unify classical attention with quantum state-based similarity retrieval; and (3) intelligent fusion controllers that allocate tasks for optimal performance. This architecture maintains full compatibility with classical Transformers while progressively incorporating quantum advantages. Experiments on sentiment analysis demonstrate that AQCF achieves competitive performance, significantly improves quantum resource efficiency, and operates successfully within typical NISQ constraints. By providing a seamless integration pathway, AQCF offers both immediate practical value on current quantum hardware and a clear evolution path toward mature Quantum LLMs.

[Paper]

Bridging Classical and Quantum Computing for Next-Generation Language Models
Bridging Classical and Quantum Computing for Next-Generation Language Models

Yi Pan, Hanqi Jiang, Junhao Chen, Yiwei Li, Huaqin Zhao, Lin Zhao, Yohannes Abate, Yingfeng Wang†, Tianming Liu†(† corresponding author)

AAAI Symposium on Quantum Information & Machine Learning (QIML) 2025 Conference

Integrating Large Language Models (LLMs) with quantum computing is a critical challenge, hindered by the severe constraints of Noisy Intermediate-Scale Quantum (NISQ) devices, including barren plateaus and limited coherence. Current approaches often fail due to static quantum-classical partitioning. We introduce… Integrating Large Language Models (LLMs) with quantum computing is a critical challenge, hindered by the severe constraints of Noisy Intermediate-Scale Quantum (NISQ) devices, including barren plateaus and limited coherence. Current approaches often fail due to static quantum-classical partitioning. We introduce Adaptive Quantum-Classical Fusion (AQCF), the first framework to bridge this gap through dynamic, quantum-classical co-design. AQCF's core principle is real-time adaptation: it analyzes input complexity to orchestrate seamless transitions between classical and quantum processing. The framework features three key innovations: (1) entropy-driven adaptive circuits that circumvent barren plateaus; (2) quantum memory banks that unify classical attention with quantum state-based similarity retrieval; and (3) intelligent fusion controllers that allocate tasks for optimal performance. This architecture maintains full compatibility with classical Transformers while progressively incorporating quantum advantages. Experiments on sentiment analysis demonstrate that AQCF achieves competitive performance, significantly improves quantum resource efficiency, and operates successfully within typical NISQ constraints. By providing a seamless integration pathway, AQCF offers both immediate practical value on current quantum hardware and a clear evolution path toward mature Quantum LLMs.

[Paper]

Quantum-Classical Hybrid Molecular Autoencoder for Advancing Classical Decoding
Quantum-Classical Hybrid Molecular Autoencoder for Advancing Classical Decoding

Afrar Jahin, Yi Pan, Yingfeng Wang†, Tianming Liu†, Wei Zhang†(† corresponding author)

First AAAI Symposium on Quantum Information & Machine Learning (QIML) 2025 Conference

Although recent advances in quantum machine learning (QML) offer significant potential for enhancing generative models, particularly in molecular design, many classical approaches still struggle to achieve high fidelity and validity. We propose a hybrid quantum–classical architecture for SMILES reconstruction that… Although recent advances in quantum machine learning (QML) offer significant potential for enhancing generative models, particularly in molecular design, many classical approaches still struggle to achieve high fidelity and validity. We propose a hybrid quantum–classical architecture for SMILES reconstruction that integrates quantum encoding with classical sequence modeling to improve quantum fidelity and classical similarity. Our approach achieves around 84% quantum fidelity and 60% classical reconstruction similarity, surpassing existing quantum baselines, and provides a promising foundation for quantum‑aware sequence models in molecular and drug discovery.

[Paper]

Quantum-Classical Hybrid Molecular Autoencoder for Advancing Classical Decoding
Quantum-Classical Hybrid Molecular Autoencoder for Advancing Classical Decoding

Afrar Jahin, Yi Pan, Yingfeng Wang†, Tianming Liu†, Wei Zhang†(† corresponding author)

First AAAI Symposium on Quantum Information & Machine Learning (QIML) 2025 Conference

Although recent advances in quantum machine learning (QML) offer significant potential for enhancing generative models, particularly in molecular design, many classical approaches still struggle to achieve high fidelity and validity. We propose a hybrid quantum–classical architecture for SMILES reconstruction that… Although recent advances in quantum machine learning (QML) offer significant potential for enhancing generative models, particularly in molecular design, many classical approaches still struggle to achieve high fidelity and validity. We propose a hybrid quantum–classical architecture for SMILES reconstruction that integrates quantum encoding with classical sequence modeling to improve quantum fidelity and classical similarity. Our approach achieves around 84% quantum fidelity and 60% classical reconstruction similarity, surpassing existing quantum baselines, and provides a promising foundation for quantum‑aware sequence models in molecular and drug discovery.

[Paper]

HELENE: Hessian Layer-wise Clipping and Gradient Annealing for Accelerating Fine-tuning LLM with Zeroth-order Optimization
HELENE: Hessian Layer-wise Clipping and Gradient Annealing for Accelerating Fine-tuning LLM with Zeroth-order Optimization

Huaqin Zhao, Jiaxi Li, Yi Pan, Shizhe Liang, Xiaofeng Yang, Wei Liu, Xiang Li, Fei Dou, Tianming Liu, Jin Lu†(† corresponding author)

Empirical Methods in Natural Language Processing (EMNLP 2025) 2025 Conference

Fine-tuning large language models (LLMs) poses significant memory challenges, as the back-propagation process demands extensive resources, especially with growing model sizes. Recent work, MeZO, addresses this issue using a zeroth-order (ZO) optimization method, which reduces memory consumption by matching the… Fine-tuning large language models (LLMs) poses significant memory challenges, as the back-propagation process demands extensive resources, especially with growing model sizes. Recent work, MeZO, addresses this issue using a zeroth-order (ZO) optimization method, which reduces memory consumption by matching the usage to the inference phase. However, MeZO experiences slow convergence due to varying curvatures across model parameters. To overcome this limitation, we introduce HELENE, a novel scalable and memory-efficient optimizer that integrates annealed A-GNB gradients with a diagonal Hessian estimation and layer-wise clipping, serving as a second-order pre-conditioner. This combination allows for faster and more stable convergence. Our theoretical analysis demonstrates that HELENE improves convergence rates, particularly for models with heterogeneous layer dimensions, by reducing the dependency on the total parameter space dimension. Instead, the method scales with the largest layer dimension, making it highly suitable for modern LLM architectures. Experimental results on RoBERTa-large and OPT-1.3B across multiple tasks show that HELENE achieves up to a 20x speedup compared to MeZO, with average accuracy improvements of 1.5%. Furthermore, HELENE remains compatible with both full parameter tuning and parameter-efficient fine-tuning (PEFT), outperforming several state-of-the-art optimizers. The codes will be released after reviewing.

[Paper]

HELENE: Hessian Layer-wise Clipping and Gradient Annealing for Accelerating Fine-tuning LLM with Zeroth-order Optimization
HELENE: Hessian Layer-wise Clipping and Gradient Annealing for Accelerating Fine-tuning LLM with Zeroth-order Optimization

Huaqin Zhao, Jiaxi Li, Yi Pan, Shizhe Liang, Xiaofeng Yang, Wei Liu, Xiang Li, Fei Dou, Tianming Liu, Jin Lu†(† corresponding author)

Empirical Methods in Natural Language Processing (EMNLP 2025) 2025 Conference

Fine-tuning large language models (LLMs) poses significant memory challenges, as the back-propagation process demands extensive resources, especially with growing model sizes. Recent work, MeZO, addresses this issue using a zeroth-order (ZO) optimization method, which reduces memory consumption by matching the… Fine-tuning large language models (LLMs) poses significant memory challenges, as the back-propagation process demands extensive resources, especially with growing model sizes. Recent work, MeZO, addresses this issue using a zeroth-order (ZO) optimization method, which reduces memory consumption by matching the usage to the inference phase. However, MeZO experiences slow convergence due to varying curvatures across model parameters. To overcome this limitation, we introduce HELENE, a novel scalable and memory-efficient optimizer that integrates annealed A-GNB gradients with a diagonal Hessian estimation and layer-wise clipping, serving as a second-order pre-conditioner. This combination allows for faster and more stable convergence. Our theoretical analysis demonstrates that HELENE improves convergence rates, particularly for models with heterogeneous layer dimensions, by reducing the dependency on the total parameter space dimension. Instead, the method scales with the largest layer dimension, making it highly suitable for modern LLM architectures. Experimental results on RoBERTa-large and OPT-1.3B across multiple tasks show that HELENE achieves up to a 20x speedup compared to MeZO, with average accuracy improvements of 1.5%. Furthermore, HELENE remains compatible with both full parameter tuning and parameter-efficient fine-tuning (PEFT), outperforming several state-of-the-art optimizers. The codes will be released after reviewing.

[Paper]

ChatRadio-Valuer: A Chat Large Language Model for Generalizable Radiology Impression Generation on Multi-institution and Multi-system Data
ChatRadio-Valuer: A Chat Large Language Model for Generalizable Radiology Impression Generation on Multi-institution and Multi-system Data

Tianyang Zhong, Wei Zhao, Yutong Zhang, Yi Pan, Peixin Dong, Zuowei Jiang, Hanqi Jiang, Yifan Zhou, Xiaoyan Kui, Youlan Shang, Lin Zhao, Li Yang, Yaonai Wei, Zhuoyi Li, Jiadong Zhang, Longtao Yang, Hao Chen, Huan Zhao, Yuxiao Liu, Ning Zhu, Yiwei Li, Yisong Wang, Jiaqi Yao, Jiaqi Wang, Ying Zeng, Lei He, Chao Zheng, Zhixue Zhang, Ming Li, Zhengliang Liu, Haixing Dai, Zihao Wu, Lu Zhang, Shu Zhang, Xiaoyan Cai, Xintao Hu, Shijie Zhao, Xi Jiang, Xin Zhang, Wei Liu, Xiang Li, Dajiang Zhu, Lei Guo, Dinggang Shen, Junwei Han, Tianming Liu, Jun Liu, Tuo Zhang†(† corresponding author)

IEEE Transactions on Biomedical Engineering (TBME) 2025 Journal

Achieving clinical level performance and widespread deployment for generating radiology impressions encounters a giant challenge for conventional artificial intelligence models tailored to specific diseases and organs. Concurrent with the increasing accessibility of radiology reports and advancements in modern general AI… Achieving clinical level performance and widespread deployment for generating radiology impressions encounters a giant challenge for conventional artificial intelligence models tailored to specific diseases and organs. Concurrent with the increasing accessibility of radiology reports and advancements in modern general AI techniques, the emergence and potential of deployable radiology AI exploration have been bolstered. Here, we present ChatRadio-Valuer, the first general radiology diagnosis large language model for localized deployment within hospitals and being close to clinical use for multi-institution and multi-system diseases. ChatRadio-Valuer achieved 15 state-of-the-art results across five human systems and six institutions in clinical-level events (n=332,673) through rigorous and full-spectrum assessment, including engineering metrics, clinical validation, and efficiency evaluation. Notably, it exceeded OpenAI's GPT-3.5 and GPT-4 models, achieving superior performance in comprehensive disease diagnosis compared to the average level of radiology experts. Besides, ChatRadio-Valuer supports zero-shot transfer learning, greatly boosting its effectiveness as a radiology assistant, while ensuring adherence to privacy standards and being readily utilized for large-scale patient populations. Our expeditions suggest the development of localized LLMs would become an imperative avenue in hospital applications.

[Paper]

ChatRadio-Valuer: A Chat Large Language Model for Generalizable Radiology Impression Generation on Multi-institution and Multi-system Data
ChatRadio-Valuer: A Chat Large Language Model for Generalizable Radiology Impression Generation on Multi-institution and Multi-system Data

Tianyang Zhong, Wei Zhao, Yutong Zhang, Yi Pan, Peixin Dong, Zuowei Jiang, Hanqi Jiang, Yifan Zhou, Xiaoyan Kui, Youlan Shang, Lin Zhao, Li Yang, Yaonai Wei, Zhuoyi Li, Jiadong Zhang, Longtao Yang, Hao Chen, Huan Zhao, Yuxiao Liu, Ning Zhu, Yiwei Li, Yisong Wang, Jiaqi Yao, Jiaqi Wang, Ying Zeng, Lei He, Chao Zheng, Zhixue Zhang, Ming Li, Zhengliang Liu, Haixing Dai, Zihao Wu, Lu Zhang, Shu Zhang, Xiaoyan Cai, Xintao Hu, Shijie Zhao, Xi Jiang, Xin Zhang, Wei Liu, Xiang Li, Dajiang Zhu, Lei Guo, Dinggang Shen, Junwei Han, Tianming Liu, Jun Liu, Tuo Zhang†(† corresponding author)

IEEE Transactions on Biomedical Engineering (TBME) 2025 Journal

Achieving clinical level performance and widespread deployment for generating radiology impressions encounters a giant challenge for conventional artificial intelligence models tailored to specific diseases and organs. Concurrent with the increasing accessibility of radiology reports and advancements in modern general AI… Achieving clinical level performance and widespread deployment for generating radiology impressions encounters a giant challenge for conventional artificial intelligence models tailored to specific diseases and organs. Concurrent with the increasing accessibility of radiology reports and advancements in modern general AI techniques, the emergence and potential of deployable radiology AI exploration have been bolstered. Here, we present ChatRadio-Valuer, the first general radiology diagnosis large language model for localized deployment within hospitals and being close to clinical use for multi-institution and multi-system diseases. ChatRadio-Valuer achieved 15 state-of-the-art results across five human systems and six institutions in clinical-level events (n=332,673) through rigorous and full-spectrum assessment, including engineering metrics, clinical validation, and efficiency evaluation. Notably, it exceeded OpenAI's GPT-3.5 and GPT-4 models, achieving superior performance in comprehensive disease diagnosis compared to the average level of radiology experts. Besides, ChatRadio-Valuer supports zero-shot transfer learning, greatly boosting its effectiveness as a radiology assistant, while ensuring adherence to privacy standards and being readily utilized for large-scale patient populations. Our expeditions suggest the development of localized LLMs would become an imperative avenue in hospital applications.

[Paper]

MolQAE: Quantum Autoencoder for Molecular Representation Learning
MolQAE: Quantum Autoencoder for Molecular Representation Learning

Yi Pan, Hanqi Jiang, Wei Ruan, Dajiang Zhu, Xiang Li, Yohannes Abate, Yingfeng Wang†, Tianming Liu†(† corresponding author)

IEEE International Conference on Quantum Artificial Intelligence (QAI 2025) 2025 Conference

We introduce Quantum Molecular Autoencoder (MolQAE), the first quantum autoencoder to leverage the complete molecular structures. MolQAE uniquely maps SMILES strings directly to quantum states using parameterized rotation gates, preserving vital structural information. Its quantum encoder–decoder framework enables latent space… We introduce Quantum Molecular Autoencoder (MolQAE), the first quantum autoencoder to leverage the complete molecular structures. MolQAE uniquely maps SMILES strings directly to quantum states using parameterized rotation gates, preserving vital structural information. Its quantum encoder–decoder framework enables latent space compression and reconstruction. A dual-objective strategy optimizes fidelity and minimizes trash state deviation. Our evaluations demonstrate effective capture of molecular characteristics and strong fidelity preservation under substantial dimensionality reduction, establishing a quantum pathway in cheminformatics for NISQ-era hardware and promising advances in drug and materials discovery.

[Paper]

MolQAE: Quantum Autoencoder for Molecular Representation Learning
MolQAE: Quantum Autoencoder for Molecular Representation Learning

Yi Pan, Hanqi Jiang, Wei Ruan, Dajiang Zhu, Xiang Li, Yohannes Abate, Yingfeng Wang†, Tianming Liu†(† corresponding author)

IEEE International Conference on Quantum Artificial Intelligence (QAI 2025) 2025 Conference

We introduce Quantum Molecular Autoencoder (MolQAE), the first quantum autoencoder to leverage the complete molecular structures. MolQAE uniquely maps SMILES strings directly to quantum states using parameterized rotation gates, preserving vital structural information. Its quantum encoder–decoder framework enables latent space… We introduce Quantum Molecular Autoencoder (MolQAE), the first quantum autoencoder to leverage the complete molecular structures. MolQAE uniquely maps SMILES strings directly to quantum states using parameterized rotation gates, preserving vital structural information. Its quantum encoder–decoder framework enables latent space compression and reconstruction. A dual-objective strategy optimizes fidelity and minimizes trash state deviation. Our evaluations demonstrate effective capture of molecular characteristics and strong fidelity preservation under substantial dimensionality reduction, establishing a quantum pathway in cheminformatics for NISQ-era hardware and promising advances in drug and materials discovery.

[Paper]

All publications