Juanxi Tian/田隽熙

About

I am currently a senior student at Hong Kong Baptist University, majoring in artificial intelligence. And long-term with Westlake University, Peking University and other universities and research institutions research partners to maintain cooperative relations.

My areas of interest include Multimodal Large Language Models, Optimization in Deep Learning, and Computer Vision.

You can find me at the following websites:

My research interests are focused on vision and learning, in-depth understanding of the training and generalization of complex deep learning systems (coupling and counter-intuitive phenomena), and continuous in-depth analysis and improvement from the perspective of optimization, so as to promote the generation of more robust artificial (general) intelligence technologies.

At present, I am actively looking for PhD opportunities, please contact me if you are interested.

I am also open to interesting research opportunities. If you are interested in collaborating with me, please feel free to contact me:


News

Stay updated with the latest news and achievements:

  • Sep 26th, 2024: CARES received by NIPS'24! Congratulations to all collaborators!
  • Aug 23th, 2024: I am invited to serve as a Reviewer for ICLR 2025, which will be held in Singapore.
  • Aug 11th, 2024: I founded the Black-Box Optimization & Coupling Bias (BOCB) Organization with Siyuan Li and Zedong Wang, which will make for some interesting series work!
  • July 14th, 2024: I maintain an open source repository on the optimizer called Awesome-Optimizers, with Zedong Wang and Siyuan Li.
  • July 3th, 2024: A shorter version of CARES was accepted by ICML 2024 Workshop on Foundation Models in the Wild.
  • July 1th, 2024: LMFlow supports custom optimization fine-tuning by major optimizers. And it won the Best Demo Paper Award at NAACL 2024

Organization

BOCB Logo

Black-Box Optimization & Coupling Bias (BOCB)

  • Our team is committed to advancing the scientific understanding of black-box optimization techniques and the pervasive Coupling Bias phenomena observed in contemporary AI systems, particularly deep neural networks. Through rigorous theoretical analysis and empirical investigation, we aim to shed light on the fundamental mechanisms underlying these prevalent challenges in the field of artificial intelligence.
  • Key Focus Areas:
    • Black-Box Optimization: Exploring advanced methods to enhance the efficiency and effectiveness of optimization processes in complex, non-linear systems.
    • Coupling Bias: Investigating the ubiquitous bias that arises from the interdependence of variables in AI models, with a focus on its impact on model performance and generalization.
    • Deep Neural Networks: Delving into the intricacies of deep learning architectures to understand and mitigate the effects of coupling bias.

Internship Experience

Chinese Academy of Sciences

Research Intern at Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences

Start Date: 2023-07-26, End Date: 2023-09-6

Westlake University

Research Intern at Center for Artificial Intelligence Research and Innovation Lab, Westlake University

Start Date: 2023-07-27, End Date: Present

International Machine Learning Research Center, Peking University

Research Intern at International Machine Learning Research Center, Peking University

Start Date: 2024-03-9, End Date: Present

International Machine Learning Research Center, Peking University

Research Intern at MMLab, CUHK

Start Date: 2024-10-20, End Date: Present


International Collaboration

UNC

Collaboration with UNC

Start Date: 2024-04-23, End Date: 2024-9-25

Supervised by Professor Huaxiu Yao.

UIUC

Collaboration with UIUC

Start Date: 2024-06-15, End Date: Present

Supervised by Professor Tong Zhang.


Preprints

OpenMixup: Open Mixup Toolbox and Benchmark for Visual Representation Learning

OpenMixup

OpenMixup is currently available on the arXiv preprint website.

Switch EMA: A Free Lunch for Better Flatness and Sharpness

Switch EMA

SEMA is currently available on the arXiv preprint website.

CARES: A Comprehensive Benchmark of Trustworthiness in Medical Vision Language Models

CARES

CARES is currently available on the arXiv preprint website.

Trans4D: Realistic Geometry-Aware Transition for Compositional Text-to-4D Synthesis

Trans4D

Trans4D is currently available on the arXiv preprint website.

Unveiling the Backbone-Optimizer Coupling Bias in Visual Representation Learning

BOCB

BOCB is currently available on the arXiv preprint website.


Services

Reviewer & Program Committee

  • Conference: 2024 - International Conference on Learning Representations (ICLR), 2025.
  • Conference: 2024 - Association for the Advancement of Artificial Intelligence (AAAI), 2025.
  • Academic Journal: 2024 - IEEE’s Transactions on Knowledge and Data Engineering (IEEE TKDE)
  • Academic Journal: 2024 - Neural Networks

Membership

  • 2024 - Student Member of IEEE
  • 2024 - Student Member of China Computer Federation (CCF)
  • 2023 - Student Member of Chinese Association for Artificial Intelligence (CAAI)

Visitor Map