Zhiqi Kang

About Me

I’m a third-year Ph.D. candidate at Inria’s Thoth team, working under the supervision of Dr. Karteek Alahari. Since May 2024, we have started a collaboration with Dr. Rahaf Aljundi from Toyota Motor Europe to explore Continual learning with Vision Language Models.

Prior to joining the Ph.D. program, I had the opportunity to work as a Research Engineer with Inria and Meta Reality Labs where I applied my skills to audio-visual speech processing. The project was under the mentorship of Dr. Radu Horaud (Inria), Dr. Mostafa Sadeghi (Inria) and Dr. Jacob Donley (Meta) and Dr. Anurag Kumar (Meta).

I completed my master’s degree in Machine Learning and Optimization of Complex Systems from Université de Technologie de Compiègne (Sorbonne). I obtained my engineer’s degree in Decision Data Mining from Université de Technologie de Troyes.

Research Interests

• Continual Learning: Lifelong knowledge acquisition in realistic scenarios such as limited labeled data (ICCV 2023), imbalanced data (IJCAI 2024), online learning with stochastic data stream (ICLR 2025), and domain generalization or post-training (ongoing work).
• Vision-Language Learning and Reasoning: Enhancing vision-language models’ performance via in-context learning, test-time scaling, and model-based annotation (ongoing work).
• Audio-visual Speech Processing: Enhancing audio-visual processing with face frontalization (IJCV 2023 and ICASSP 2022).

News

• [2025.01] One paper about general continual learning based on prompt initialization and output layer masking is accepted by ICLR 2025. Huge thanks to all my co-authors! More details to come soon.

• [2024.04] I will be attending ICVSS 2024 summer school in Sicily. The topic of this year is Computer Vision in the Age of Large Language Models. Looking forward to it!

• [2024.04] One paper about task-imbalanced continual learning based on dynamic prompt learning is accepted by IJCAI 2024. Congrats to my collaborators! More details to come soon.

• [2023.08] One paper about semi-supervised continual learning is accepted by ICCV 2023 and is selected as an oral presentation (selection rate 1.8%)!

Publications

Services

Review

• ACM Multimedia: 2022-2024
• CVPR : 2023-2025
• NeurIPS : 2022-2024
• ICLR : 2023-2025
• ACCV, WACV: 2024
• ICML: 2025
• Computer Vision and Image Understanding (CVIU)
• IEEE Transactions on Neural Networks and Learning Systems (TNNLS)
• IEEE Transactions on Mobile Computing

Membership & Activities

• [2023-present] Member of the The BAAI Young Scientist Association
• [2023-present] Member of the The Computer Vision Foundation
• [2022-2023] Vice president of the Section Grenoble of Chinese Student and Scholar Association in Grenoble, France