# Jingxuan Kang

> PhD Student in the Department of Electrical and Electronic Engineering, Imperial College London (London, UK). Research: multimodal large language models (vision-language models, unified multimodal models, video generation), toward world models; AI for science; recursive self-improvement (RSI).

Hi, I'm Jingxuan (Jensen)! I'm a PhD student in the [Department of Electrical and Electronic Engineering](https://www.imperial.ac.uk/electrical-engineering/) at [Imperial College London](https://www.imperial.ac.uk/).

I work on multimodal large language models (MLLMs) and am moving toward world models. I also explore AI for science and recursive self-improvement (RSI). My papers have appeared at CVPR, ICCV, ICML, NeurIPS, and ACM MM.

I serve as a reviewer for ICML, AAAI, CVPR, ICLR, NeurIPS, and ECCV.

Contact: j.kang26@imperial.ac.uk

## Research interests

- Vision-Language Models: Vision-language models that answer from the image itself rather than from prior expectation, from fine local detail to long multi-turn context.
- Unified Multimodal Models: Single models for image understanding and generation: the representations they learn, and the control they give over generation.
- Video Generation: Video models that keep objects and physical state consistent over time.
- AI for Science: Representations of cell morphology learned from Cell Painting images of chemical and genetic perturbations.

## Publications

- [Learning from Noisy Prompts: Saliency-Guided Prompt Distillation for Robust Segmentation with SAM](https://openaccess.thecvf.com/content/CVPR2026F/html/Kang_Learning_from_Noisy_Prompts_Saliency-Guided_Prompt_Distillation_for_Robust_Segmentation_CVPRF_2026_paper.html). J. Kang, Z. Zhang, S. Zheng, S. Li, U. B. Patel, A. H. Fitzhugh, P. Lung, Y. Kiberu, N. Jathanna, S. Jamil-Copley, B. Kainz, C. Qin. CVPR 2026 Findings (first author).
- [Thermal Diffusion Matters: Infrared Spatial-Temporal Video Super-Resolution through Heat Conduction Priors](https://openaccess.thecvf.com/content/CVPR2026/html/Zhou_Thermal_Diffusion_Matters_Infrared_Spatial-Temporal_Video_Super-Resolution_through_Heat_Conduction_CVPR_2026_paper.html). M. Zhou, S. Li, Y. Zhang, J. Geng, Y. Shen, J. Kang, F. Zhuang, S. Wang. CVPR 2026.
- [From Local Details to Global Context: Advancing Vision-Language Models with Attention-Based Selection](https://proceedings.mlr.press/v267/cai25d.html). J. Kang*, L. Cai*, S. Li, W. Ma, B. Xie, Z. Qin, J. Liang. ICML 2025 (co-first author; * equal contribution).
- [Learning Modality Knowledge Alignment for Cross-modality Transfer](https://arxiv.org/abs/2406.18864). W. Ma, S. Li, L. Cai, J. Kang. ICML 2024.
- [Enhancing Cross-Modal Fine-Tuning with Gradually Intermediate Modality Generation](https://arxiv.org/abs/2406.09003). L. Cai, S. Li, W. Ma, J. Kang, B. Xie, Z. Sun, C. Zhu. ICML 2024.
- [Translating Simulation Images to X-ray Images via Multi-Scale Semantic Matching](https://link.springer.com/chapter/10.1007/978-3-031-73748-0_10). J. Kang, T. Jianu, B. Huang, B. Bhattarai, N. Le, F. Coenen, A. Nguyen. MICCAI Workshop 2024 (first author).
- [Borrowing Knowledge From Pre-trained Language Model: A New Data-efficient Visual Learning Paradigm](http://openaccess.thecvf.com/content/ICCV2023/html/Ma_Borrowing_Knowledge_From_Pre-trained_Language_Model_A_New_Data-efficient_Visual_ICCV_2023_paper.html). W. Ma, S. Li, J. Zhang, C. H. Liu, J. Kang, Y. Wang, G. Huang. ICCV 2023.
- [Style Transfer Meets Super-Resolution: Advancing Unpaired Infrared-to-Visible Image Translation with Detail Enhancement](https://dl.acm.org/doi/abs/10.1145/3581783.3612492). J. Kang*, Y. Shen*, S. Li, Z. Yu, S. Wang. ACM MM 2023 (co-first author; * equal contribution).
- [Language Semantic Graph Guided Data-efficient Learning](https://proceedings.neurips.cc/paper_files/paper/2023/hash/4bbd69ce4cea6aa3a08bde08a40fd65a-Abstract-Conference.html). W. Ma, S. Li, L. Cai, J. Kang. NeurIPS 2023.

## News

- 2026.09: One paper accepted to NeurIPS 2026: Beyond Pairwise Supervision: Spectral Characteristic Matching for Data-Efficient Multimodal Alignment.
- 2026.06: ICML 2026: Selected as a Gold Reviewer.
- 2026.04: Started PhD at Imperial College London.
- 2026.03: Two papers accepted to CVPR 2026: Learning from Noisy Prompts: Saliency-Guided Prompt Distillation for Robust Segmentation with SAM (Findings); Thermal Diffusion Matters: Infrared Spatial-Temporal Video Super-Resolution through Heat Conduction Priors.
- 2025.05: One paper accepted to ICML 2025: From Local Details to Global Context: Advancing Vision-Language Models with Attention-Based Selection.
- 2024.07: Two papers accepted to ICML 2024: Learning Modality Knowledge Alignment for Cross-modality Transfer; Enhancing Cross-Modal Fine-Tuning with Gradually Intermediate Modality Generation.
- 2024.06: One paper accepted to MICCAI Workshop 2024: Translating Simulation Images to X-ray Images via Multi-Scale Semantic Matching.
- 2023.10: One paper accepted to NeurIPS 2023: Language Semantic Graph Guided Data-efficient Learning.
- 2023.10: One paper accepted to ICCV 2023: Borrowing Knowledge From Pre-trained Language Model: A New Data-efficient Visual Learning Paradigm.
- 2023.07: One paper accepted to ACM MM 2023: Style Transfer Meets Super-Resolution: Advancing Unpaired Infrared-to-Visible Image Translation with Detail Enhancement.

## Service

- Reviewer: ICML, AAAI, CVPR, ICLR, NeurIPS, ECCV

## Blog

- [我的 AI 订阅清单：一份主观推荐](https://jingxuan.uk/blog/my-ai-stack-2026.html) (2026.05.19): 把主流 AI 的最高档订阅都买了一遍之后，我对 Claude / ChatGPT / Gemini / Grok / 豆包 / Coze 的主观推荐。
- [AI 科研不完全指北第二期：如何 Propose 一个 idea](https://jingxuan.uk/blog/propose-idea.html) (2026.05.19): 从 Evaluate Idea 到 Propose Idea 的科研思考——好的 idea 不是被想出来的，而是从实验信号中长出来的。
- [REPA：让 DiT 训得快一点](https://jingxuan.uk/blog/repa.html) (2026.05.19): 一句话讲清楚 REPA：给 DiT 中间层加一条对齐 DINOv2 的 loss，训练直接快一档。
- [AI 科研不完全指北：Top10 PhD 烧了上万刀的真实经验。Context Is All You Need](https://jingxuan.uk/blog/context-is-all-you-need.html) (2026.05.18): 一位 Top10 CS PhD 用 Claude / GPT / Gemini 协作做深度学习科研的踩坑经验——所有 workflow 设计都在做一件事：管住每一个 session 的上下文。

## Links

- [Homepage](https://jingxuan.uk/)
- [Google Scholar](https://scholar.google.com/citations?user=Ij5yAYIAAAAJ&hl=en)
- [GitHub](https://github.com/JingxuanKang)
- [X](https://x.com/NeverMoreNg)