Sitemap
A list of all the posts and pages found on the site. For you robots out there, there is an XML version available for digesting as well.
Pages
Kevin Tian
Kevin Tian, also publishing as Mu Tian / Tian, Mu, is a Senior Machine Learning Engineer at ByteDance/TikTok Singapore and ex-Meta Research Scientist working on LLM post-training, agentic/RAG search, multimodal foundation models, and generative vision.
Posts
Understanding Reinforcement Learning for LLMs — Part I: From Supervised Learning to Sequential Decision Making
Published:
Reinforcement learning has become one of the central ideas behind modern LLM post-training. Yet it is often discussed in a confusing way: sometimes as a mathematical framework, sometimes as an alignment recipe, sometimes as a set of algorithms such as PPO, DPO, GRPO, or RLHF.
community
Singapore Red Cross
I volunteer with Singapore Red Cross, supporting companionship, feeding assistance, and meaningful engagement activities for differently-abled residents at the Red Cross Home for the Disabled. I also initiated and hosted fundraising campaigns on Giving.sg to support community and humanitarian causes.
AWWA Dementia Day Care Centre
Volunteering with AWWA’s dementia and senior-care community in Singapore, supporting companionship, feeding assistance, meaningful engagement activities, and fundraising for eldercare and caregiver support.
portfolio
Resource-efficient LLM system for local-service search
A production-oriented LLM/SLM system for local-service search: structured extraction, POI grounding, LLM-as-a-judge verification, agent planning, RL-style post-training, and nearline cache-based deployment. It improved multilingual query understanding, increased cache reuse, and reduced online serving pressure under real search-engine constraints.
Multimodal Video Transformer
A multimodal video Transformer system for harmful-content understanding, combining video/audio/text contrastive pretraining, temporal sequence modeling, supervised fine-tuning, attention-based harmful-content localization, and PyTorch/TorchScript productionization. It contributed to substantial reductions in harmful-content prevalence, including a major reduction in political misinformation, while improving reviewer efficiency through temporal localization.
Generative AI & Interactive Learning for Annotation-efficient 3D Medical Vision
A generative and interactive medical AI project for annotation-efficient 3D segmentation, combining diffusion-based image-mask synthesis, scribble-supervised learning, attention/CRF regularization, and online human-in-the-loop adaptation. The system reduced dependence on dense expert labels while improving segmentation robustness under limited-label and real-world deployment constraints.
publications
MCMC Guided CNN Training and Segmentation for Pancreas Extraction
Published in IEEE Access, 2021
A probabilistic CNN-based pancreas segmentation method that uses MCMC sampling to guide 3D patch selection during training and segmentation, addressing anatomical variability, small organ size, and fuzzy boundaries in abdominal CT.
Recommended citation: Tian, M., He, J., Yu, X., Cai, C., & Gao, Y. (2021). "MCMC Guided CNN Training and Segmentation for Pancreas Extraction." IEEE Access, 9, 90539–90554.
Download Paper
Open-source Algorithm and Software for Computed Tomography-based Virtual Pancreatoscopy and Other Applications
Published in Visual Computing for Industry, Biomedicine, and Art, 2022
An open-source 3D Slicer-based virtual pancreatoscopy platform that combines pancreatic duct segmentation, optimal path planning, and super-resolution to enable CT-based fly-through visualization and quantitative duct analysis.
Recommended citation: Huang, H., Yu, X., Tian, M., He, W., Li, S. X., Liang, Z., & Gao, Y. (2022). "Open-source Algorithm and Software for Computed Tomography-based Virtual Pancreatoscopy and Other Applications." Visual Computing for Industry, Biomedicine, and Art, 5, Article 20.
Download Paper
Multi-scale Multi-task Distillation for Incremental 3D Medical Image Segmentation
Published in Computer Vision – ECCV 2022 Workshops, Lecture Notes in Computer Science, 2023
A knowledge-distillation framework for incremental 3D medical image segmentation, designed to preserve prior task knowledge while adapting segmentation models to new structures or datasets.
Recommended citation: Tian, M., Yang, Q., & Gao, Y. (2023). "Multi-scale Multi-task Distillation for Incremental 3D Medical Image Segmentation." Computer Vision – ECCV 2022 Workshops, Lecture Notes in Computer Science, 13803, 369–384.
Download Paper
A Dynamic Interactive Learning Framework for Automated 3D Medical Image Segmentation
Published in arXiv / CoRR, 2023
An interactive online-learning framework for 3D medical image segmentation that combines sparse user input, proxy-mask propagation, weak supervision, replay, and label smoothing to reduce annotation effort while maintaining robust segmentation performance.
Recommended citation: Tian, M., Chen, X., & Gao, Y. (2023). "A Dynamic Interactive Learning Framework for Automated 3D Medical Image Segmentation." arXiv preprint arXiv:2312.06072.
Download Paper
Synthesizing Images With Annotations for Medical Image Segmentation Using Diffusion Probabilistic Model
Published in International Journal of Imaging Systems and Technology, 2024
A diffusion-based medical image synthesis framework that generates annotation-aligned image-mask pairs, using texture style injection and frequency-domain attention to improve realism and downstream segmentation under limited-label settings.
Recommended citation: Huang, Z., Yang, Q., Tian, M., & Gao, Y. (2024). "Synthesizing Images With Annotations for Medical Image Segmentation Using Diffusion Probabilistic Model." International Journal of Imaging Systems and Technology, 35(1), e70007.
Download Paper
AttenScribble: Attention-enhanced Scribble Supervision for Medical Image Segmentation
Published in Journal of Visual Communication and Image Representation, 2025
A weakly supervised medical image segmentation framework that learns from sparse scribble annotations using pluggable spatial self-attention, attentive similarity regularization, partial segmentation loss, and masked CRF regularization.
Recommended citation: Tian, M., Yang, Q., & Gao, Y. (2025). "AttenScribble: Attention-enhanced Scribble Supervision for Medical Image Segmentation." Journal of Visual Communication and Image Representation, 110, 104476.
Download Paper
talks
Talk 1 on Relevant Topic in Your Field
Published:
This is a description of your talk, which is a markdown file that can be all markdown-ified like any other post. Yay markdown!
Conference Proceeding talk 3 on Relevant Topic in Your Field
Published:
This is a description of your conference proceedings talk, note the different field in type. You can put anything in this field.
teaching
Teaching experience 1
Undergraduate course, University 1, Department, 2014
This is a description of a teaching experience. You can use markdown like any other post.
Teaching experience 2
Workshop, University 1, Department, 2015
This is a description of a teaching experience. You can use markdown like any other post.