About Me

Hi, my name is Liyang Song.

I am a deep learning researcher at the Institute for Experiential AI at Northeastern University in Portland, Maine, working on computer vision and AI for healthcare. I completed my M.S. in Data Science at Northeastern University and previously worked as a data engineer and software developer in industry. I like building systems that connect carefully-designed vision models with real-world clinical data.

While working as a data engineer at a pharmaceutical company in China, I visited some of the country’s poorest villages, where “barefoot doctors” without medical degrees cared for thousands of patients in small clinics. We helped set up remote video consultations so specialists in Shanghai could review difficult cases. I still remember patients lining up in cramped rooms while a remote oncologist quietly pointed out a tiny shadow on a scan that turned out to be an early cancer.

Those days were inspiring but also unsettling: every extra consultation meant another late night. That tension is why I care about AI systems that make high-quality care more scalable and reliable, instead of just adding another dashboard for doctors to babysit.

My research lies at the intersection of robust computer vision, multimodal learning, and healthcare. I am especially interested in:

• Robust computer vision under domain shift – understanding how models fail when the distribution changes, and designing methods that transfer across patients, devices, and environments.
• Grounded multimodal and long-form / egocentric video understanding – connecting visual signals with language, context, and temporal structure in real-world videos.
• Reliable and trustworthy ML – uncertainty, evaluation, and failure analysis for models that will be used in high-stakes settings.
• AI for healthcare & computational medicine – including video-based physiological monitoring and infant development, where data are small, noisy, and clinically constrained.

In the long term, I hope to help build computer vision systems that clinicians can trust for remote monitoring and triage, so that high-quality care is accessible even where specialists are scarce.

Recently, I led a project on video-based infant respiration estimation that resulted in a first-author paper at WACV 2026, and I now work on tools to analyze children’s pose and gaze in autism therapy sessions as part of the DREAM project.

Outside of work, I am an occasional INFJ, a long-time fan of strategy and RPG games, and an amateur photographer.

Reach Me

• 📧 Email: song.liy@northeastern.edu
• 😸 GitHub: https://github.com/LiyangSong
• 💼 LinkedIn: https://www.linkedin.com/in/ly-song
• 🏠 This site: https://liyangsong.github.io
• 📄 CV / Resume (updated Jan 2026): resume_liyang_song_012026.pdf

Selected projects & publications

• Overcoming Small Data Limitations in Video-Based Infant Respiration Estimation (WACV 2026)
  We introduce the AIR-400 benchmark and pipeline for estimating infant respiration from heterogeneous clinical RGB/IR videos, including infant ROI detection, multiple optical-flow methods, spatiotemporal models, and PSD-based loss / post-processing.

• DREAM-ASD Toolkit
  Ongoing open-source toolkit for exploring skeleton, head pose, and gaze data from robot-enhanced autism therapy sessions. Provides preprocessing, visualization, and analysis utilities to support research on early ASD markers and engagement.

• Systems & backend projects
  I have also built PaxMesh, a transport-agnostic RPC framework with Paxos, CodeArena, a distributed online judge platform, a distributed key-value store, and several cloud-based ML pipelines using AWS, PostgreSQL, Docker, and message queues.

More details and links are available on the Projects page.

Books

I am currently reading:

• Deep Learning with Python, 2nd Edition
• Designing data-intensive applications: the big ideas behind reliable, scalable, and maintainable systems
• Fluent Python: Clear, Concise, and Effective Programming, 2nd Edition

Read in 2025:

• Spring Start Here: Learn what you need and learn it well
• Code: The Hidden Language of Computer Hardware and Software, 2nd Edition

Read in 2024:

• Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow, 3rd Edition
• Distributed Systems: Concepts and Design, 5th Edition
• Grokking Algorithms, 2nd Edition

On my list:

• Machine Learning Engineering in Action
• JavaScript: The Definitive Guide, 7th Edition
• How Linux Works: What Every Superuser Should Know, 3rd Edition
• Effective Java, 3rd Edition

Photography

I like shooting with my Fujifilm X-T30 II, mostly natural landscapes and everyday life.

So far I’ve taken it to highland snow mountains in northwestern China and to the coasts on both the East and West coasts of the United States. I also use it to record quieter moments with my family and our two cats.

I may start posting small photo selections on this site in the future.

Games

I mostly play strategy and simulation games, and story-driven RPGs.

On the strategy side, I enjoy games like Oxygen Not Included, Civilization, and Stellaris.
What I like most is designing self-sustaining systems—watching a base or a city run almost automatically once all the pieces fit together.

On the RPG side, some of my favorites are The Legend of Zelda and NieR:Automata. I enjoy slow, immersive exploration: wandering through worlds that feel coherent and alive, and uncovering their stories at my own pace.

Movies

I love fantasy, adventure, sci-fi, epic, and road movies.

My all-time favorite is The Lord of the Rings trilogy.

Music

C-Pop, J-Pop, City Pop, OST, light music, and classical.

My playlists keep growing faster than I can organize them.

About this site

This site is built with Hugo framework, styled with Diary theme by Rise.

It serves as my personal homepage and a place to collect notes on research, systems, and things I enjoy.

Last modified on 2025-12-30