Michael Sun

Major Experiences

For some of these, see Project reflections section for more details. Most code repos are public. For specific dates, see resume or LinkedIn. Enjoy!

Research Assistant @ MIT CSAIL

• Supervised by Wojciech Matusik
• I work on multiple projects within the MIT-IBM Watson AI lab (PI: Jie Chen), working with client companies on graph-based design and generative AI for scientific discovery in future industries
• I also collaborate with domain experts from other MIT departments to simultaneously develop new ML methods and close the loop experimentally
• I actively think about what’s the most compact way to describe the design space of graph-structured objects, and how can large language models aid in the loop of scientific discovery

Research Intern @ IBM Watson-AI Lab

• Supervised by Jie Chen
• Multi-modal Large Language Models for molecular inverse design and planning
• Multi-modal Foundation Models for learning domain-specific graph languages and expert-level molecular design
• Applications in design of circuits, molecules, graphical models, neural architectures

Machine Learning Engineer @ TikTok

• Onboarded the team powering the search engine for >10B videos and >1B users on TikTok
• Got bored, realized what I really wanted, and left

Research @ Stanford Artificial Intelligence Laboratory (SAIL)

• Supervised by Ananya Kumar and Percy Liang
• Devise SOTA representation learning approaches to continual learning settings

Research @ SLAC National Accelerator Laboratory (feat. SNAP)

• Supervised by Tailin Wu and Jure Leskovec
• Accelerating large-scale laser particle acceleration simulations with deep learning
• Devised approach can model complex physical phenomena at multiple scales

Machine Learning Engineer Intern @ Spotify

• Supervised by Brian Martin and Funmi Doro
• One of 3 inaugural interns in Spotify’s ML Platform team
• Began feature in ML Home, Spotify’s productivity hub for all applied ML teams in the company, on Bayesian optimization over pipeline parameters
• Proposed pipeline to identify causes for underperforming user segments, using Kubeflow Pipelines for orchestrating end-to-end ML workflows

Research @ Stanford Network Analysis Project (SNAP)

• Supervised by Antoine Bosselut and Jure Leskovec
• Surpassed SOTA results on the “cold start” problem in recommendation systems via a hybrid matrix factorization and “content basket” approach
• Introduced a novel deep autoencoder architecture DeepNaniNet using jointly learned graph and language encoders for reconstructing user-item preferences
• Compiled largest anime recommendation dataset featuring 10,000 shows, 13,000+ users, 130,000 reviews, etc. by crawling online discussion forums (data)
• Founded and maintaining only existing NLP driven anime recommendation site OtakuRoll built on DeepNaniNet!
• Published in Springer‘s special issue on recommender systems

Machine Learning Intern @ Samsung Research America Think Tank Team (deck)

• Supervised by Curt Aumiller and Kate Hajash
• Trained the SARAM robot to stir kitchen pots in a goal-based OpenAI gym environment
• Replicated two OpenAI papers on dexterous hand solving a Rubik’s cube and ran hundreds of experiments with policy gradient actor-critic algorithms
• Developed a new exploration-based environment for more human-like movements and presented successful PoC to team including SRA VP Joon Lee
• Worked on practical methods to making RL policies transfer sim2real

Research @ Quantitative Imaging and AI Laboratory

• Supervised by Daniel Rubin
• Applying techniques of multi-modal representation learning on chest X-ray images, following this dataset with ideas from this paper (some initial code)
• Training object detection models on Stanford’s largest video/EEG dataset for seizure detection with SOTA video understanding technologies, like this unsupervised deep tracker to track patients and this for video event composition

Software Engineer – Machine Learning Intern @ Synaptics (demo/code/deck)

• Supervised by Utkarsh Guar and Gaurav Arora
• Replicated the SOTA paper (SSD) in object detection
• Engineered datasets and data pipeline for the task of logo detection
• Developed a high-recall end-to-end realtime logo detection pipeline for smart TV
• Experimented and tested deployment on Synaptics VSR 371 SoC
• Presented findings to company AI/CV team and CTO

Co-Founder, CTO @ Demodraft (site/demo/code/deck) – Led a team of part-time volunteers to deploy our Beta within 1.5 months, got accepted into Berkeley Sky Deck, one of nation’s top college accelerators with <10% acceptance rate [update: exited by merging with OTP! see legacy landing page]

Quizkly (Live app/demo/code)– Full-stack React + Django web app with end-to-end deep learning pipeline that auto-generates quizzes from any corpus of text [update: service no longer sponsored or maintained]

aiFood (App Store/demo/code) – iOS native app that automates macro-counting and meal preparation with custom made API (code)

aiRoute (App Store/demo/code) – iOS native app that randomly generates a running route in your area and provides turn-by-turn map and voice navigation [update: some SDKs since 2018 have changed and may not work on new devices/OS anymore]

Publications/preprints/presentations

(Selected works are visualized)

Post-Hoc Regression Refinement via Pairwise Rankings
Neural Information Processing System (NeurIPS) 2025 [preprint]
Kevin T. Wijaya, Michael Sun, Minghao Guo, Hans-peter Seidel, Wojciech Matusik, Vahid Babaei

Foundation Molecular Grammar: Multi-Modal Foundation Models Induce Interpretable Molecular Graph Languages
International Conference of Machine Learning (ICML) 2025 [link, preprint, code]
Michael Sun, Gang Liu, Weize Yuan, Wojciech Matusik, Jie Chen

Directed Graph Grammars for Sequence-based Learning
International Conference of Machine Learning (ICML) 2025 [link, preprint, code]
Michael Sun, Orion Foo, Gang Liu, Wojciech Matusik, Jie Chen

Multimodal Large Language Models for Inverse Molecular Design with Retrosynthetic Planning
International Conference on Learning Representations (ICLR) 2025 [link, code, preprint, press]
Gang Liu, Michael Sun, Wojciech Matusik, Meng Jiang, Jie Chen

Procedural Synthesis of Synthesizable Molecules
International Conference on Learning Representations (ICLR) 2025 [link, code, preprint]
Michael Sun, Alston Lo, Minghao Guo, Jie Chen, Wojciech Matusik, Connor W. Coley

Representing Molecules as Random Walks Over Interpretable Grammars
International Conference of Machine Learning (ICML) 2024 (Spotlight) [link, code, preprint, video]
Michael Sun, …, Zachary P Smith, Jie Chen, Wojciech Matusik

PP-GNN: Pretraining Position-aware Graph Neural Networks with the NP-hard metric dimension problem
Neurocomputing 2023 [link, code]
Michael Sun

X-RiSAWOZ: High-Quality End-to-End Multilingual Dialogue Datasets and Few-shot Agents
Association for Computational Linguistics (ACL) 2023 [link, code]
Mehrad Moradshahi, …, Michael Sun, …, Monica Lam

Privacy Preserving Inference of Personalized Content for Out of Matrix Users
Springer Journal of Data Science 2023 [link]
Michael Sun, Andrew Wang

---

Protein Structure Tokenization via Geometric Byte Pair Encoding
under review [code, preprint]
Michael Sun, Weize Yuan, Gang Liu, Wojciech Matusik, Marinka Zitnik

AI-driven Discovery and Experimental Validation of Polymer Structures for Gas Separation Membranes
under review [code]
Michael Sun*, Benjamin Pedretti*, Aristotle Grosz, Minghao Guo, Zachary Smith, Wojciech Matusik

A Domain-Specific Language for Programmable Chemistry with LLMs
under review
Alston Lo, Benjamin Tod Jones, Michael Sun, Armando Solar-Lezama, Wojciech Matusik

---

Autonomous Agents for Scientific Discovery: Orchestrating Scientists, Language, Code, and Physics [preprint]
Lianhao Zhou, …, Michael Sun, …, Marinka Zitnik, Shuiwang Ji

Two-Stage Pretraining for Molecular Property Prediction in the Wild
GenBio @ International Conference of Machine Learning (ICML) 2025 [preprint]
Kevin T. Wijaya, Minghao Guo, Michael Sun, Hans-peter Seidel, Wojciech Matusik, Vahid Babaei

Improving Representational Continuity with Supervised Continued Pretraining
CLVISION @ Conference on Computer Vision and Pattern Recognition (CVPR) 2023 [link]
Michael Sun, Ananya Kumar, Divyam Madaan, Percy Liang

Learning Efficient Hybrid Particle-continuum Representations of Non-equilibrium N-body Systems
AI4Science @ Neural Information Processing System (NeurIPS) 2022 [poster, preprint]
Tailin Wu, Michael Sun, …[5 authors]…, Zhitao Ying, E. Paulo Alves, Jure Leskovec, Frederico Fiuza

---

(Dataset) User-Item Feature Graph for Content Based Recommendations of Japanese Anime Shows
Harvard Dataverse [DOI]
Michael Sun

(Industry) Teaching the Samsung Bot Chef to Stir in a Open-Ended Environment
Samsung Research America Think Tank Team Internship (Presented at Intern Showcase) [presentation]
Michael Sun

(Industry) Logo Detection with VGG/MobileNet SSD
Synaptics Internship [presentation]
Michael Sun

(Misc.) Bounds on metric dimension for families of planar graphs
International Science and Engineering Fair (ISEF) (2017 2nd Place Grand Award, Category of Mathematics) [abstract, preprint]
Carl Joshua Quines, Michael Sun

(Misc.) Investigating Effect of Dialogue History in Multilingual Task Oriented Dialogue Systems [preprint]
Stanford CS224V Final Project
Michael Sun, Kaili Huang, Mehrad Moradshahi

(Misc.) Alpha-Mini: Minichess Agent with Deep Reinforcement Learning [preprint]
Stanford CS238 Final Project (Perfect Score)
Michael Sun, Robert Tan

(Misc.) Do Neural Networks Generalize from Self-Averaging Sub-classifiers in the Same Way As Adaptive Boosting? [preprint]
Stanford CS154 Final Project
Michael Sun, Peter Chatain

(Misc.) Almost-Nash Sequential Bargaining [preprint]
Stanford CS261 Final Project
Gokul Dharan, Hunter Guru, Michael Sun

(Misc.) How Fourier Transform Can Speed Up Training CNNs
Towards Data Science [link]
Michael Sun

(Misc.) LightGCN for MovieLens-100K
Medium [link]
Michael Sun, Chuanxiu Xiong, Sidharth Goel

Personal reflections

1. Stanford MS RA
2. TikTok
3. OtakuRoll (NLP Rec Sys)
4. Demodraft
5. Samsung Research
6. Stanford QIAI
7. Synaptics
8. aiFood + MealApp
9. aiRoute
10. Quizkly
11. KnowledgeTree

Code Samples

PixelCNN (GitHub) – Experimentation/evaluation of better ways to project caption embeddings for the PixelCNN conditional generative model (report/poster)

LightGCN for MovieLens-100K – Tutorial and Colab walkthrough for movie recommendations using GNN (final draft/code)

KnowledgeTree (GitHub) – Console C++ and Python application that builds a high-to-low level concept tree from scraping Wikipedia pages with NLP techniques (poster)

GoodNews (GitHub)  – Machine learning project that predicts news article virality and popularity using its content and metadata (report/poster/code)

aiFriend (GitHub) – Website that generates reports (in R Markdown) analyzing and visualizing Messenger conversations (code)

Monetic (GitHub) – Funding platform to help struggling individuals during COVID-19 raise funds by sharing linked TikToks (devpost)

YangNoYang (GitHub) – Inspired to Silicon Valley’s “Hot Dog”, uses FaceID to detect faces of Andrew Yang in pictures (site)

InstaBot (GitHub) – Bot that can auto-login, auto-click posts, send follow requests, and auto grow following on Instagram (devpost)

OtakuRoll (GitHub) – Implementation of AnimeRecSys (see below), uses variety of algorithms to generate recommendations based on past shows enjoyed (prototype)

Education history

I am a US citizen. I was born in LA, California. My family moved and I studied overseas from elementary to high school, as an expat.

Here are some selected highlights from my education.

During high school (2014-2018):

• Second Place Grand Award in Mathematics Category at Intel International Science and Engineering Fair (ISEF) 2017
  • Project was a conjecture and proof of new upper bounds on the metric dimension for planar graphs with applications to GPS-less navigation systems
• Ranked top ~100 2017 United States of America Mathematical Olympiad (USAMO), one of only three from China and first qualifier in my high school’s history
• Co-authored ISEF-winning paper in combinatorial graph theory on the Metric Dimension for Planar Graphs (arxiv)
• Distinguished Honor Roll (top 1%) on both AMC 12 and AMC 10
• Top 30 in the national American Invitational Mathematics Examination (AIME)
• Ross Mathematics Camp at Ohio State University 2017 (one of USA’s oldest, most prestigious pure math summer programs)
• Founded school’s only math club, grew to 20+ members as president, lecturer, problem writer, organizer and promoter, now it has 100+ members
• Started a legendary blog to document progress in math competitions, becoming one of the most popular blogs on AoPS, the biggest online forum for math problem solvers around the world (fun fact: where I met my ISEF partner)

In college (2018-2022):

• Pear VC Garage Fellow program (one of three freshmen in cohort of two dozen out of hundreds of Stanford applicants)
• Published two iOS apps on App Store
• Quizkly, ML MCQ quiz generator (funded by Pear VC)
• Forum moderator in China-US student conferences (Stanford FACES)
• Effective Altruism fellowships (Stanford EA)
• RA at the QIAI Lab (PI: Daniel Rubin) on deep learning in medical applications
• RA in the SNAP Group (PI: Jure Leskovec) on deep NLP recommender systems and knowledge-augmented language models
• Summer associate in Stanford National Accelerator Laboratory on neural n-body simulations (SLAC) (PI: Frederico Fiuza)
• RA in p-Lambda Group (PI: Percy Liang) on continual representation learning
• Major Student Advisor senior year
• Grad RA in US-Asia Tech Center studying high tech startups in Asia

I graduated from Stanford University in June 2022 with BS (Honors) in MCS and MS in Computer Science.

During PhD:

• Spotlight at ICML 2024
• Intern at IBM Watson-AI Lab 2024 summer, working on LLMs for graphs
• 2025 D.E. Shaw Research Fellowship – 1 of 20 selected

Relevant coursework:

• Computer Science
  • CS 103 Mathematical Foundations for Computing, CS 106L Standard C++ Programming Laboratory, CS 106X Programming Abstractions Accelerated, CS 107 Computer Organization & Systems, CS 110 Principles of Computer Systems, CS 145 Data Management and Data Systems, CS 154 Introduction to the Theory of Computation, CS 161 Design and Analysis of Algorithms, CS 199 Independent Research (2x), CS 221 Artificial Intelligence: Principles and Techniques, CS 224N Natural Language Processing with Deep Learning, CS 224V Conversational Virtual Assistants with Deep Learning, CS 224W Machine Learning with Graphs, CS 246 Mining Massive Data Sets, CS 229 Machine Learning, CS 236 Deep Generative Models, CS 238 Decision Making Under Uncertainty, CS 261 Optimization and Algorithmic Paradigm, CS 330 Deep Multi-task and Meta Learning (audited), CS 47 Cross Platform Mobile Development (audited)
• Mathematics + Statistics
  • MATH 104 Applied Matrix Theory, MATH 171 Fundamental Concepts of Analysis, MATH 61DM, 62DM, 63DM Modern Mathematics: Discrete Methods
  • STATS 116 Theory of Probability, STATS 200 Introduction to Statistical Inference
• Others
  • MS&E 111X Introduction to Optimization (Accelerated), MS&E 221 Stochastic Modeling, PHIL 150 Mathematical Logic
• deeplearning.ai Deep Learning Specialization [certificate]
• Alberta Machine Intelligence Institute Reinforcement Learning Specialization

Alumni programs:

• AwesomeMath Summer Program twice – an Olympiad-prep math camp
• Worldwide Online Olympiad Training – high school online Olympiad-prep math camp
• Ross Mathematics Camp 2017 – one of US’s oldest/most prestigious math enrichment camps, guiding students on proof-based exploration from classical to modern number theory results
• Summer Program on Applied Rationality and Cognition 2018 – a highly selective program of nationwide’s top STEM talent
• Make School Summer Academy 2018 – program with curriculum in iOS app dev and design thinking in which students ship apps by the end
• Multiple Stanford Effective Altruism Fellowships (2019)

Mentoring Resources (more to come)

• My advice on LeetCode and problem collection grouped by category
• My Machine Learning (Stanford CS229) flashcards and some interview prep for ML engineering positions
• Collaborative math blog back from HS
• Research paper collection

---

Contact

---

Hobbies: gym, chess, anime, hiking, and NBA.
Whenever I can, I go exploring Japan.