CU
Columbia University
M.S. in Computer Engineering
Selected Coursework
Deep Learning for Computer VisionComputer ArchitectureAI-Native ComputingAI of Things
Computer Engineering · Machine Learning
Yishan
Song
M.S. in Computer Engineering
Columbia University
I’m interested in machine learning, deep learning, computer vision, AI systems, and computer architecture — especially the space where mathematical ideas become practical intelligent systems.
Columbia University · Expected 2027
Yishan SongComputer Engineering · Columbia
About
Math, systems,
and intelligent machines.
An interdisciplinary path from applied mathematics and statistics to computer engineering, with a focus on understanding both models and the systems that make them useful.
I’m a Computer Engineering master’s student at Columbia University with an undergraduate background in Applied Mathematics and Statistics from UC Davis.
My academic interests span machine learning, deep learning, computer vision, AI systems, and computer architecture. I enjoy moving between theory, modeling, and implementation — and turning ideas into practical experiments and tools.
I like building at the intersection of technical depth and clear, thoughtful design.
Current Research Interests
I’m especially interested in efficient AI systems and heterogeneous neural–symbolic computing — including how intermediate representations, data movement, and producer–consumer overlap can shape end-to-end performance.
Education
Academic path.
UCD
University of California, Davis
B.S. in Applied Mathematics · B.S. in Statistics
Machine Learning Focus
Machine Learning Focus
Selected Coursework
Statistical LearningOptimizationProbability TheoryMathematical StatisticsStatistical InferenceMultivariate Data AnalysisTime Series AnalysisNumerical AnalysisLinear & Applied Algebra
Selected Work
Research, competitions,
and ML studies.
A selection of projects from research collaboration, machine-learning competitions, and undergraduate coursework across optimization, clustering, density estimation, and time-series modeling.
Research
Cell Segmentation with Deep Learning
Developed and evaluated deep-learning-based cell segmentation workflows using public Cellpose datasets and fluorescent microscopy images, with reproducible preprocessing, tuning, visualization, and evaluation pipelines.
Kaggle · Top 20%
DRW Crypto Market Prediction
Led an ensemble-learning pipeline for minute-level price-movement prediction, combining time-slicing and decay weighting with XGBoost, LightGBM, and CatBoost to address temporal drift.
Kaggle · Top 15%
Spacecraft Telemetry Anomaly Detection
Built anomaly-detection models for spacecraft telemetry with ensemble trees and neural networks, using temporal visualizations and feature attribution to interpret anomalous behavior.
Coursework
Selected ML & Statistical Studies
A broader set of undergraduate studies spanning optimization, unsupervised learning, density estimation, and statistical forecasting.
Sophia Optimizer StudyReplicated and analyzed the Sophia stochastic second-order optimizer, compared its design and reported behavior with AdamW and Lion, and wrote a 25-page technical report.
USPS Handwritten Digit ClusteringApplied PCA and K-Means to 9,298 handwritten digits; used Hungarian matching to reach 74.1% cluster-to-label accuracy at 11 principal components.
Clustering & Density EstimationImplemented K-Means and K-Medoids from scratch, explored PCA image compression and KDE bias–variance behavior, and compared DBSCAN with Mean Shift.
Time-Series ForecastingModeled Central African Republic export data from 1960–2017, selected an ARIMA(2,1,3) model using AIC, and produced 10-year forecasts with uncertainty intervals in R.
Skills
Tools I work with.
Programming & Data
PythonRSQLMATLABNumPyPandas
Machine Learning
PyTorchTensorFlowscikit-learnXGBoostLightGBMCatBoostPCA / SVDClustering
Focus Areas
Machine LearningDeep LearningComputer VisionAI SystemsComputer Architecture
Creative Lab
Things I build
outside the classroom.
Small apps, games, and media projects where I can experiment with interaction, visual design, and storytelling.
铸金录
JX3 Gear Ledger
A cross-platform gear and spending tracker for JX3, designed to organize equipment upgrades, enhancement costs, character progression, and market-price comparisons.
CompletedRelease pendingCross-platform
小人退散
Shoo, Villain!
A playful Chinese-style stress-relief mobile game centered on a customizable straw doll, tactile interactions, sound effects, and themed visual design.
CompletedRelease pendingMobile Game
珠落潮生
Pearlfall: Heart Tide
A mobile-first endless arcade game about two mermaids: catch pearls, dodge hazards, maintain combos, and repeatedly trigger Heart Tide for a burst of invincibility and doubled scoring.
In ProgressCasual GameMobile-first
Media & Storytelling
Educational videos, visual storytelling, and student media.
A separate portfolio for video production, short-form content, motion graphics, visual design, and educational storytelling.
View Media Portfolio