Yueran Song
Los Angeles, California
USC Analytics grad student · Applied AI
I’m interested in how models behave when the data gets messy, and how to use them in real decisions. My current projects span wafer defect detection and small language models.
01Work
Manufacturing ML, model training, and generative AI applications
I compared a random forest and a small CNN on WM-811K wafer maps. The random forest’s drop on the test split led me to investigate differences between batches and performance on rare defects.
Next I’ll test how to use model scores to prioritize reinspection. If only 200 of every 1,000 wafers can be checked, how many defects would the ranking catch, and which rare defects would it miss?
I’m building a small language model in PyTorch. I’ve set up the notebook and am working through tokenization, attention, and the training loop.
I’ll train on a small text corpus and inspect validation loss alongside generated samples to see what patterns the model learns while predicting the next token.
I built a text-to-SQL prototype around a delivery incentive case and used validation cases to check whether generated queries followed the business definitions.
A query can run and still answer a different question. The checks look at whether the time window, population, and aggregation match the original request.
02Background
I’m a master’s student in Analytics at USC Viterbi, where I study predictive analytics and data exploration.
As a data analyst intern at a semiconductor equipment company in Shanghai, I used Python and SQL to analyze operational data for recurring reports. At an IoT startup in Chengdu, I built web interfaces and explored next-day electricity forecasting.
M.S. in Analytics
Expected May 2028Bachelor’s in Information Management and Information Systems
