Scale AI Machine Learning Research Internship Interview Experience

Role: Machine Learning Research Intern
Location: San Francisco
Result: Rejected

Applied for an NLP internship in November, but only Computer Vision internship roles were available. A recruiter reached out the next day.

The second round was a take-home Computer Vision assignment. The task was to build a CNN with fewer than two million trainable parameters. The model had to predict an object's width, height, center coordinates, and rotation angle while maximizing bounding-box IoU.

Several architectures and custom loss functions were tested. Results were written up in a PDF and submitted. The take-home passed.

The next round was a 45-minute technical phone screen.

Ten random values between 0 and 1 had to be generated and sorted. The gap between the fourth and fifth values was recorded over many trials. The same process was repeated for the fifth and sixth values, then both distributions were visualized and compared.

Increasing the number of trials was suggested for a clearer visual comparison. Numerical comparison methods such as summary statistics and statistical tests were also discussed. The distributions were right-skewed because the values were sorted before taking adjacent gaps.

The FastPrep version of this exercise is available here: Adjacent Order-Statistic Gaps.

The rest of the phone screen was a deep dive into previous ML internship work. The process ended after this round, before the planned onsite.