ByteDance New Grad Machine Learning Engineer Interview Experience — Research Talk, Model Questions and a Fishing Pond Coding Problem

First was a self-introduction, then I introduced my research. He said I could share slides if that would make things clearer. I happened to be preparing my proposal recently, so I shared my slides, and it definitely felt better than just talking through it.

He asked some questions based on my work, normal follow-ups. Interestingly, he also brought up jev and asked whether I follow it.

Then on the technical side he asked whether I had read the reports for different models, what the main improvements were in each generation, whether I understand multimodal, and why LLMs now generally use linear normalization.

The coding question was:

There are N fish ponds, and M[i] is the number of fish you can currently catch per hour in pond i.
You can fish for K hours in total.
Every hour you fish in a pond, that pond's yield for the next hour goes down by 1.
Each hour you can freely choose which pond to fish in.
Find the maximum number of fish you can catch.

N = 2
M = [90, 100]
K = 100
Output: 7075

My first thought was to use a heap, but he required better than O(K*log(N)), so it was only with some hints that I stumbled my way to a sorting + math summation approach. Overall I feel positive about it, although my own performance was just so-so.