高频题（0）
1. Vision Research and ML Knowledge
Discuss the sizes of classic computer vision datasets and images, efficient methods for downloading and reading large datasets, evaluating experimental results, and common metrics for MLLM benchmarks.
2. Implement and debug a neural network in PyTorch
Using PyTorch, write a simple neural network as specified in a given Colab notebook. Analyze dimension changes, determine an appropriate loss function and implement it without using PyTorch's built-in loss functions. Identify and fix any bugs in the provided partial code. Ensure the final implementation runs successfully.
3. Optimize a multinomial sampler
Write a function to generate a sample distribution that matches a given list of probabilities using a multinomial sampler. Discuss and implement performance optimizations for this program in a real-world scenario, focusing on practical improvements rather than algorithmic optimizations like changing linear search to binary search.
4. Describe multi-GPU DNN training
In the context of multi-GPU deep neural network (DNN) training, explain what is distributed and what is returned during the process. Additionally, discuss how the network traffic changes if the size of the neural network doubles.
