Case Study

Scaling Decision for Deep Learning with Massive Datasets

Case context: A machine learning team has access to a massive dataset of millions of labeled images. However, when they train their existing small neural network on this dataset, they notice that the network's performance saturates and does not improve despite the large volume of data.

Question: Based on the relationship between computational scale and dataset size described in the text, what should the team diagnose as the primary bottleneck, and what decision should they make regarding their model size and computational resources to resolve this issue?

Sample answer: The team should diagnose that their current neural network is too small to take advantage of the huge dataset. Based on the concept of computational scale, they should decide to increase their model size (train a bigger neural network) and invest in computational scale (more compute resources) to enable training this larger network to exploit the dataset.

Key points:

  • Diagnose that the current small network cannot take advantage of the huge dataset.
  • Increase the size of the neural network (make it big enough).
  • Utilize computational scale to train this larger neural network.

Rubric: The response must correctly diagnose that the current network is too small to utilize the massive dataset, and recommend increasing both model size and computational scale/resources.

0

1

Updated 2026-05-27

Contributors are:

Who are from:

Tags

Machine Learning

Deep Learning

Supervised Learning

Dive into Deep Learning @ D2L

Data Science

Machine Learning Strategy

Machine Learning Yearning @ DeepLearning.AI