Learn Before
Essay

Analyzing Potential Improvement Actions for an Inaccurate Cat Detector

Question: Assume your cat detector is not accurate enough. Analyze three potential improvement ideas mentioned in the text (either data-related or network-related). Explain how each proposed change aims to improve performance.

Sample answer: To improve an inaccurate cat detector, a team can explore multiple avenues. 1. Collecting more cat pictures increases the overall training data size, helping the model learn generalizable patterns. 2. Collecting a more diverse training set (e.g., containing cats in unusual positions, coloration, or different camera settings) helps the model handle edge cases and perform better under varied conditions. 3. Adjusting neural network capacity, such as trying a bigger neural network (with more layers, hidden units, or parameters) or trying a smaller neural network, allows the team to find the optimal model complexity for the task.

Key points:

  • Identifies at least three valid improvement strategies from the text (e.g., more data, diversity, training time, network capacity, regularization).
  • Explains how data collection strategies (more data or more diverse data) help the model generalize.
  • Explains how modifying the neural network (size/architecture) or training duration impacts performance.

Rubric: The response should identify three valid improvement ideas from the text (e.g., collecting more pictures, collecting a diverse training set, training longer, changing neural network size/architecture, or adding regularization) and explain how they aim to improve the system.

0

1

Updated 2026-05-26

Contributors are:

Who are from:

Tags

Machine Learning

Deep Learning

Supervised Learning

Dive into Deep Learning @ D2L

Data Science

Machine Learning Strategy

Related