Learn Before
Improving an Underperforming Fruit Classifier
Question: Suppose a fruit-ripeness classifier is not accurate enough. Describe three improvement ideas, drawn from data-related or model-related choices, and explain how each one could help.
Sample answer: A team could improve the classifier in several ways. First, it could collect more labeled fruit images so the model has more examples to learn from. Second, it could expand the training set with more varied cases, such as different lighting, camera angles, backgrounds, and fruit varieties, so the model better matches real-world use. Third, it could adjust the model setup by trying a larger network if the model seems too simple, a smaller network or stronger regularization if it is overfitting, or more training time if optimization has not yet converged.
Key points:
- Names at least three valid improvement strategies.
- Explains that more data can reduce generalization error.
- Explains that more varied data can reduce mismatch between training and deployment conditions.
- Explains that changing model capacity, regularization, or training time can address underfitting or overfitting.
Rubric: The response should identify three valid improvement ideas, such as collecting more examples, making the data more diverse, changing model size or architecture, training longer, or adding regularization, and explain how each one improves performance.
0
1
Tags
Machine Learning
Deep Learning
Supervised Learning
Dive into Deep Learning @ D2L
Data Science
Machine Learning Strategy
Machine Learning Yearning @ DeepLearning.AI
Related
A practical improvement for a weak cat detector
True or False: One suggested fix for an underperforming image classifier is to train it for more gradient-descent iterations.
To improve a model that misses rare cases, one option is to collect a more _____ training set, such as examples from unusual conditions.
A practical improvement for an underperforming image classifier
Trying a smaller neural network is one possible improvement for an underperforming message classifier.
To improve a spam filter, one idea is to train the model _____, by taking more optimization steps.
Match each road-sign classifier improvement idea to its description.
Put the steps in a sensible order when a spam filter is not accurate enough and the team wants to improve it.
Which option best describes collecting a more varied training set for a cat classifier?
Regularization is not one of the suggested ways to improve a cat detector that is performing poorly.
A weaker wildlife classifier may improve if you change the neural network _____, which means how its layers and hidden units are arranged.
Match each improvement type to the appropriate idea for a package detector.
Order the steps for improving a road sign detector by expanding the training data.
Improving an Underperforming Fruit Classifier
Improving an Image Classifier for Airport Luggage Checks
Broadening a Bicycle Detector Training Set