Specialist AI Testing Ct AI · Free Practice Question Medium
Question 6
6.1 A technology company is developing an AI-based camera system to identify different types of objects in real-time video feeds. The system is built using a deep neural network with multiple layers. During the training phase, the output layer of the network indicates the likelihood that a specific object, such as a cat, is present in the input image. After extensive training, the network consistently misclassifies dogs as cats. Which of the following adjustments should the company consider to address this issue, and why?
-
A
Increase the number of neurons in the hidden layers to provide the network with more computational capacity, potentially improving its ability to distinguish between similar objects.
-
B
Adjust the bias values of the neurons to ensure that the activation functions are centered more accurately, which could help in refining the classification process.
-
C
Change the activation function used in the neurons to a different function that might better capture the differences between cats and dogs.
-
D
Increase the amount of labeled training data, particularly more examples of both cats and dogs, to improve the network's ability to learn the differences between these two classes.
Reveal correct answer
Correct answer: D
Explanation
The misclassification issue indicates that the neural network might not have enough diverse training data to accurately distinguish between cats and dogs. Increasing the amount of labeled training data (Option D) can help the network learn the subtle differences between these classes, leading to better performance. Option A, increasing the number of neurons in the hidden layers, may improve the network's capacity, but without sufficient training data, this alone won't solve the misclassification problem. Option B, adjusting the bias values, is a less direct approach and not likely to address the core issue of data insufficiency. Option C, changing the activation function, might have some impact, but the primary issue seems to be with the training data quality and quantity.
Discussion
Think the marked answer is wrong, or have a better explanation? Share it below — comments appear after review.
