Google Professional Machine Learning Engineer
Get started today
Ultimate access to all questions.
Ultimate access to all questions.
You are a Machine Learning Engineer working on an image segmentation model for a self-driving car application. After deploying the first version of the model, you observe a significant decline in the area under the curve (AUC) metric. Upon further investigation through video recordings, you notice that the model's performance is suboptimal in scenarios with highly congested traffic, whereas it performs as expected in less congested conditions. Considering the need for the model to perform reliably across all traffic conditions to ensure safety and compliance with regulatory standards, what is the most probable cause for this outcome? Choose the best option.
Explanation:
The most plausible explanation for the observed decrease in AUC and the model's underperformance in highly congested traffic is that the model is overfitting to less congested scenarios, compromising its ability to generalize to more congested conditions. Overfitting indicates that the model has learned the training data too well, including its noise and outliers, leading to inadequate performance in new, more challenging scenarios. The vanishing gradient problem (Option D) could also contribute to the model's inability to learn effectively from highly congested scenarios, especially in deep networks, making E a plausible option. However, the primary issue is overfitting (Option A), as indicated by the model's differential performance across traffic conditions. Incorrect options include B, which misrepresents the data imbalance issue, and C, which incorrectly dismisses AUC as a valid metric for this context.