Free Amazon MLS-C01 Actual Exam Questions - Question 10 Discussion
During mini-batch training of a neural network for a classification problem, a Data Scientist notices
that training accuracy oscillates What is the MOST likely cause of this issue?
D The oscillation in training accuracy is classic when the learning rate is set too high, causing the model to overshoot the minima during optimization. Even without exact numbers, this is the most common cause compared to batch size or shuffling. Big batches usually stabilize training rather than cause oscillations, and an imbalanced dataset would affect accuracy differently, not typically causing back-and-forth swings in training metrics.
B Disabling shuffling means batches stay similar every epoch, which can cause accuracy to jump instead of smooth out. It’s a common but subtle issue that fits the oscillation clue well.
B tbh, if shuffling is disabled, the model could be seeing very similar or ordered samples in each batch and that might cause accuracy to bounce around instead of settling. It’s often overlooked but can really mess with training stability. This seems like a plausible cause without needing info on learning rate or batch size.
Guessing D since a high learning rate can easily cause accuracy to swing up and down.
Makes sense that D could cause oscillations, but I’d actually go with C here. If the batch size is too big, the model might not be updating weights frequently enough, causing those swings in accuracy between batches. Smaller batches usually smooth out training metrics, so C seems like a solid pick.
It’s B since no shuffling can cause batches to vary a lot, making accuracy jump.
It’s D, but does the question specify the learning rate value?