Debug and Tune an MNIST Image Classifier
Given a buggy NumPy-based neural network notebook for MNIST digit classification, identify and fix issues such as softmax overflow and incorrect gradient backpropagation using the chain rule. Once debugged, tune hyperparameters to achieve greater than 80% classification accuracy.
Asked at:
Datology
Question Timeline
See when this question was last asked and where, including any notes left by other candidates.
Late July, 2026
Datology
Senior
It is a ML coding problem. A notebook is given to debug an ML classifier for MNIST digits. Common bugs are softmax overflow, buggy gradient backprop. It has to be implemented in numpy, so you need to know chain rule very well and be able to derive the gradients at each step. Finally, tune hyperparameters to achieve >80% accuracy.
Hello Interview Premium
Your account is free and you can post anonymously if you choose.