Describe the scenario, what "job" will the neural network do in the context of a software application?
- Is this a "classification" problem? What are the labels?
- Is this a "regression" problem?
What approach will be used to train the model?
- Assuming supervised learning, what is the training dataset and how is it labeled?
- If this is a "reinforcement learning"-style problem, how will you "reward" the network. Examples might be a player's score for a game or how long an agent survives in an environment.
Describe, in informal terms, the architecture of the model. What are the inputs and what are the outputs?