In Question 1 what about task X?

Post date: Nov 01, 2011 2:22:0 PM

I don't really want to be drawn into commenting on specific task ideas (for the assignment), however I can give some general advice on the selection of appropriate Machine Learning (ML) approaches.

The key in selecting what ML approach to apply is to identifying the characteristics of the task you're trying to solve and how well they fit against the type of the task that that specific ML approach was designed for.

The lecture slides describes the characteristics of typical Reinforcement Learning (RL) tasks. Of these I think delayed reward is probably the most important. If a task doesn't have any form of delayed reward, i.e. you always know immediately that a decision is bad or good, then that sort of task is possibly better cast as a Supervised Learning (Classification) task, rather than RL.