What are some examples of sequential-decision tasks?

Question

I ask this because I'm currently learning about Neural Networks as a subset of the machine learning algorithms

Just trying to get some intuition on what sort of problems out there are categorized as "sequential" while others "non-sequential"

Thanks :)

Answer 1

One kind of sequential problem is making control decisions in real time. Consider the problem of driving a vehicle. The system will have a state, you can observe some measurements, and you must make a decision. You decision generates and input which then affect the future state of the vehicle and the whole process repeats. A problem which interacts with an external entity that changes its behavior based on a decision is sequential in nature. Other examples include things like optimally managing a stock portfolio, searching for an evading opponent, optimally playing a card game.

One non-sequential problem is classifying the fruit on a scanner in a grocery store. A measurement is taken, the data is processed and and output is determined. When the output generated does not alter the nature of of future outputs, it is non-sequential.

Answer 2

The most obvious example in the field of neural networks:

• image classification is a non-sequential task
• neural machine translation (and most other NLP problems) are sequential tasks, because the translation of a given word in a sentence does not only depend on the word itself, but also on the translation of the previous word (e.g. whether arm was interpreted as limb or as weapon).