I've been working on applying some ML approaches to a particular problem, and it's repeatedly led me to into an interesting topic related to the process of 'directed search'. In particular, I've been pondering these questions:
* What are the 'essential' characteristics and requirements of a successful directed search? (Particularly when dealing with extremely large solution spaces.)
* How can we quantify the efficiency of a directed search, estimate the resources required, and thereby design a successful one (and avoid futile ones)?
These do not appear to be profound questions. But to the extent I can find this topic addressed, it seems to be in one of two extreme forms.
In the first form, the topic and these related questions are considered to be fairly obvious and even straightforward and so the topic is just given brief mention, almost in passing. At the other extreme is published research focused on a particular or new optimized search technique, where the topic is a secondary, indirect concern, and so the questions above are taken for granted and not treated as worthy topics in and of themselves.
Probably I am not using the accepted jargon and so am just not finding or recognizing the relevant published material. If anyone can point me toward any resources around this specific topic, I'd greatly appreciate it.
The text search literature addresses this. A text search engine applies a scoring function to every document in the index, and chooses the document with the highest score. Text search indexes are all about pruning low-scoring documents early.
My article 4 Easy Steps to Structure Highly Unstructured Big Data, via Automat... might be of some help.