State-space search, finding paths in huge, implicitly given graphs, is a fundamental problem in artificial intelligence and other areas of computer science. State-space search algorithms like A*, IDA* and greedy best-first search are major success stories in artificial intelligence, and hundreds of papers based on variations of these algorithms are published every year. Due to this success, the major assumptions of these algorithms are rarely questioned.
We argue that the current generation of state-space search algorithms has three significant deficiencies that impede further progress in the field:
1. They explore a monolithic model of the world rather than applying a factored perspective.
2. They do not learn from mistakes and hence tend do commit the same mistake thousands of times.
3. In the case of satisficing (i.e. suboptimal) search, the design of the major algorithms has been based on ad-hoc intuitions rather than sound theoretical principles.
This proposal targets these three issues. We propose to develop a rigorous theory of factored state-space search, a rigorous theory of learning from information gathered during search, and a
decision-theoretic foundation for satisficing search algorithms. Based on these insights we will design and implement new state-space search algorithms addressing the deficiencies of current methods. Finally, we will apply the new algorithms to application domains of state-space search to raise the state of the art in these areas.
Fields of science
Call for proposal
See other projects for this call