Approximate Novelty Search

Abstract

Width-based search algorithms seek plans by prioritizing states according to a suitably defined measure of novelty, that maps states into a set of novelty categories. Space and time complexity to evaluate state novelty is known to be exponential on the cardinality of the set. We present novel methods to obtain polynomial approximations of novelty and width-based search. First, we approximate novelty computation via random sampling and Bloom filters, reducing the runtime and memory footprint. Second, we approximate the best-first search using an adaptive policy that decides whether to forgo the expansion of nodes in the open list. These two techniques are integrated into existing width-based algorithms, resulting in new planners that perform significantly better than other state-of-the-art planners over benchmarks from the International Planning Competitions.

Novelty and Lifted Helpful Actions in Generalized Planning

“…We proposed a characterization of lifted helpful actions in GP and experimented with different evaluation function combinations using new functions f ha , f ln , and f cn . Other lifted HA extraction methods (Corrêa et al 2021;Wichlacz, Höller, and Hoffmann 2022) and noveltybased search strategies (Lei and Lipovetzky 2021;Singh et al 2021;Corrêa and Seipp 2022) proposed for classical planning could be adopted by research on GP as heuristic search.…”

Using phylogenetic analysis to enhance genetic improvement

“…It does not by itself, however, guarantee good coverage of a fitness landscape [3]. Novelty search also has an inherent limit to how many individuals can be usefully stored before the cost of enumerating them, when evaluating new individuals, slows down the GI process to a point of being non-viable [14]; because the entire genetic material of each individual must usually be stored, there are memory constraints on the total number of points that can be saved. So while useful novelty search is not a substitute for well-tuned GI parameters which guide a process towards high-value areas of a fitness landscape.…”