An efficient evolutionary algorithm for solving incrementally structured problems

Abstract

Many real world problems have a structure where small problem instances are embedded within large problem instances, or where solution quality for large problem instances is loosely correlated to that of small problem instances. This structure can be exploited because smaller problem instances typically have smaller search spaces and are cheaper to evaluate. We present an evolutionary algorithm, INCREA, which is designed to incrementally solve a large, noisy, computationally expensive problem by deriving its initial population through recursively running itself on problem instances of smaller sizes. The INCREA algorithm also expands and shrinks its population each generation and cuts off work that doesn't appear to promise a fruitful result. For further efficiency, it addresses noisy solution quality efficiently by focusing on resolving it for small, potentially reusable solutions which have a much lower cost of evaluation. We compare INCREA to a general purpose evolutionary algorithm and find that in most cases INCREA arrives at the same solution in significantly less time.

Autotuning algorithmic choice for input sensitivity

“…In our example, many of the sorting routines (QuickSort, MergeSort, and RadixSort) will recursively call Sort again, thus, the either...or statement will be executed many times dynamically when sorting a single list. The autotuner uses evolutionary search to construct polyalgorithms which make one decision at some calls to the either...or statement, then different decisions in the recursive calls [8].…”

Nitro: A Framework for Adaptive Code Variant Tuning

“…Petabricks [2] supports user specification of transforms that are analogous to functions. Transforms are automatically composed together to form hybrid algorithms using a compiler framework and an adaptive algorithm [33]. Petabricks, however, implicitly tunes variants for the size of the input data set.…”

“…Guo proposes the use of Bayesian Networks to learn the mapping from input features to code variants [9]. Petabricks uses a bottom-up evolutionary algorithm named INCREA [33] which builds a tuned algorithm for a specific problem size by incrementally composing tuned algorithms for smaller problem sizes. Other work in this area includes [46], [47], [48], [49].…”

Continuous learning of compiler heuristics

“…PetaBricks [Ansel et al 2009] is an open-source compiler and programming language developed at MIT that uses machine learning and evolutionary algorithms to autotune [Ansel et al 2011] programs, by making both fine-grained and algorithmic choices. PetaBricks programs work on numerical matrices as their input and output data.…”

“…First of all, the number of tests contained in PetaBricks's benchmark suite is bigger than that of cBench. Furthermore, every program compiled by PetaBricks needs to be tuned by the underlying autotuner that adapts runtime parameters exposed by each program to the system it is running on, using the algorithm described in Ansel et al [2011]. It is a genetic algorithm, and as such it requires the program to be run multiple times to find out the best configuration.…”