William B. Langdon

Fitness Causes Bloat

The problem of evolving an artificial ant to follow the Santa Fe trail is used to study the well known genetic programming feature of growth in solution length. Known variously as “bloat”, “fluff” and increasing “structural complexity”, this is often described in terms of increasing “redundancy” in the code caused by “introns”.

Schema theory for genetic programming with one-point crossover and point mutation

We review the main results obtained in the theory of schemata in genetic programming (GP), emphasizing their strengths and weaknesses. Then we propose a new, simpler definition of the concept of schema for GP, which is closer to the original concept of schema in genetic algorithms (GAs). Along with a new form of crossover, one-point crossover, and point mutation, this concept of schema has been used to derive an improved schema theorem for GP that describes the propagation of schemata from one generation to the next. We discuss this result and show that our schema theorem is the natural counterpart for GP of the schema theorem for GAs, to which it asymptotically converges.

Optimizing Existing Software With Genetic Programming

We show that the genetic improvement of programs (GIP) can scale by evolving increased performance in a widely-used and highly complex 50000 line system. Genetic improvement of software for multiple objective exploration (GISMOE) found code that is 70 times faster (on average) and yet is at least as good functionally. Indeed, it even gives a small semantic gain.

A SIMD interpreter for genetic programming on GPU graphics cards

Mackey-Glass chaotic time series prediction and nuclear protein classification show the feasibility of evaluating genetic programming populations directly on parallel consumer gaming graphics processing units. Using a Linux KDE computer equipped with an nVidia GeForce 8800 GTX graphics processing unit card the C++ SPMD interpretter evolves programs at Giga GP operations per second (895 million GPops). We use the RapidMind general processing on GPU (GPGPU) framework to evaluate an entire population of a quarter of a million individual programs on a non-trivial problem in 4 seconds. An efficient reverse polish notation (RPN) tree based GP is given.

Size Fair and Homologous Tree Crossovers for Tree Genetic Programming

Size fair and homologous crossover genetic operators for tree based genetic programming are described and tested. Both produce considerably reduced increases in program size (i.e., less bloat) and no detrimental effect on GP performance.GP search spaces are partitioned by the ridge in the number of program v. their size and depth. While search efficiency is little effected by initial conditions, these do strongly influence which half of the search space is searched. However a ramped uniform random initialization is described which straddles the ridge.With subtree crossover trees increase about one level per generation leading to subquadratic bloat in program length.

Some Considerations on the Reason for Bloat

A representation-less model for genetic programming is presented. The model is intended to examine the mechanisms that lead to bloat in genetic programming (GP). We discuss two hypotheses (“fitness causes bloat” and “neutral code is protective”) and perform simulations to examine the predictions deduced from these hypotheses. Our observation is that predictions from both hypotheses are realized in the simulated model.

Efficient multi-objective higher order mutation testing with genetic programming

In academic empirical studies, mutation testing has been demonstrated to be a powerful technique for fault finding.However, it remains very expensive and the few valuable traditional mutants that resemble real faults are mixed in with many others that denote unrealistic faults.These twin problems of expense and realism have been a significant barrier to industrial uptake of mutation testing.Genetic programming is used to search the space of complex faults (higher order mutants). The space is much larger than the traditional first order mutation space of simple faults.However, the use of a search based approach makes this scalable, seeking only those mutants that challenge the tester,while the consideration of complex faults addresses the problem of fault realism; it is known that 90% of real faults are complex (i.e. higher order).We show that we are able to find examples that pose challenges totesting in the higher order space that cannot be represented in thefirst order space.

Evolving Problems to Learn About Particle Swarm Optimizers and Other Search Algorithms

We use evolutionary computation (EC) to automatically find problems which demonstrate the strength and weaknesses of modern search heuristics. In particular, we analyze particle swarm optimization (PSO), differential evolution (DE), and covariance matrix adaptation-evolution strategy (CMA-ES). Each evolutionary algorithm is contrasted with the others and with a robust nonstochastic gradient follower (i.e., a hill climber) based on Newton-Raphson. The evolved benchmark problems yield insights into the operation of PSOs, illustrate benefits and drawbacks of different population sizes, velocity limits, and constriction (friction) coefficients. The fitness landscapes made by genetic programming reveal new swarm phenomena, such as deception, thereby explaining how they work and allowing us to devise better extended particle swarm systems. The method could be applied to any type of optimizer.

The GISMOE challenge: constructing the pareto program surface using genetic programming to find better programs (keynote paper)

Optimising programs for non-functional properties such as speed, size, throughput, power consumption and bandwidth can be demanding; pity the poor programmer who is asked to cater for them all at once! We set out an alternate vision for a new kind of software development environment inspired by recent results from Search Based Software Engineering (SBSE). Given an input program that satisfies the functional requirements, the proposed programming environment will automatically generate a set of candidate program implementations, all of which share functionality, but each of which differ in their non-functional trade offs. The software designer navigates this diverse Pareto surface of candidate implementations, gaining insight into the trade offs and selecting solutions for different platforms and environments, thereby stretching beyond the reach of current compiler technologies. Rather than having to focus on the details required to manage complex, inter-related and conflicting, non-functional trade offs, the designer is thus freed to explore, to understand, to control and to decide rather than to construct.

Search based software engineering for software product line engineering: a survey and directions for future work

This paper presents a survey of work on Search Based Software Engineering (SBSE) for Software Product Lines (SPLs). We have attempted to be comprehensive, in the sense that we have sought to include all papers that apply computational search techniques to problems in software product line engineering. Having surveyed the recent explosion in SBSE for SPL research activity, we highlight some directions for future work. We focus on suggestions for the development of recent advances in genetic improvement, showing how these might be exploited by SPL researchers and practitioners: Genetic improvement may grow new products with new functional and non-functional features and graft these into SPLs. It may also merge and parameterise multiple branches to cope with SPL branchmania.