William Mydlowec

Automatic Creation of Human-Competitive Programs and Controllers by Means of Genetic Programming

Genetic programming is an automatic method for creating a computer program or other complex structure to solve a problem. This paper first reviews various instances where genetic programming has previously produced human-competitive results. It then presents new human-competitive results involving the automatic synthesis of the design of both the parameter values (i.e., tuning) and the topology of controllers for two illustrative problems. Both genetically evolved controllers are better than controllers designed and published by experts in the field of control using the criteria established by the experts. One of these two controllers infringes on a previously issued patent. Other evolved controllers duplicate the functionality of other previously patented controllers. The results in this paper, in conjunction with previous results, reinforce the prediction that genetic programming is on the threshold of routinely producing human-competitive results and that genetic programming can potentially be used as an “invention machine” to produce patentable new inventions.

Reverse engineering of metabolic pathways from observed data using genetic programming.

Recent work has demonstrated that genetic programming is capable of automatically creating complex networks (such as analog electrical circuits and controllers) whose behavior is modeled by linear and non-linear continuous-time differential equations and whose behavior matches prespecified output values. The concentrations of substances participating in networks of chemical reactions are also modeled by non-linear continuous-time differential equations. This paper demonstrates that it is possible to automatically create (reverse engineer) a network of chemical reactions from observed time-domain data. Genetic programming starts with observed time-domain concentrations of input substances and automatically creates both the topology of the network of chemical reactions and the rates of each reaction within the network such that the concentration of the final product of the automatically created network matches the observed time-domain data. Specifically, genetic programming automatically created metabolic pathways involved in the phospholipid cycle and the synthesis and degradation of ketone bodies.

Automatic creation of both the topology and parameters for a robust controller by means of genetic programming

The paper describes a general automated method for synthesizing the design of both the topology and parameter values for controllers. The automated method automatically makes decisions concerning the total number of processing blocks to be employed in the controller, the type of each block, the topological interconnections between the blocks, the values of all parameters for the blocks, and the existence, if any, of internal feedback between the blocks of the overall controller. Incorporation of time-domain, frequency-domain, and other constraints on the control or state variables (often analytically intractable using conventional methods) can be readily accommodated. The automatic method described in the paper (genetic programming) is applied to a problem of synthesizing the design of a robust controller for a plant with a second-order lag. A textbook PID compensator preceded by a lowpass pre-filter delivers credible performance on this problem. However, the automatically created controller employs a second derivative processing block (in addition to proportional, integrative, and derivative blocks and a pre-filter). It is better than twice as effective as the textbook controller as measured by the integral of the time-weighted absolute error, has only two-thirds of the rise time in response to the reference (command) input, and is 10 times better in terms of suppressing the effects of disturbance at the plant input.

Evolution of a Controller with a Free Variable Using Genetic Programming

A mathematical formula containing one or more free variables is “general” in the sense that it provides a solution to an entire category of problems. For example, the familiar formula for solving a quadratic equation contains free variables representing the equation’s coefficients. Previous work has demonstrated that genetic programming can automatically synthesize the design for a controller consisting of a topological arrangement of signal processing blocks (such as integrators, differentiators, leads, lags, gains, adders, inverters, and multipliers), where each block is further specified (“tuned”) by a numerical component value, and where the evolved controller satisfies user-specified requirements. The question arises as to whether it is possible to use genetic programming to automatically create a “generalized” controller for an entire category of such controller design problems — instead of a single instance of the problem. This paper shows, for an illustrative problem, how genetic programming can be used to create the design for ‘both the topology and tuning of controller, where the controller contains a free variable.

Automatic Computational Discovery of Chemical Reaction Networks Using Genetic Programming

The concentrations of substances participating in networks of chemical reactions are often modeled by non-linear continuous-time differential equations. Recent work has demonstrated that genetic programming is capable of automatically creating complex networks (such as analog electrical circuits and controllers) whose behavior is modeled by linear and non-linear continuous-time differential equations and whose behavior matches prespecified output values. This chapter demonstrates that it is possible to automatically induce (reverse engineer) a network of chemical reactions from observed time-domain data. Genetic programming starts with observed time-domain concentrations of substances and automatically creates both the topology of the network of chemical reactions and the rates of each reaction of a network such that the behavior of the automatically created network matches the observed time-domain data. Specifically, genetic programming automatically created a network of four chemical reactions that consume glycerol and fatty acid as input, use ATP as a cofactor, and produce diacyl-glycerol as the final product. The network was created from 270 data points. The topology and sizing of the entire network was automatically created using the time-domain concentration values of diacyl-glycerol (the final product). The automatically created network contains three key topological features, including an internal feedback loop, a bifurcation point where one substance is distributed to two different reactions, and an accumulation point where one substance is accumulated from two sources.

Automatic Synthesis, Placement, and Routing of an Amplifier Circuit by Means of Genetic Programming

The complete design of a circuit typically includes the tasks of creating the circuits placement and routing as well as creating its topology and component sizing. Design engineers perform these four tasks sequentially. Each of these four tasks is, by itself, either vexatious or computationally intractable. This paper describes an automatic approach in which genetic programming starts with a high-level statement of the requirements for the desired circuit and simultaneously creates the circuits topology, component sizing, placement, and routing as part of a single integrated design process. The approach is illustrated using the problem of designing a 60 decibel amplifier. The fitness measure considers the gain, bias, and distortion of the candidate circuit as well as the area occupied by the circuit after the automatic placement and routing.

Automatic Synthesis of Both the Topology and Parameters for a Controller for a Three-Lag Plant with a Five-Second Delay Using Genetic Programming

This paper describes how the process of synthesizing the design of both the topology and the numerical parameter values (tuning) for a controller can be automated by using genetic programming. Genetic programming can be used to automatically make the decisions concerning the total number of signal processing blocks to be employed in a controller, the type of each block, the topological interconnections between the blocks, and the values of all parameters for all blocks requiring parameters. In synthesizing the design of controllers, genetic programming can simultaneously optimize prespecified performance metrics (such as minimizing the time required to bring the plant output to the desired value), satisfy time-domain constraints (such as overshoot and disturbance rejection), and satisfy frequency domain constraints. Evolutionary methods have the advantage of not being encumbered by preconceptions that limit its search to well-traveled paths. Genetic programming is applied to an illustrative problem involving the design of a controller for a three-lag plant with a significant (five-second) time delay in the external feedback from the plant to the controller. A delay in the feedback makes the design of an effective controller especially difficult.

Use of conditional developmental operators and free variables in automatically synthesizing generalized circuits using genetic programming

This paper demonstrates that generic programming can be used to create a circuit-constructing computer program that contains both conditional operations and inputs (free variables). The conditional operations and free variables enable a single genetically evolved program to yield functionally and topologically different electrical circuits. The conditional operations can trigger the execution of alternative sequences of steps based on the particular values of the free variables. The particular values of the free variables can also determine the component value of the circuits components. Thus, a single evolved computer program can represent the solution to many instances of a problem. This principle is illustrated by evolving a single computer program that yields a lowpass or a highpass filter whose passband and stopband boundaries depend on the programs inputs.