Genejector is a project I did for my computer science bachelor thesis. It's all about doing crazy new things with genetic programming in Java. This project is neither active nor maintained, but merely available here for its conceptual value and curious eyes.

The foundation of this project is the field of genetic programming. Genetic programming is an interesting combination of computer science and biology, in which programmers write caretaker software to create and evolve masses of automatically generated programs. Inspired by evolution, these programs are mated and mutated through numerous generations while enduring constant selection pressure by the caretaker to solve a difficult programming problem. This complex process is all put into action in the hope that, at some point, one of the programs will excel at solving the problem. Classic targets of genetic programming include optimization problems, artificial intelligence and novel algorithms. The goal of this project is to explore the possibility of introducing several modern programming concepts to the field, which is otherwise dominated by conventional mathematical methodologies.

I seek to implement a framework to explore the following concepts in relation to genetic programming:

• Context reflection: Genetic programs are traditionally expected to be completely self-contained and only take advantage of custom crafted components. This limitation is problematic for problems which are not easily isolated but are part of a larger codebase and thus very context-dependent. Implementation of a framework capable of reflecting upon the problem context and reuse the codebase could significantly ease the barriers of using genetic programming to solve these problems.

• Data structures: Modern programming has heavy reliance on data structures. However, these are rarely available in genetic programming frameworks unless the genetic programs evolve them on their own. Common data structures like lists, maps and sets could be introduced, and these are often available in the programming language of choice already. Leveraging these could open up an opportunity for shorter and more readable programs while taking advantage of proven and efficient ways of managing data at runtime.

• On-the-fly compilation and injection: Genetic individuals are traditionally evaluated by custom evaluation functions coded into the custom crafted components. By taking advantage of on-the-fly compilation and injection, the individuals can be evaluated in the original problem codebase while relieving the programmer of tedious work.

• Execution sandboxing: To support modern programming features such as data structures and loops, genetic programs need to be able to either directly or indirectly allocate memory as well as execute code that might not terminate. This implies that genetic programs can act maliciously by, among other things, consuming a critical amount of system resources. Execution sandboxing is an interesting solution to the issue that gives the genetic program free rein to consume resources and perform actions while the framework can enforce hard limits on consumption and protect the host system from damage.