Berbalang

Evolutionary Exploration of Emergent Execution

Berbalang is a genetic programming laboratory principally dedicated to the exploration of "weird machines" (or "emergent execution") through evolutionary methods. At its core is a complete reimplementation of ROPER, a system I designed while at Dalhousie University, in 2018, to study the evolution of return-oriented programming (ROP) payloads on the ARM architecture. A substantially overhauled and improved version of ROPER can be found in the berbalang::roper module.

In addition to roper, Berbalang includes a few other genetic programming testbeds as well, including berbalang::linear_gp, a basic linear genetic programming system, which evolves programs for a simple, register-transfer machine, and berbalang::hello_world, a simple genetic algorithm system for evolving strings to match a given target.

To Build

We'll assume that you plan to house Berbalang into ~/src/berbalang. First, clone the repository:

[~/src]$ git clone https://github.com/oblivia-simplex/berbalang
[~/src]$ cd berbalang

On NixOS

[~/src/berbalang]$ nix-shell
...
[nix-shell:~/src/berbalang]$ cargo build --release

Docker

TODO

Gentoo

TODO

Other Linux Distributions

TODO

Running Experiments

First, set up a log directory. We'll assume that you chose ../logs for this purpose. Then create a directory containing experiment specification files, following the example provided in berbalang/config.toml. Let's say you're using ./experiments/foobar/ as your experiment specification directory. Each toml file in that directory should specify the parameters for a series of trial runs.

Then, run, say, 100 trials for under each specification with

[~/src/berbalang]$ ./trials.sh ./experiments/foobar 100 ../logs

The resulting logs will be available under

../logs/berbalang/$JOB/$SELECTION_METHOD/$YEAR/$MONTH/$DAY

Python scripts and tools for analysing experiment data can be found in berbalang/analysis.