Home_eac3

You are on EACirc 3 wiki, for current version, go to home site.

EACirc is a framework for automatic problem solving. It uses supervised learning techniques based on evolutionary algorithms to construct and optimize software circuits in order to solve the given problem.

Problems are solved by the means of hardware-like circuits - small, software-emulated circuits consisting of gates and interconnecting wires transforming input data into desired output data. The layout of these circuits is designed randomly at first. They are subsequently optimized in the process of supervised learning (inputs are provided alongside with correct outputs) until the the desired success rate is achieved.

The learning stage incorporates genetic programming principles:

• a handful of these circuits (circuit 'population') is considered simultaneously;
• each individual circuit is evaluated on the data and its 'fitness' is determined by comparison of its outputs with the expected outputs;
• individuals with low 'fitness' are deleted (survival of the fittest);
• individuals with high 'fitness' are altered ('sexiual crossover' and a small chance of 'mutation');
• the process starts over with this new 'generation' of circuits.

EACirc Framework

The EACirc framework consists of main application and several supporting tools and scripts. The modular design allows for easy addition of new problem modules ('projects') and output interpretation modules ('evaluators'). Currently, the project has following main parts:

• EACirc - the main application, constructs circuits using evolutionary principles.
• utils - set of scripts and small programs used for results processing.

EACirc framework user documentation

• Building instructions
  Step-by-step instructions for building EACirc in MS Visual Studio, Qt Creator or GNU Make.
• Running instructions
  Basic run requirements, command line arguments.
• Recommencing computation
  It is possible to continue the computation wron the save state without the loss of determinism and reproducibility.

Framework development documentation

• Framework components
  Components of EACirc, globally accessible resources.
• Circuit back-end
  Basic principles of circuits and their representation.
• Main computation loop
  Description of main program loop, order of actions.
• Projects
  Interface and requirements for project modules.
• Evaluators
  Evaluator interface and description of standard evaluators.
• CUDA support
  Details of CUDA support for evaluating circuits.
• Testing EACirc
  Build-in self-tests using Catch testing framework.
• Known bugs
  Bugs we know about but will not fix in foreseeable future.
• Third party libraries
  Description, licence and credits of used third-party libraries.
• Input/Output files structure
  EACirc produces number of output files containing circuits, statistics and saved state.

Project specific docementation

• Project eStream
  Security of eStream ciphers.
• Project SHA-3
  Security of SHA-3 hash candidates.
• Project CAESAR
  Security of authenticated encryption candidates.
• Project Files
  Creating test vectors from external files.

Supporting tools, scripts, published work

• Scripts
  Supporting script: downloading and processing results.
• Automation tools
  Oneclick documentation easing running multiple computations on CRoCS BOINC server.
• Published work
  Published papers and presentations about EACirc.

Authors and licence

The framework is developed at the Centre for Research on Cryptography and Security, Masaryk University, Brno, Czech Republic. It's licensed under MIT licence, see the file in the repository root. The project uses some third party code, licence details for these parts can be found in the LicenceReadMe file in the repository root.

• Michal Hajas
  2015-now (Java bytecode simulator)
• Karel Kubíček
  2014-now (TEA)
• Jiří Novotný
  2014-now (build system, CUDA core implementation)
• Ľubomír Obrátil
  2014-now (automation tools)
• Zdenek Říha
  2013-now (Java bytecode emulator)
• Marek Sýs
  2013-now (project concept, results interpretation)
• Petr Švenda
  2008-now (project lead, initial implementation)
• Martin Ukrop
  2012-now (framework, CAESAR, polynomial backend)

Former:

• Milan Čermák
  2012-now (initial CUDA implementation)
• Ondrej Dubovec
  2011-2012 (SHA-3 candidates testing)
• Dušan Klinec
  2014-2014 (polynomial backend)
• Matěj Prišťák
  2011-2012 (object model and refactoring, batch mode for BOINC via XML, eStream candidates testing)
• Tobiáš Smolka
  2011-2012 (BOINC related support)