The Hypernetwork Architecture

A novel architecture for biological information processing

This site contains the program and samples of the hypernetwork architecture on solving classification problems [1]. More information about hypernetworks is found here.

The programs:

Download the source code, and install it with:

1. gunzip hypernetwork.tar.gz
2. tar -xvf hypernetwork.tar
3. cd hypernet/src
4. make
5. make install (install the programs in the directory "~/bin", -please update you path)

Download the compiled programs for linux:

• hypertrain2: training with two hypernetwork organisms.

  Parameters: parameter_file training_data #epochs result_file [Optional plotting: 1=display learning, 2 =display organism]

• hypertrainp: training with a population of two or more organisms.

  Parameters: parameter_file training_data #epochs result_file #population [Optional plotting: 1 =display learning, 2=display organism]

• hypertest: testing the performance of the best organism obtained in training, with a test data set.

  Parameters: parameter_file test_file plotting [1=display learning, 2=display organism]

Examples:

Here are examples of classification problems addressed by hypernetworks.

Please download the files in the following directories and test the performance with the "test.sh" script. If you want you can run the training program "train.sh".

1. The 4-bit identity function, described in the paper. The parameter file, training data, and the hypernetwork . See the paper for details.
2. Full adder.
3. (4-10)-Parity tasks.
4. The Tic-tac-toe endgame problem.
5. The 2x2 bit multiplier.
6. Two spirals.
7. The DNA splice problem . This is one of the best hypernetworks obtained in building the Hyperexon DNA splice site recognition system described in [2].

You can download the complete set of examples and the source code from here.

References:

1. Segovia-Juarez, J., and Colombano S. 2004. A molecule interaction based architecture for evolutionary learning and biocomputing. Submitted.

2. Segovia-Juarez, J., and Colombano S., and Kirschner, D. 2004. Identifying splice sites using hypernetworks with artificial molecular evolution.  In review.

Comments?, please email to: Jose L. Segovia-Juarez, jlsj@umich.edu