Projects / bnns

bnns

BNNS is a research tool for interactive training of artificial neural networks based on the Response Function Plots visualization method. It enables users to simulate, visualize, and interact in the learning process of a Multi-Layer Perceptron (MLP) on tasks that have a 2D character. Tasks include the famous two-spirals task or classification of satellite image data.

Tags
Licenses
Operating Systems
Implementation
Translations

Recent releases

  •  12 May 2014 14:42

    Release Notes: This release contains fixes for configuration and compilation issues on OS X.

    •  10 Dec 2013 19:35

      Release Notes: This release improves CLI based on user comments, and adds the use of readline and history libraries. It adds config file support, --conf-file cmd line opt. It adds command line options to control the row size of windows. It adds the mandatory command line option --task-name. It includes a section 1 man page installed by 'make install`. It improves pattern preparation scripts. It fixes loss of frozen weight info on update of learning rate. It fixes processing of events from empty RFP slots, making empty slots visible. It fixes calculation of MSE across the output layer. It applies a threshold of 255 to PGM images in order to have values in {0,255}.

      •  10 Feb 2013 23:03

        Release Notes: This maintenance release adds support for gcc versions above 4.4, uses a default position for the extrapolation window to avoid a deadlock in XCB-enabled Xlib reported on Ubuntu 12.04.1 LTS, drops an obsolete xpm.h include causing compilation errors because libxpm presence was not checked by the configure script, changes interpretation of the --scale-factor value, and introduces a set of supported values.

        •  12 Jun 2012 15:50

          Release Notes: This beta was focused on refactoring and usability. The command line option sets in Perl scripts that assist in pattern preparation were extended. All control variables have corresponding command line options to make the pattern preparation process comfortable and user friendly. A new interaction feature is included: Freezing and unfreezing of weights coming into a neuron via the Space key at the corresponding RFP. The RFP becomes highlighted to inform users of weights being frozen. Positioning of performance analysis tools was improved.

          •  01 Jun 2012 15:43

            Release Notes: This release disables user interaction during the training process to improve overall stability, extends the set of commandline options and extends the main menu to include previously unstable options, implements an error-energy preview, improves the Perl scripts which assist pattern preparation, and improves the manual draft.

            Screenshot

            Project Spotlight

            OpenStack4j

            A Fluent OpenStack client API for Java.

            Screenshot

            Project Spotlight

            TurnKey TWiki Appliance

            A TWiki appliance that is easy to use and lightweight.