Main Page

Overview emergent is a comprehensive, full-featured deep neural network simulator that enables the creation and analysis of complex, sophisticated models of the brain in the world. The software differs from other tools (e.g., Matlab, python) in providing a full-featured GUI for constructing, visualizing, and interacting with the neural models (in a 3D display), so that people with little to no programming experience can use it effectively. This is important for teaching applications. It also supports the workflow of professional neural network researchers, with a powerful scripting language system called css (not the other css), which uses the familiar C++ syntax, including python/matlab like Matrix extensions. Programs also have a full GUI that also allows novices to automate network training & testing, and construction of the input environment, while simultaneously supporting the expert with a text-based editor interface. Full interactive debugging and error-checking facilities are provided. Model outputs can be analyzed using DataTable data processing operations (filtering, grouping, sorting, dimensionality reduction, etc). The same DataTable functionality is used for presenting inputs to the networks, and it is straightforward to write programs to generate any sort of input (interactively or statically) for the networks. In addition, the 3D GUI also features a complete Newtonian physics simulator, allowing you to create rich robotics simulations, including a biophysically realistic human arm with 12 muscles, and realistic visual processing of images (which can come from virtual cameras in the virtual environment, or from the real world) according to principles of early visual processing. As a direct descendant of PDP (1986) and PDP++ (1999), emergent has been in development for decades, and has been used in hundreds of scientific publications from a variety of different labs. Detailed models of the hippocampus, prefrontal cortex, basal ganglia, visual cortex, cerebellum, and other brain areas are available (and described in our textbook). A large number of classic neural network algorithms and variants are supported, including Backpropagation, Constraint Satisfaction, Self Organizing, and the Leabra algorithm which incorporates many of the most important features from each of these algorithms, in a biologically consistent manner. In addition, the symbolic / subsymbolic ACT-R architecture is now supported as well. News and updates now: Current SVN Status 8/25/14: 7.0.1, See 7.0.1 ChangeLog. 6/23/14: 7.0.0, See 7.0.0 ChangeLog. 8/28/13: 6.3.2, See 6.3.2 ChangeLog. 4/7/13: 6.2.0, See 6.2.0 ChangeLog. 12/20/12: 6.1.1, See 6.1.0 ChangeLog. 11/9/12: 6.1.0, See 6.1.0 ChangeLog. 10/8/12: 6.0.3, See 6.0.0 ChangeLog. 9/26/12: 6.0.2, See 6.1.0 ChangeLog. 8/31/12: 6.0.1, See 6.0.0 ChangeLog. fixes brain view bug 1500 8/30/12: 6.0.0, See 6.0.0 ChangeLog. 6/5/12: 5.3.10 beta, See 6.0.0 ChangeLog. fixes broken MIN, MAX, ABS css functions -- critical fix. 5/4/12: 5.3.9 beta, See 5.4.0 ChangeLog. 5/3/12: 5.3.8 beta, See 5.4.0 ChangeLog. 4/25/12: 5.3.7 beta, See 5.4.0 ChangeLog. 4/12/12: 5.3.6 beta, See 5.4.0 ChangeLog. 3/13/12: 5.3.2 beta, See 5.4.0 ChangeLog. 2/15/12: 5.3.1 beta, See 5.4.0 ChangeLog. fixes the gelin activation function's temporal dynamics to be more graded over time, as the old vm-based activation function was 12/23/11: 5.3.0, See 5.3.0 ChangeLog. 10/31/11: 5.2.6 beta, See 5.3.0 ChangeLog. 10/24/11: 5.2.4 beta, See 5.3.0 ChangeLog. (has bug in Epoch loop; don't use) 10/1/11: 5.2.3 beta, See 5.3.0 ChangeLog. 9/20/11: 5.2.2 beta, See 5.3.0 ChangeLog. 9/1/11: 5.2.0, See 5.2.0 ChangeLog. 5/24/11: 5.1.7 beta, See 5.2.0 ChangeLog. 4/11/11: 5.1.6 beta, See 5.2.0 ChangeLog. 3/21/11: 5.1.5 beta, See 5.2.0 ChangeLog. 3/09/11: 5.1.4 beta, See 5.2.0 ChangeLog. 3/07/11: 5.1.3 beta, See 5.2.0 ChangeLog. (has an annoying bug) 2/23/11: 5.1.2 beta, See 5.2.0 ChangeLog. 8/16/10: 5.1.0, See 5.1.0 ChangeLog. 4/21/10: 5.0.2, See 5.0.2 ChangeLog. 12/5/09: 5.0.1, See 5.0.1 ChangeLog. 9/9/09: All textbook projects have been updated to emergent 5.0.0 and are available on the CompCogNeuro wiki. 8/25/09: 5.0.0, See 5.0.0 ChangeLog.

Documentation and useful links Documentation, About emergent, Screenshots emergent-users mailing list, Submit a bug Keyboard shortcuts, Color coding, Matrix css Development, ViewVC, CodeReview, OpenGrok Comparison of Neural Network Simulators Download emergent 7.0.1 Windows instructions 64bit only: First install MSVC++ 2010 and MSVC++ 2012 (runtime C++ libraries, select vcredist_x64.exe for each -- only needed if you haven't done this already) and then install emergent-7.0.1-7737-win64.exe or emergent-7.0.1-7464-win64.exe. OSX instructions 10.8 / 10.9, 10.10: Install emergent_install_mac_deps-7.0.0.dmg (note: this release uses same dependencies as 7.0.0 -- no need to re-install) and then either emergent-7.0.1-7737-mac64.dmg (10.9, 10.10) or emergent-7.0.1-7464-mac64-10.8.dmg (10.8). After downloading, right click and select open. 10.7: Only 6.3.2 is avail: Install emergent-6.3.2-6447-mac64-lion.dmg, qt-mac-opensource-4.8.2.dmg and Subversion-1.7.13_10.7.x.pkg. 10.6 32 bit: only 6.1.0 is avail: Install emergent-6.1.0-5751-mac32.dmg Ubuntu Instructions Our apt repository contains packages for emergent 7.0 for precise and trusty releases. There are older versions of emergent for karmic, lucid, maverick, natty, oneiric, precise, quantal, raring, saucy and trusty. See Build (Linux-Ubuntu) for help building the latest version from scratch. Debian Build Instructions See Build (Linux-Debian) for help building any revision of emergent from source. HTTP download or FTP download You can find older binary packages for emergent and source packages for all of its prerequisites on the HTTP download or FTP download sites (same source, just different TP). Source code To get the emergent source code using Subversion run this command: svn checkout --username anonymous --password emergent https://grey.colorado.edu/svn/emergent/emergent/trunk ~/emergent