Brain Simulator 0.6.0

Early access release (2016-12-01)

School for AI

New features

Python API - Added Python interface for MyProjectRunner that allows you to control simulation of a brain project from code. You can also use some classes of the School for AI.

Tiling observer for tensors - The memory block observer can now show the data in tiles according to tensor dimensions. See the new option ObserveTensors.

Improvements and fixes

New nodes

Brain Simulator 0.5.0

Early access release (2016-04-12)

Brain Simulator is now licensed under Apache License, version 2.0!

New features

Undo - it's now possible to undo and redo model changes such as connection or node removal

Dynamic model (experimental) - the ability to programmatically change the model at runtime, add and connect new nodes, etc. (see dynamic model API description)

Changed memory block dimensions API - improved API for manipulating with memory block dimensions (the TensorDimensions class), removed user defined dimensions from UI (dimensions can be adjusted only on dedicated places such as Join node output or the memory block observer)

Brain Unit testing framework (experimental) - the ability to easily create node and brain tests using Brain Unit node or by defining a simple class (see the node and brain testing guide)

Memory block metadata - the ability to attach any metadata to memory blocks (such as preferred visualization method)

Select backward edges - select which connections in a cycle are backward to adjust simulation order

Scripting node group - a node group that allows to define a custom task scheduler in a script

Improvements and fixes

Known bugs

Brain Simulator 0.4.0

Early access release (2015-12-16)

New features

Dashboard – a new panel where node and task properties can be pinned and controlled

Node categories & Toolbar – node toolbox is now sorted into categories; redesigned node selection UI with improved full-text search

Profiling – when turned on, collects profiling information and shows it in the debug/profiling window and the main graph view

Dynamic memory – blocks marked as dynamic can be reallocated in node/task code

Memory blocks dimensions (experimental) – memory block attribute that allows arbitrary dimension setting both from the UI and from code (please do not use it from code yet, as API will probably change)

New worlds and nodes

Improvements and fixes

For developers

Known Bugs

Brain Simulator 0.3.0

Early access release (2015-10-29)

New Features

AssociativeNetworkWorld: serves for importing information about some objects / concepts and their relationship to each other into Brain Simulator. In particular, it works with this information in a textual form, such as "the mouse is small." Here, the words "mouse" and "small" represent the concepts and "is" represents the relation between them

CSharpNode and PythonNode: allows internal scripting inside Brain Simulator. It is possible to program simple tasks right inside the Brain Simulator GUI (without the need to have Visual Studio)

Batch learning: enables the training of neural networks on multiple data samples (called mini-batches) at one time. Mini-batches help smooth out noise present in individual samples. This can lead to better results and quicker-than-standard online training

Brainz: makes it possible to save both a project and trained state data into one file

More Features and Improvements

New Worlds

Bug Fixes and Updates

Known Bugs

Brain Simulator 0.2.0

Early access release (2015-08-31)

New Features

TextWorld - We`ve enabled the use of text as input for Brain Simulator, a feature that was previously unavailable. TextWorld now converts text to a format that neural networks understand, allows the user to choose between two types of sample text inputs, and makes it possible to load external plain text files to be used as input

SoundWorld - SoundWorld update is similar to TextWorld, but uses sound rather than text as input. It brings the processing of raw audio samples inside Brain Simulator (for example, for speech recognition tasks)

StackLayers - StackLayers provides users with a mechanism for building more complex neural networks. The update enables the stacking of several neural networks in parallel and also putting another layer above them, and connects all networks in the stack to it. This ensures that information between the stack and the higher layer flows correctly

GaussLayer - the layer samples from the distribution that is parametrized by the output of the previous layer and this sampling is further used as the input to the following layer. Then it can be applied to sample novel (unseen) data from the distributions.


Bug Fixes

Known Bugs

Same as in the release 0.1. Sorry.-)

Brain Simulator 0.1

Early access release (2015-07-07)

Known Bugs