A Scanner window has multiple “receptors”, where each receptor overlaps one pixel in the image. When a receptor is over a pixel, it sends the pixel color values to the network.

There is no filtering of the image pixels.

The Definition program uses a GUI to define the network structure. The network structure data is written to a “NetFile”.

A very common neural structure is lateral connections from a neuron to surrounding neurons in the same layer. TANSAV can model this by mapping an origin node’s fanouts to same-layer target nodes within a defined radius of the origin node.

The drawing shows a small square Scanner window, but windows can be much larger and can be rectangular or elliptical. Scanning is similar to human saccades, but basic raster scanning can be used.

The drawing shows a layer looking from above. It depicts how the output of the origin node (Orange-colored) only affects the inputs of nearby nodes on the same layer.

As shown, for a mapped connection each origin node output can connect only to a group of nodes within a

user-defined radius in a target layer. The program allows different types of mapping

The files are, of course, encoded binary with checksums.

Overall, the model looks something like a standard ANN configuration, with multiple inputs, a body, and a single output. However, all input connections to the node are to “synapses”, represented by the dark dots. The synapses are attached to “dendritic segments” which send the input signals to the node body.

Although the drawing shows connections between two node layers, connections can also be from source outputs to layer node inputs.

An additional source type is the “Reader”. A Reader reads rows of numerical data values from a table stored in a disk file, and sends these values to network node inputs.

The data values in the table have to be converted to the signal levels used in the network, but a utility program can be used for the conversion.

With the overlapping of targeted connections, a given node in a target layer will only be affected by the outputs from a limited number of origin nodes. This contrasts to fully-random connections, where any target layer node can be affected by the output of any origin layer node.

A primary requirement is that all nodes in a layer have similar characteristics (the same “node-type”). Different layers can have different dimensions and node-types.

A simplified drawing of the TANSAV "node" (neuron) model is shown below.

The drawing shows only four segments with a few synapses each, but nodes can (and should) have many segments and thousands of inputs.

However, some brain connections are not random, so TANSAV allows the specification of “mapped” connections.

The Simulation program uses a CLI to read the NetFile and run the simulation using image data files. As the simulation proceeds, selected results are written to an output file.

A standard network structure in TANSAV is the Layer array, which is designed to model brain layers.

About

Sources provide inputs to the network. Sources are similar to layers in that they have outputs that connect to node inputs.

The primary source type in TANSAV is the “Scanner”. A Scanner has a window that moves over an image, as depicted in the drawing below.

Connection between brain regions in humans are not that well understood. But connectivity is at least as important as neuron operation in determining the network functionality. What is well known is that different areas of the brain use different types of connections.

Sources

Program Suite Usage

Neuron Model

As shown, a Reader sequentially scans each line and sends the values to the network. A Reader is something like a linear Scanner that inputs numbers rather than pixel values.

The random connections are indicated with the green lines. Obviously the drawing greatly over-simplifies the actual number of connections, which can equal the number of total node inputs in a target layer.

The above drawings show only a single origin node’s connections. When multiple origin nodes in one layer connect to the same layer, their connections to the target nodes can overlap. The drawing below on the left shows how two origin nodes’ connections can overlap on another target layer. The drawing on the right depicts how connections from an origin layer can overlap with lateral connections on the target layer. As with biological neural development, origin nodes compete for connections to a target layer.

Layers

The drawing below shows the overall program usage method. As can be seen, the main suite consists of three programs linked by two data files. A third set of files is used to input the image data.

Connections

The above drawing is very simplified and doesn’t show several additional program capabilities.

As is typical, a layer is composed of multiple nodes arranged in a rectangular configuration. Thus, layers are defined by an X-length and a Y-length, defined in node counts.

Like other ANN simulators, TANSAV can create random connections from an “origin” output to a “target” layer. Each output can connect to any unconnected node input in a target layer. Such random connections may actually model the connectivity between certain biological brain areas.

The Analysis program uses another GUI to read the output file and present the simulation results in various text and graphical formats.