tvachnad / neural-network Goto Github PK

View Code? Open in Web Editor NEW

2.0 2.0 2.0 45.66 MB

License: MIT License

CSS 4.07% HTML 8.31% JavaScript 77.20% Processing 2.83% Python 7.59%

neural-network's People

Contributors

Stargazers

Watchers

Forkers

jburroni aamgalan

neural-network's Issues

when logging the input, log both the source and target neuron

neuron should not use neighbours' astrocytes

Astrocyte.replenish should use setTimeout

Currently, it's using setInterval, which means that it's called every 20 seconds. But we only need to call it just once

keep js dependencies internal (for offline work)

Namely, in the simulation_page.html, can we just download anything like this one needed for logging: https://ajax.googleapis.com/ajax/libs/jquery/1.11.3/jquery.min.js and keep it in the vendor js folder, rather than get it every time?

Also, the main homepage is waayy to scripty for our minimal needs (sorry Cassie). We could accomplish the same with something vastly simpler than bootstrap.

For both of these, for example if we're at an ONR conference with no internet (they have high security and no wifi allowed), we want to be able to launch the simulation without needing to be online.

the number of signals is bigger than the number of axons

This is not feasible so we should check what are we doing wrongly.

Add this assert to spot the problem:

        // if(n.astrocyte.availableEnergy<=astrocyte_settings.minEnergy)
        // n.astrocyte.replenish();
    }
    assert(this.allSignals.length < this.allAxons.length);
    if(this.logger != null){
        if(this.logger.getLastEntry() >= 19){
            this.logger.sendToServer();

add analysis code to fulfil item 5 of the scales behavior's document

Deprecate old menu options

Max signals and max axon distance are no longer relevant or accurate. Max neuron connections is also broke/invalid? On a related note, Master branch has had many menu terminology updates, and there might be a conflict during merge w/ the newly deleted menu options.

Need to change checking of if the neuron can fire during update

// update neurons state and release signal
        for (ii = 0; ii < this.allNeurons.length; ii++) {
            n = this.allNeurons[ii];
            // the astrocyte we're taking energy from
            if (n.receivedSignal) { // Astrocyte mode
                if (n.fire() === true) {
                    var a = n.canFire();
                    if(a === true){
                        if(this.logger != null){
                            this.logger.addToLastEntry(ii+1);
                            this.logger2.addToLastEntry(n.acc+0.125);                   
                        }
                        a.deplete();
                        n.lastSignalRelease = currentTime;
                        this.releaseSignalAt(n);
                    } else if(a >= 0){
                        //log misssed activation astrocyte energy
                    }
                }
            }

To log the energy of the astrocyte when it didn't have enough energy to fire, but the neuron would have fired, the order of checking if the neuron will fire needs to change. First check if the neuron will fire, then check if the neuron can fire from the astrocyte. This also means changing the way that canFire() is defined. A helper function is also needed to return the value of the astrocyte.

d3.csv is asynchronous

this code:

    //load connectivity matrix
    d3.csv("./static/models/connectivity.csv", function(data) {

        for(var q = 0;q<188;q++){
            self.connectivityMatrix[q]=data[q];
        }
    });

is the code to load the connectivity matrix. The code that follows assume that when its run, the connectivity matrix is already loaded, but that is not true as d3.csv is asynchonous

simulations get slow when the number of simultaneous signals reach 40k

Astrocyte Regeneration is on global timer and not synced on ticks

The regeneration is not synced to the ticks that update the network but a clock on the users computer. This leads to the regeneration time getting out of sync with the simulation when the simulation slows down.

Proposing this change to the function to have it update when neuralNetwork.update() is called

    NeuralNetwork.prototype.updateRegeneration = function(currentTime) {
        var timeSinceLastUpdate = currentTime - this.lastRegenUpdate;
        if(timeSinceLastUpdate >= astrocyte_settings.frequency){
            astrocyte_settings.replenishEnergy += this.regenSign*astrocyte_settings.amplitude;
            this.lastRegenUpdate = currentTime;
            if(astrocyte_settings.replenishEnergy > astrocyte_settings.maxThreshold){
                astrocyte_settings.replenishEnergy = astrocyte_settings.maxThreshold; 
                this.regenSign *= -1;
            } else if(astrocyte_settings.replenishEnergy < astrocyte_settings.minThreshold){
                astrocyte_settings.replenishEnergy = astrocyte_settings.minThreshold;
                this.regenSign *= -1;
            }
        }
    }

Neurons Decay membrane potential on a timer not on ticks

The Neuron.prototype.decay function isn't used and neurons only decay based on the decay function which use a timer. Need to make this depend on a tick based clock. Proposing this change where if a neuron doesn't receive a signal then it decays using the decay function:

Neuron.prototype.decay = function(){
    this.acc = 0.9 * this.acc;
    if (this.acc < 0)
        this.acc = 0;
}
Neuron.prototype.fireIfCan = function(neuralNet, currentTime)
{
    var ret = [false, null];
    if (this.willFire())
    {
        if (this.receivedSignal) {
            var astrocyte = this.astrocyteWithEnergy();
            if (astrocyte != null) { // Astrocyte mode
                var prevacc = this.acc;
                this.fire();
                ret = [true, prevacc]
                astrocyte.deplete();
                this.lastSignalRelease = currentTime;
                neuralNet.releaseSignalAt(this);
            } else {
                ret = [false, this.astrocyte.availableEnergy];
            }
            this.receivedSignal = false; // if neuron received signal but still in delay reset it
        } else {
            this.decay();
        }
    } 
    return ret;
}

tvachnad / neural-network Goto Github PK

neural-network's People

Contributors

Stargazers

Watchers

Forkers

neural-network's Issues

Recommend Projects

Recommend Topics

Recommend Org

Jobs