Modelling Biological Neural Networks

GIF references: Screenshot capture by H Muzart. All structural/functional models fully or partially (i.e. modified from original scripts) built and made by H Muzart, using open-source computational tools by others (SimBrain 3.0, Emergent PDP++, and some Blender 3D, Matlab nntool, Human Brain Project platform, Tensorflow Playground).

Buiding BNNs with ANNs underlying human cognition, in SimBrain

   So here, I have used SimBrain 2.0 & 3.0 (Tosi & Yoshimi 2016, simbrain.net) for part of my CCNBS model, inspired by other works (e.g. Hebb/Rosenblatt et al, O'Keefe/Burgess et al, O'Reilly/McClelland et al, Dayan/Rolls et al, Sejnowski et al, Hinton/Rumelhart et al, etc). The model was built with Java and XML files. 

     Since 2015-2016 (final year UCL BSc), I started getting more of an interest in this type of software, and then started to use these more in 2017 and used those again more recently in 2020. This also relates in parallel to the neural networks that one may observe in histology that I have been interested in since 2012. Here in SimBrain however, it is based in silico in software, rather than silicon hardware or biological wetware. 

     In software like SimBrain, One can model fully recurrent networks, dynamic attractor states, backpropagation networks, Pavlovian conditioning, etc. Interesting intelligent patterns can come from simple rules, as systems are scaled up. It has also been applied to many other issues in scientific research [cognitive attention control, dopaminergic systems, etc; links 1, 2, 3, 4], and here I have tried to fit it to my model of emotional episodic memories from my novel cognitive paradigm [CognTech section] and neural data from the literature [CognTech section]. 

The neural network I made is for now only a demonstrable proof of concept for:

• Training and learning a simple backpropagation system as a classifier that can differentiate 2 feature types between 2 room objects.

• Pattern completion within hippocampal attractor state dynamics for episodic memory, where partial cues help recall the whole memory, with damage to this mimicking neurological/psychiatric conditions.

• Modifiable Hebbian learning rules, spiking neurons, architectures including memory stores.

• Interactive inhibitory competitive network conditioned to pattern associate features that spatially and temporally belong together on various scales (with 4 room scenes out of 12).

These can help model both cognition at the whole-brain/organismal macroscopic level, but also at the neuronal level. There are emulations of molecular effects on neural circuits, and purely psychological phenomenology.

Ultimately this contributes to my other work (CCNBS res pub pres). 

There are so many possibilities to take all this further.

BNNs with ANNs, with Emergent PDP++

   Emergent PDP++ (v.1.0 to v.8.0, 1990, 2008, 2012, 2020) is a software on computational explorations of the mind (from R O'Reilly, Aisa, Munakata, McClelland, Frank, et al; there is the official book, html-formatted documentation at grey.c.edu/e & psych.c.edu/e , emergent.c/..../cortex , code , vid playlist ; written in C++ and Go, with a graphical (UI) user interface).

Like SimBrain - it can model almost anything from the last 100 years in computational cognitive neurobiosciences.

See how I have briefly used it below.

Future plan includes: TBC

BNNs with ANNs, with HBP

     Since 2010, there have been new brain initiatives - the Human Connectome Projects ([1] , [2]) (USA), BRAIN initiative (NIH), Blue Brain Project (EU), Human Brain Project (EU), dHBP (dev HBP; KCL/Imperial/etc), RIKEN (.jp), etc. One of interest is the Human Brain Project (see HBP HM profile, HBP Forum HM profile, e-brains profile) , and this has become fascinating as it allows open-source collaboratory work on anything from brain simulations to AI agent to animal molecular circuits to neurorobotics to neuromorphic computing, etc. These e-platforms, I am still learning about, but I soon intend on making many contributions.

< Word doc (on Medion) of screenshots >

BNNs with ANNs, Blender3D

     Here, I have used Blender 2.0 (Blender can be used for almost anything in 3D modelling) to start to construct 3D visuo-spatial models of networks of neurons. Animations have been added for effects over time. There is open-source code out there that exists for neuronal processes (e.g. spikings, activation functions, ...) which I plan to incorporate, in Python, so that these structural models also become functional ones. This is similar to many other projects out there.

[---X---] Blender API -- installation on OS, with 3D builds with C & Python interoperability.

    > web-embed implementation on Sketchfab.com - https://skfb.ly/oETZ6  (VR-view enabled & device compatible) , https://sketchfab.com/harry.f.muzart

TF.PG, Matlab NNs, other, etc

What I've also used:

[---X---] Tensorflow Playground (Git) -- browser-based visualisation models of deep machine learning ANNs and statistical inference models (not necessarily BNNs)

[---X---] Matlab NN Designer / Simulink -- this is not strictly about biological neural networks, it is more to do with data processing pipelines in deep machine learning - for example neuroimaging data files or camera vision data files, etc.

Future plan to carry on everything all further, based on the above:

• Like others (Brian NN Simulator (Imperial College); DigiCortex SpikeFun; NeurophStudio; TVB; neuroConstruct (UCL) using neuroML; etc) which have biologically-plausible ANN models, different spiking models and topological neuromorphometric components.

• Dopamine, serotonin, oxytocin, etc, can be explored in terms of neuronal circuits computational models (see CognTech Mol and ANNs/BNNs) and can be used for remote open-source online human cognitive experimenting potentially using oral/intranasal (non-IV) doses, which are relatively safe and cheap ( ...... , ...... ).

Other models of interest includes ones with place cells (Bicanski, Burgess, et al), grid cells (UCL x DeepMind), hierarchical visual neocortex (Dayan, Sejnowski, Carandini, et al), etc....

My aim with all this is ultimately to use my experience in neurobiology, computational science, and 3D digital visual rendering, to produce models similar to those: [1] [2] [3] [4] [5] [more tbc]; and these would not just be merely aesthetic and for-education-purposes, but they will be actual functional models, based on scientific research and they themselves will further expand scientific research. Furthermore to establish DIY home-based labs for everyone, with multiple monitor screens, just like NASA/SpaceX command control centers, but instead of the 3D geographical cartography of Earth & outer space, it would be neurons in the human brain.

I have already created my on projects and have been contributing to other larger projects by others.

More to do:

TBC

For example, useful other external links of interest:

TBC