Orangutan: A Multiscale Brain Emulation-Based Artificial Intelligence Framework for Dynamic Environments

TL;DR

A novel brain-inspired AI framework, Orangutan, that simulates the structure and computational mechanisms of biological brains on multiple scales, encompassing multi-compartment neuron architectures, diverse synaptic connection modalities, neural microcircuits, cortical columns, and brain regions, as well as biochemical processes including facilitation, feedforward inhibition, short-term potentiation, and short-term depression.

Abstract

Achieving General Artificial Intelligence (AGI) has long been a grand challenge in the field of AI, and brain-inspired computing is widely acknowledged as one of the most promising approaches to realize this goal. This paper introduces a novel brain-inspired AI framework, Orangutan. It simulates the structure and computational mechanisms of biological brains on multiple scales, encompassing multi-compartment neuron architectures, diverse synaptic connection modalities, neural microcircuits, cortical columns, and brain regions, as well as biochemical processes including facilitation, feedforward inhibition, short-term potentiation, and short-term depression, all grounded in solid neuroscience. Building upon these highly integrated brain-like mechanisms, I have developed a sensorimotor model that simulates human saccadic eye movements during object observation. The model's algorithmic efficacy was validated through testing with the observation of handwritten digit images.

Figures (17)

• Figure 1: $Orangutan$ Architecture
• Figure 2: Figure \ref{['fig:neuron:neuron']} depicts a typical $Orangutan$ multi-compartment neuron model. All compartments, from top to bottom, are sequentially arranged as dendrites (star), soma (round), axons (square), and synapses (the bulb-ended line). Figure \ref{['fig:neuron:dendrite']} shows a multi-compartment dendritic structure. Two primary dendritic compartments (MAX and MIN) simultaneously receive excitatory inputs from three presynaptic neurons (green), performing maximum and minimum calculations respectively, before passing their outputs to a secondary dendritic compartment (ADD) for summation. The numbers illustrate the excitatory inputs from the presynaptic neurons and the final excitation obtained by the soma (blue circle) after dendritic integration.
• Figure 3: Various synaptic connection types. From left to right, they are axo-dendritic, axo-somatic, axo-axonic, axo-synaptic, somato-somatic synapses, and autapses. To save space, other synaptic structures listed in Table \ref{['table:synapse']} are not shown in the figure.
• Figure 4: Neural Microcircuits, where both the red and purple neurons are inhibitory neurons. In practical scenarios, $Orangutan$ simplifies neural circuits by replacing excited inhibitory interneurons with the inhibitory axons of presynaptic neurons, thereby reducing the neural network's complexity and operational overhead. \ref{['fig:micro_circuit:side_inhibit']} Lateral inhibition features a neuron (blue) that excites a target (yellow), whilst inhibiting nearby cells (green) via an interneuron (red). \ref{['fig:micro_circuit:back_inhibit']} In recurrent inhibition, the neuron (blue) inhibits itself through an interneuron (red). \ref{['fig:micro_circuit:mutual_inhibit']} For mutual inhibition, each neuron (blue) suppresses the other via an interneuron (red). \ref{['fig:micro_circuit:not_logic']} A NOT gate arises when the excitatory neuron’s (blue) activity is prevented by an inhibitory neuron (red). \ref{['fig:micro_circuit:and_logic']} In an AND gate circuit, the green excitatory neuron activates both yellow and red neurons equally, but red's inhibition prevents yellow from activating. Only with both green and active inhibitory purple neurons does the yellow output neuron activate. \ref{['fig:micro_circuit:or_logic']} An OR gate triggers the output (yellow) when either the green or blue cell is active. When active together, the green neuron uses the red neuron to inhibit the blue synapse, protecting the yellow neuron from overactivation. \ref{['fig:micro_circuit:xor_logic']} An XOR gate activates the output (yellow) only when activity is present in one of two antagonistic neurons (green, blue).
• Figure 5: Predictive Coding Neural Circuit. Prediction cell (green) and input cell (blue) separately excite the negative prediction bias cell (orange) and positive prediction bias cell (purple), respectively. They form a NOT gate microcircuit with each other through their respective inhibitory axons and synapses (red). When the excitation of the prediction neuron is greater than that of the input neuron, the negative prediction bias neuron is activated, and vice versa for the positive prediction bias neuron. If the excitations of both are exactly equal, then neither of the prediction bias neurons will be activated.