The ERU Hub

In the past two weeks, I’ve focused on improving the auditory cortex, to match the detail and requirements in the Beads primer / design notes.

The Auditory Cortex and the ERU Hub

A1

Let us talk some more about population coding. In essence, it means viewing the cortical stimulus as a whole. When we move from independent, intricate action potentials to abstract ideas, higher neurons have to make a holistic sense of detail. The spectrotemporal receptive fields in the primary auditory cortex are examples of such neurons. This group of neurons have affinities to frequency spectra, periodicity and temporal spikes. For example, if a neuron has to match with the sound aaaaaa, the frequency spectrum will be narrow, the temporal window wide and periodicity low. In contrast, if a neuron is looking for the sound ur, the frequency spectrum will be wide while the temporal window is narrow. So a massive population of such neurons sit in the primary cortices, looking for patterns, and firing when close matches are found. As a result, those synapses trigger abstract sections of the brain.

A2 and beyond

As we know, spike trains typically represent instantaneous stimulus. For example, consider an image of a car, a shrill sound, a tangy taste or a sweet smell. You can probably infer that these stimuli, in silo, lack context. However, a mature brain will tag multimodal temporal information and prior cognitive processing as contextual information to them; so “I'm currently in a Mercedes dealership, so the car is a Mercedes”, the image of a metal piece links it to the shrill sound, “I'm eating pasta”, “I'm in a honey market, so the sweet smell must be from honey” become the corresponding context for those examples. This principle is at the core of biological reasoning.

To further understand why cortices are essential, let's take a look at their Beads-specific conceptual twin - the ERU hub. The ERU hub acts as the gateway manager for information before it enters the brain. The multilevel ERU hub (equivalent to A2, A3 and beyond) routes stimulus for different bursts of information. The sound aaaa can be the start of “Amelia” or “Amazing”, so the hub deciphers the correct section of the brain to route to. However, a natural follow up to the above would be “how shallow or deep should the hub be? If it's too shallow, you end up with a single level of routes. If it's too deep, you risk neural fatigue”. The systemic brain has different answers to that for different velocities of thinking -

1. Slow thinking / Neurogenesis - When new neurons are created, detail is prioritized over efficiency, so there aren’t many direct connections to the Hub for that particular stimulus. Side note: notice how you hold great nuance for new skills / topics you just learn about? This is because newly created “long chain” synapses are biologically “fresh” and can hold neurotransmitter links and vesicles for longer
2. Fast thinking - When mature neurons / topics are “invoked”, the brain focuses on moving quickly from synapse to synapse. So, the abstract neurons representing these intuitive topics are closer to the hub

Algorithms / Development

Building the Environmental Response System (ERS)

1. For each neuron in A1
   • Initialize a gaussian spectral profile, tuned to a center frequency and its neighbors
   • Initialize a temporal Gabor kernel with cosine and gaussian components
   • Outer product of the two vectors makes up the STRF for that neuron
   • Apply STRF to the incoming spike stream, to get the temporal drive of a neuron

1. Define the facilitation dynamics i.e., if a neuron is under focus, boost its vesicle flow
2. Define the depression dynamics i.e., if a synapse produces too many vesicles, exponentially decay the flow to factor in refractory periods

1. Define the leaky potential mechanics of the entire neuron
2. Define the conductance-based excitatory and inhibitory mechanics of the dendrites
3. Define the spike frequency adaptation dynamics of the soma
4. Define the spike generation mechanism of the AIS (axon initial segment)

Development Activity - https://github.com/akhil-reddy/beads/graphs/commit-activity
Building the auditory cortex components - https://github.com/akhil-reddy/beads/blob/main/beads/core/eru/hub/audio/a1_cortex.py

Please note that some code (class templates, function comments, etc) is AI generated, so that I spend more of my productive time thinking and designing. However, I cross-verify each block of generated code with its corresponding design choice before moving ahead.

Next Steps

Building the Environmental Response System (ERS)

1. Building the visual cortex
2. Building the ERUs
3. Neurotransmitters - Fed by vision’s bipolar and amacrine cells, for example, to act on contrasting and/or temporal stimulus

Deployment

1. Overlaying video frames onto the retina, including code optimization for channel processing
2. Overlaying audio clips onto the cochlea, including optimization for wave segment processing
3. Parallelization / streaming of cellular events via Flink or equivalent