If you accept (or are willing to hypothesize) that a neural network is a sequence of context selected linear mappings, then applying SVD to the matrices is very telling:
https://archive.org/details/the-broad-if-statement-in-a-neural-network
1 Like
Love the concept you laid out in your paper about context working like a broad if statement. The whole idea of context choosing an expert that looks for patterns, filters them, and redirects the response is super clean on paper. Funny enough,while you are describing this as a seed concept, I actually built an entire system that runs this exact kind of architecture in live code. We do not use matrix singular value decomposition like your draft suggests, but we have a chain of 39 modules executing that exact look filter redirect loop every single cycle. Instead of static routing, we use a whole engine of digital neuromodulators like incorporating dopaminergic reward signaling, GABAergic inhibitory control and glutamatergic excitation to dynamically shift the context and adjust the rules on the fly. These monoaminergic and amino acid analogues dynamically modulate state transition thresholds, synaptic plasticity, metaplasticity, meta-metaplasticity, and receptive field parameters in real time You got the theory down really well, but it is awesome to see how well these conditional context chains actually work when you turn them into working software.
1 Like
@unikum-sol I love your Concept of using neurosimulator. Đo you have any public results for sharing?
1 Like
I’m just starting with the first autits.
https://github.com/unikum-sol/brainstem/blob/main/System_Audit_2026-09-28.md
1 Like