Explore

Extradynamical Evolution

9-17-2025 — DAF meeting with others, AK and JH, regarding

Extradynamical Evolution⁠

⁠

Andrius sees the Kalman Filter in terms of the

Three Minds⁠

— What is known, questioning, and investigatory/integration.

Kalman filter.

1 — Sensory evidence, Observations, Measurements

2 — (Hidden) Regularities.

Classical SPM setting for this, y is a vector of neuroimaging data, and x is the time evolution of the dynamical neural system

1 & 2 come with errors —

How do you integrate this information? What information to lean more in favor of? How to weigh?

The weighing factor, is the Kalman Gain. Between 1 and 0. When the errors of 1 and 2 are equal, the KG is .5

Error of the 2 is assumed fixed for here.

Error of the 1 can be updated.

⁠

AK associates the KG blending with the 3rd Mind.

If the measurement is good, you go largely with it. If the measurement is poor, it is largely ignored.

Then AK turns to

Active Inference⁠

and Free Energy Principle.

y = m1 = evidence

x = m2 = state

P is the information with regards to sensory world.

Questions — What does it mean to have a niche, be a niche-coevolve?

⁠

Three Minds⁠

Evidence + States in the mental world.

Comparing sensory P + variational distribution Q.

Andrius points out that Q can be simpler. That is true, also a key general feature here, is that Q is partitioned or articulatable, completely separate from P’s structure. So there can be an isomorphism or coarse-graining between Q-P, or Q can be arbitrarily under- or over-specified. Further, Q gives us access to an entire suite of computational tools, diagramming, incremental optimizations via Variational Bayesian methods, sampling and filter/particle meth.ds

Perception-Action loops — update your mind, update your actions (can have different gradients, like being at a saddle point).

JH —

Linear Dynamical System — What does it mean for a system to stand on its own.

x vectors, 1,2,3. — The way to get from one state to another state, is multiply with a matrix A (this can be read in the inbound or perceptual direction, as the mapping from Observation-State “Tale of Two Densities”, and can be read through time as the Markov Transition Matrix, the time evolution — that gives us a “chessboard” of moves through Kairos and Chronous).

When the time evolution matrix is fixed, the number of timesteps is exponentiated.

In Biology, there is breakdown into Niches.

⁠

The single niche is broken down — “ideal” is A3 = 0, the bottom left quadrant.

⁠

[— In this situation, the larger dynamics are not in play at all.

There is feedback from Subsystem to Larger. HOWEVER there is no cross-talk downwards. So from the perspective of the living organism, it is “perfect niche”, it is like an acidity-producing fermenting bacteria, that continually pumps.,

This is like the Infinite Heat Bath situation.

If there WERE a perfect niche — how would it evolve? JH does not know!

In terms of Kalman Filter — Could there be an evolutionary algorithm, that adjusts the matrix / subspace angle.

“In the laboratory, that is what we do”. Wall off the world & get a sub-system that operates independently — e.g. Heat bath (simulates “infinite heat bath” for the reaction), behavioral setting with stationary (static or limit attractor) stimuli. If we allowed the laboratory animal to engage in open-ended bidirectional co-evolution, there would be an open-endedness in principle, and a historiocity in practice, so that there would be no “replicates” within or across time series.

Could the Kalman filter context, help us?

Here a key feature of ACTION is needed. That takes the observer off the bench. E.g. we could have Kalman filtering radio enthusiastic, purely recieving and processing signal to refine their sound. One style of action, would be changing the frequency dial on the radio. Deeper manipulations to the radio apparatus could be done, also fitting under the auspice of action.

And then separate from playing with the Radio dial (fine-tuning, online inference), and Radio anatomy (structure learning, weight updating, topological changes to neural networks m1 and symbolic networks m2) — there is the interaction with the other radio sender.

What are the Systems that allow this learnability and breakdown? Are Markov Blankets general, or an example of this class?

What are Learnable systems?

Is there a hierarchy or breakdown? Theoretical composition, of what systems are learnable or not? —

Could that help us understand, why the universe is learnable? The laws of physics might be there, but there is too much interaction/noise/crosstalk. Cosmologists think about this in terms of decreasing temperature, allowing the parsing? Non-Locality in Quantum.

Load content from www.math4wisdom.com?

Loading external content may reveal information to 3rd parties. Learn more

Allow

Want to print your doc?
This is not the way.

Try clicking the ⋯ next to your doc name or using a keyboard shortcut (

CtrlP

) instead.