A basic conceptron implementation [wip].
Paper: Paper Cell Assemblies in the Cerebral Cortex, V. Braitenberg 1977
Reading:
- Prof. Santosh Vampala Assemblies of Neurons
- Prof. Günther Palm Cell Assemblies
- Prof. Murray Sherman summarizing thalamo-cortical circuits [Talk 1, 2, 3 these are all super dense and elucidating].
The concept is a hypothetical formalism to model a rich set of essential, computational properties of the human cerebral cortex implements and its nuclei.
In the spirit and fashion of Hebbs Cell Assemblies [Hebbian theory, Braitenberg 1977[fn:1]]. This is an attempt to find a level up from the neurons, trying to see the forest instead of the trees so to speak.
The goal is to find and describe a computational substrate, or building blocks. The goal is not a detailed model of biological neuronal nets, each element is idealized and abstracted, but should be evolutionarily plausible.
—
I want to give the assembly calculus:
- some wires (between nuclei and areas), inspired by M. Sherman’s work on Thalamus
- a thought-pump, from Braitenberg [1977] and G. Palm.
- … travling waves? geometry? holographic encoding? hdv?
That would then be the complete formalism Conceptron
.
The name conceptron is from Braitenbergs 1977 paper Cell Assemblies in the Cerebral Cortex. It’s the functional unit that implements concepts, I suppose. Comes from Perceptron and Neuron. Its fundamental property is how it represents interpretations as ‘excitation modes’ in its network, called cell assemblies [Hebb].
- Very fast (100ms to recognize a new object)
- Doesn’t need many data points to make sense of new objects or something, is on the fly
- It is highly dynamic and robust
- It confabulates
- It supports creative thought and explanation structures
- It supports multi-sensory integration
- It creates magical interfaces to user-level entities, which it uses to navigate the world
- It is energy efficient (presumably its information processing is sparse)
- Presumably very parallel (how else so fast)
- Presumably high-dimensional (how else this malleability of ideas, sub-ideas make sense, vague ideas make sense, perspective and so forth).
- Language acquisition is presumably a bit of a window into the ad-hoc epistemology it creates.
- It makes analogies
- It generalizes rules and regularities
- It can associate and pattern complete
- It creates an on-the-fly interpretation of the world and the animal which it is navigating.
- It can create highly derived and abstract ideas and idea structures on the fly, can compare them and so forth. And all that is recorded in mid-term memory, which is accessible via magic interfaces again.
- It has mid-term memory, which breaks when the hippocampus is broken (HM).
- I guess now I can insert all of cognitive neuroscience: attention, movement, planning, working memory, etc. etc.
- It needs to sleep and dream to function
Brain software is somewhat different compared to the computers we have currently. In some ways, brain software is the real deal. And our computers are just the first clumsy attempts at making a computer. But note that machine intelligences, or human-machine cyberborgs, uploaded minds, etc. Will enjoy the benefits of both current computer technology together with the high-creativity, high-on-the-fly, high-multi-processing-integration computational paradigm of cognition software.
ac.clj
is an implementation of Prof. Santosh Vampalas assembly calculus. [wip]
Assemblies of Neurons Learn to Classify Well-Separated Distributions, Santosh S. Vempala’s homepage.
Using dtype-next.
[fn:1]
Braitenberg 1977 On the Texture of Brains -An Introduction to Neuroanatomy for the Cybernetically Minded.
And the paper Cell Assemblies in the Cerebral Cortex.