-
Notifications
You must be signed in to change notification settings - Fork 1
About
The project is organized around a website that collects and abstracts information about the cerebellum as an associative memory. The website is a community effort. Ideally, it will build a table with an entry for each neuron type, the number of neurons of that type, what neurons they connect to with what fan-ins and fan-outs, nature of connections (excitatory, inhibitory), and firing rate (typical, range). The table is referred to as "Cerebellum Facts." Each "fact" is accompanied by a reference or references to the source. In addition to the list of references, the website has an annotated bibliography, to help viewers navigate the material. The annotations are informal, more like comments. They are written by us participants in the website and are meant to point out particulars about a paper that struck us as significant and prompted us to recommend the paper to others.
Human and animal memories work by association. Among the brain's circuits, the cerebellum's looks the most like an associative memory. It is highly regular, has a small number of neuron types but more neurons overall than the rest of the brain, and it has been modeled mathematically since Marr's theory of cerebellar cortex (1968). Among models of the brain's circuits, the cerebellum's is perhaps the most comprehensive and informative.
Pseudocerebellum is an associative memory for high-dimensional vectors--for vectors with thousands of dimensions. When cued with a vector, it finds the most similar vector or vectors among a set of stored vectors: its function is k-nearest-neighbor search among tens of thousands of vectors. Associative memory is a key component of an architecture for computing with high-dimensional vectors--for "hyperdimensional" computing. Building a fast and energy-efficient associative memory is a major engineering challenge.
The website is for gathering information about the cerebellum that could lead to the engineering of efficient associative memories for high-dimensional vectors. Research of this kind was began in the 1980s and presented by Egon Loebner in a paper on "Intelligent network management and functional cerebellum synthesis" at IEEE COMPCON Spring 1989, but not followed up on. However, new ideas about computing with vectors, starting with Plate's Holographic Reduced Representation (HRR), have made the building of an efficient associative memory into a problem with practical significance.
We will start with neuroanatomy. Where do inputs to the cerebellum come from and in what numbers? Where do outputs go and in what numbers. What connections are internal to the cerebellum, and again in what numbers? The paper by Loebner serves as a model. Its numbers refer to the cerebellum of the cat. We want those connections and numbers also for the human brain and would collect them into a table with an entry for each neuron type. In addition to connections and numbers, the table would include one or more references, including page numbers, to where the information is found.
In addition to cerebellar facts, the website will have PDFs of cerebellum papers and a bibliography, plus comments written by us highlighting the reasons for including the paper in the website. What caught your attention, what did you learn, what might be helpful for someone else?
Cerebellum's importance for motor coordination was established long ago. We therefore expect that information on this website will help us design more agile robots. There is increasing evidence that the cerebellum is equally important for higher cognitive functions, including language. Since the problems of language motivated the development of high-dimensional computing (HRR and its kin), cerebellum-like associative memory is likely to be important for all kinds of learning, not just for learning of motor control. Its huge number of neurons and synapses suggests a capacity that is large enough to support learning over a lifetime.
We hope to make the website into a source of information and a place of self-study. It is focused narrowly on the cerebellum as an associative memory. Even that can be much to keep up with, considering the volume of research on the cerebellum that is reported every year.