"Information compression as a unifying principle in human learning, perception, and cognition"

[Gerry Wolff]

As its title suggests, this recently-published paper: "Information compression as a unifying principle in human learning, perception, and cognition" (PDF, Complexity, vol. 2019, Article ID 1879746, 38 pages, 2019, https://doi.org/10.1155/2019/1879746) describes evidence for the idea that information compression is a unifying principle in human intelligence.

That principle is the foundation for the SP System, meaning the SP Theory of Intelligence and its realisation in the SP Computer Model, described in "The SP theory of intelligence: an overview" (PDF, Information, 4 (3), 283-341, 2013, bit.ly/1NOMJ6l).

Comments will be welcome.

Gerry Wolff

[Dmitry A. Kazakov]

Post by Gerry Wolff
As its title suggests, this recently-published paper: "Information compression as a unifying principle in human learning, perception, and cognition" (PDF, Complexity, vol. 2019, Article ID 1879746, 38 pages, 2019, https://doi.org/10.1155/2019/1879746) describes evidence for the idea that information compression is a unifying principle in human intelligence.
That principle is the foundation for the SP System, meaning the SP Theory of Intelligence and its realisation in the SP Computer Model, described in "The SP theory of intelligence: an overview" (PDF, Information, 4 (3), 283-341, 2013, bit.ly/1NOMJ6l).
Comments will be welcome.

Fundamental terminology/understanding issues:

1. Information cannot be compressed. Compression is a term related to
representation, encoding of [information].

2. In general, information is either there or absent, nothing to compress.

3. Learning, perception etc denote a process of translation from one
source (e.g. visual input) into another (cognitive model, space of
classes, parameters of a classifier, sets of decision rules etc). Since
both are different one cannot talk about compression. In older days we
would call it information/knowledge extraction.

4. Though it is a trivial fact of machine learning about the number of
states (not information) of:

raw inputs >>> features space >>> class space

I would not call it compression. It is extraction of relevant
information and throwing away the rest.

5. And finally, this process of permanent transitions and reductions
cannot be a foundation of intelligence, because this is how everything
in the world works. Relevant factors take effect, the irrelevant do not.
One cannot attribute this exclusively to intelligence. Rather the
intelligence works this way because there is no other way at all.

One could argue that this is how we view, understand, model the world.
OK, but that is a different story.

[Manuel Rodriguez]

Post by Gerry Wolff
As its title suggests, this recently-published paper: "Information
compression as a unifying principle in human learning, perception,
and cognition"
Comments will be welcome.

I've read through the paper and found some pros and other points which i
doesn't like. The paper is an overview article which means, it collects
and presents existing information to a subject. The amount of references
is 106 while the paper itself is 35 pages long. So it cites on each page
around 3 sources which is too little. If the idea was, to give an overview
then the average citation count should be at 6 sources per page and more,
because the font is small and the text is formatted in two-column mode.

And now i would like to explain what i didn't like. The SP-theory
is presented as an alternative to Bayesian reasoning. The underlying
datastructure is a probabilistic network. What does that mean? Using
a concept graph and a linked list is a good idea in computerscience
because it allows to store semantic information in an efficient way. But
introducing a random factor and uncertainty into the memory pattern will
make the overall system unpredictable. I think, in this point the paper
needs improvements.