Two years since my last post in this journal, and two years since the revolutionary events of the first half of 2020.

* The start of the major pandemic and the dynamics of the collective social response to that, the resulting impact was comparable to a soft version of the third World War (which is now gradually becoming less soft and more acute, especially this year).

* The GPT-3 revolution marking the transition from AI 2.0 (Deep Learning) to AI 3.0 (learning from a single or few examples, more and more competent program synthesis from linguistic clues, competent fast computational protein folding, "foundation models" based on Transformers or sometimes on hybrid models using artificial attention, etc).

The turbulence and the pace of change are only increasing. Very difficult and very interesting times are likely to be ahead of us this year and further down the road.

My main online platforms are still the dreamwidth and twitter accounts mentioned in the previous post.

My study (with a set of computational experiments) on the role of regularization in learning the network topology in a relatively novel neuroevolutionary scheme for intrinsically sparse networks:

https://github.com/anhinga/synapses/blob/master/regularization.md

( a shorter description here )

Crosspost: https://anhinga-drafts.livejournal.com/30356.html

https://www.youtube.com/watch?v=gL2L7otx-qc

More on linear models of computation:

( Read more... )

In anhinga-anhinga friends page:( Read more... )

The Fourth Conference on Artificial General Intelligence is hosted by Google in Mountain View, California on August 3-6, 2011:

http://agi-conf.org/2011/

A quote by C.S.Peirce.

I have not realized that he was a panpsychist. The most natural (although not the only) interpretation of this quote is that awareness is universal, and that matter comes as a result of "particles of mind" forming patterns due to "habit". I like this model.

Is this the programming environment I was dreaming about?

http://en.wikipedia.org/wiki/Scratch_%28programming_language%29

I was running an experiment since the New Year trying my own custom slow weight-loss method (with the idea of reversing a slow weight-gain long term trend, and achieving a slow weight-loss long term trend). I finally achieved the milestone I've set to myself 5 months ago (going down back to the local minimum reached in September 2008 by losing 12 pounds).

I had a nice guilt-free technique which worked for me for smoking control (a high voluntary per-cigarette "tax"), but I could not figure out how to apply that to food.

However, it turns out that there is a very simple method: do not consume calories in the evening (say, after 6pm; it makes sense to allow a couple of "grace days" a week, otherwise weight loss might be too steep and unhealthy, and one doesn't want that, and one also wants to have some penalty-free and guilt-free wiggle room). This restriction is so simple, that it is easy to self-impose a voluntary "tax" scheme for breaking this rule. The "tax" should be affordable, but sufficiently high to induce some reluctance to break the constraint.

The details should probably be tuned individually, but if someone wants to know how I tuned it for myself, I'll tell you privately.

(comments to this post might be initially screened)

I'll be placing the references to systems equipped with non-commutative conjunctions in comments to this post; mostly, this will be about generalized equalities valued in quantales.

The context here is that there is a duality between metric and logical viewpoints:

http://anhinga-anhinga.livejournal.com/70309.html

This duality comes from the fact that distance can be thought about as a degree of inequality. Whether all we have here is two notions equivalent up to a dual viewpoint, or whether there are deeper dualities lurking underneath remains open.

When people consider sets equipped with equalities valued in the algebra of open sets of a topology ("Omega-sets"), the natural metric counterpart of that is the notion of partial ultrametric valued in the algebra of closed sets of the same topology. The independently made generalizations to fuzzy equalities valued in commutative quantales and to partial metrics valued in commutative quantales also coincide up to dual viewpoint/dual notation.

There is plenty of interesting work related to fuzzy equalities valued in non-commutative quantales, the references I'll be collecting here might be helpful in doing something interesting with that on the metric side.

http://www.math.tulane.edu/~mfps/mfps26/Titles_and_Abstracts.html

В Оттаву ехать, всё таки, облом, но есть интересные доклады, особенно Эскардо..

"I'll begin by developing and applying the topological view of computation to perform seemingly impossible practical and theoretical tasks, based on some of my papers, including "Infinite sets that admit fast exhaustive search" and "Exhaustible sets in higher-type computation". For example, (1) sets that admit exhaustive search in finite time are topologically compact, [...]"

Надо будет понять, как это работает:

http://www.cs.bham.ac.uk/~mhe/