Mishka's Page -- history of the main page (since late 1995)

The site was created either in late 1995 or during 1996, when I was doing research as a PhD student at Brandeis and working at Parametric (aka PTC). I wanted to publish my research results somewhere. I was also coming out, and wanted to create a gay website.

That's when my first decent results were obtained. Those were results on computing distances between programs via related metrics on Scott domains, which we obtained together with Josh Scott working at my Newton office of PTC during July 4, 1996 holidays. That office was closed soon after that.

That was also when we lost Anya Pogosyants and Igor Slobodkin. I still remember how I heard about that on December 15, 1995 (that was Friday, the same Newton office).

This was my first visit to Russia after emigration. I presented my first decent paper at Logic at Yaroslavl conference (our paper with Josh Scott describing results from the previous Summer). During that Summer working together with Sveta Shorina we discovered co-continuous valuations.

I saw people I have not seen for many years. The country changed drastically since 1989. This was also the Summer when I was able to explore Moscow gay life.

Not much changes on this Web page. In reality, there were changes. Sasha Chislenko took me to my first rave that Fall, and my life and worldview changed drastically.

After my January 1998 trip to Moscow. Magic winter railroads in Moscow, a love story, huge psychedelic green circles around white lights on railroad platforms, Dmitrij Bushuev reads from his new book "Na kogo pohozh Arlekin", -- none of this is reflected on this page.

But there is a new section, "Essays by Mishka", and two essays were posted there in December 1997 -- on the internal time being a square root of the physical age, and on evolutionary origins of homosexuality.

After my May 1998 trip to England. I presented our joint work with Sveta, "Relaxed Metrics, Maximal Points, and Negative Information" at Mathematical Foundations of Programming Semantics'14. I got a very cool tip from Mike Smyth, which led to my and Sveta's subsequent paper "On a Smyth Conjecture" about relationship between lower bounds of relaxed metrics and tolerances. I met Steve Matthews, the mathematician who discovered partial metrics. Without this meeting I would not have returned to partial metrics in 2005, after leaving the field a few years earlier.

I've fallen in love with London, and with its rave and gay scenes during that visit.

This is the first version of Web page which reflects my involvement with rave community, and also the first version which had the "Love Search" section. That was the period when I was trying to be politically active and to contribute to the struggle for cognitive liberty.

The anti-war page: protesting NATO aggression against Yugoslavia. We actually participated in the street protests against this war during that time. We also warned our senators before the war, that the war would lead to the loss of democracy in Russia (this prediction, unfortunately, turned out to be correct).

On a brighter side: this is the first version which has the "Current Futurist Thought" section, dedicated to technological singularity, transhumanism, uploading, implants, etc.

This the the canonical Jun, 1999 page (the Yugoslavia war is over, so it is dropped from the page).

This version was redesigned on July 1, 2000 to create the new top-level section "Science of Consciousness", and to move "Psychedelic Music and Chemicals" and "Love Search" seactions to the second level to reflect my new focus and priorities.

Here I am starting to focus more and more on neuroscience, philosophy of mind, and futurist topics. Unfortunately, I don't have a pre-9/11 version of the page preserved.

The "War" page. From 9/11 to the fall of Taliban. This page had a black background.

My normal page in 2002-2010.

This was a slowly changing page. The "Science of Consciousness" section was evolving, and so was the professional interests section.

I worked for the same employer during that period. It was very much a post-9/11 thing, with a lot of defense customers.

I focused on neuroscience for a few years till approximately 2006 (and was monitoring it lightly after that). Then in 2005 my colleagues invited me to do joint research in partial metrics with them, and interesting results were obtained regarding essential equivalence of partial metrics and fuzzy equalities (the key observation was made during July 4, 2006 holidays). The fuzzy equalities arose in the context of sheaves and toposes, and this prompted me to stop avoiding categories and toposes, and to pay more attention to mainstream math (and, later, physics). I was also blogging during the last few years.

Later I might try to write a better history of this period. However I cannot illustrate it by an evolution of a single Web page.

The decade is ending, and I decided it is time to change the appearence of the main page. It's time to add this history you are reading, to remove the "War" section, to do some clean-up, to try to make sure that what actually interests me now is better visible, and to update the futurist section.

Those 5 years were a period of scientific progress for us. During this time we completely understood the nature of correspondence between partial metrics and fuzzy equalities, and published the results of our research in this area.

Then it turned out that presence of partial contradictions was tightly connected to vector semantics of programming languages, which led to our discovery of partial inconsistency landscape.

Most importantly, in 2015 we have finally obtained a model of computation which allows us to continuously deform programs and, moreover, to represent large classes of programs by matrices of real numbers.

I worked for the map division of Nokia during most of this period, which was an interesting and extremely strange experience.

I have also added a couple of new essays to this site.

We were focusing on further development of the formalism for programs as matrices. In 2016, we realized that the resulting formalism generalizes recurrent neural networks, if we replace streams of numbers with arbitrary streams admitting an operation of combining several streams with coefficients (linear streams).

Also in 2016: the term dataflow matrix machines was coined. The new space of "flexible mixed-rank tensors with tree-shaped indices" (V-values) was discovered and studied. A general-purpose research-grade open source implementation of dataflow matrix machines in Clojure was created.

The flavor of neural machines we had obtained is expressive enough to use those neural machines as a programming platform. These neural machines have powerful, flexible, and convenient self-modification facilities.

The 2017 was the year of publications. The reference paper, Dataflow Matrix Machines and V-values: a Bridge between Programs and Neural Nets (December 2017), was published. Also, a paper on loop quantum gravity was published in Annales Henri Poincaré.

The 2017 was also the year when it became apparent that the health issues are becoming more prominent as one ages. In 2020, I was reminded of that again (of course, we are also in the middle of CoViD-2019 pandemic right now, and it remains to be seen how radically, or not radically, the world is going to be reshaped by this event).

It does often take time, before new results are noticed and used by the comminity (see, for example, the history of backpropagation, of ReLU, etc), so starting from the year 2018 I was mostly focusing on further experimental and theoretical development of dataflow matrix machines, with the goal of shortening the wait before they are adopted and used by the wider community

The potential of dataflow matrix machines and V-values to lead to breakthroughs in program synthesis and in metalearning is quite strong.

Several new essays and studies on neural networks were posted in the writings on the science of consciousness section in 2017-2019, and a number of informal "github preprints" were published recently following a variety of arxiv preprints which appeared in 2015-2017. I did a lot in open source during these years (visual and audio-visual arts, design notes, various pieces of software, etc). I am also happy that "Revolutionary" short sci-fi was finally published on the internet in 2019 by its author (see Selected Sci-Fi section).

Since mid-2019, a lot of life changes are happening. The 18 year period of working for geographic software industry came to an end. So, I became even more focused on dataflow matrix machines research, and also on branding and packaging dataflow matrix machines in order to make them more accessible for the wider community (my resume and publications page also contains a white paper on dataflow matrix machines and a collaborative research agenda). I am also using the fact that I have more free time to get back to the "science of consciousness", especially given that a lot of new interesting texts appeared recently, including "The Order of Time" by Carlo Rovelli.

I used to prefer Python as the language of choice and PyTorch as the machine learning framework of choice, but in 2020 I am also exploring Julia programming language and falling in love with it and with the superflexible and lightweight Julia Flux machine learning framework.

At the end of May 2020, the next stage of Transformer revolution began.

The field experienced a qualitative jump, with the OpenAI presentation of code generator assistant demo during Microsoft Build 2020 event on May 20, 2020 (see, e.g. https://twitter.com/matvelloso/status/1263193089310461952) and then with GPT-3 paper Language Models are Few-Shot Learners on May 28, 2020 and OpenAI API private beta on June 11, 2020.

We are in the middle of July and we have seen reports from enough people using GPT-3 via that private beta, and we can confidently say that a new epoch has started, the transition at least as consequential as AlexNet had been in September-December 2012 has just happened. It's interesting that this coincides with a period of unusually intense social, political, and economic turbulence (somehow, everything is coming to head all at once).

Technologically speaking, we are at a point when things might start changing at arbitrarily fast rate at any moment, in particular because people are ready to hybridize all kinds of approaches with Transformers, just like they hybridized all kinds of approaches with "deep" nets in recent years. As Jürgen Schmidhuber is saying in recent years, "we are almost there". Now this has literally become true.

Before adding the "🇺🇦 I support Ukraine 🇺🇦" message in connection with Russian aggression against Ukraine.

This is the strangest time. The "second deep learning revolution" based on Transformers and other attention-based models continues at full speed.

Language models competently generate computer code based on the linguistic context (OpenAI Codex, GitHub Copilot, etc.). New methods of training machine learning models and new AI application areas proliferate. We now have models which can take one look at a novel neural architecture and predict reasonably well what will the values of its parameteres be after training (Parameter Prediction for Unseen Deep Architectures (Oct 2021)). Crypto currencies and NFTs are becoming more and more prominent. Charles Stross' Accelerando is gradually becoming a reality.

All that is happening on top of a gradually diminishing global pandemic, with the origin of the virus still being debated. And as the pandemic is gradually winding down, various conflicts, coups, civil wars, campaigns of repression are erupting everywhere.

Finally, Russia is invading Ukraine in September 1939 style, while rather abruptly completing its own long slide to the social system more and more closely resembling the social system of the 1939 aggressors. However, the aggressor's military is performing quite poorly and failing in a rather spectacular fashion while meeting competent resistance and taking losses. Meanwhile, the world recognizes the new acute danger and rapidly unites against the aggressor. The nuclear dimension of the conflict is quite worrying; there is also plenty of uncertainty, as Russia and the West had not conducted full nuclear tests for 30 years, and the maintainance of the weapons had been less than adequate, especially in Russia, but also in the West.

Since August 1, 2022 it has been necessary for me to physically visit the Brandeis campus in order to maintain and update my Brandeis site. As of November 20, 2022, the remote access is still down.

Therefore I have created a “partial advanced mirror” on GitHub:

https://anhinga.github.io/brandeis-mirror

• To enable advanced edits of the active parts of the site (to be backported to the original Brandeis site during my campus visits)
• To have an explicit publicly accessible backup of the most pages of this site