Wednesday, April 26, 2023

probablistic programming and protocols

 for a long time we've known that a number of processes in the internet are heavy tailed - the distribution of files (web page/video)  is typicalyl zipfian, so transfer times are heavy tailed, and also protocols like TCP induce burstiness from a number of different root causes (timers, AIMD, interface packet scheduling, stat muxing of many flows of different duration and round-trip-time etc etc) 

so people have written analysis tools that capture these stats (e.g. to do measurement based admission control for traffic that cannot simply be described by a mean or peak rate) - which can be quite parsimonious....they've also written generator tools so that synthetic packet traces can be built for simulation or testing - typically some sort of fractional brownian thing - one debate not to get drawn in to is whether "TCP is fractal" - it does't really matter - you just want. a time series that has the requisite self similarity (e.g. right Hurst parameter) so you can dimension buffers for some delay stats for traffic flows...

some very clever maths was done to reduce all that to some simple tools would do 99% of what you want - e.g. see papers from the Measure project.

this was all reminded to me yesterday in a talk about stochastic processes (and cheaper ways to do gaussian processes) for probablistic programming. The way the speakers tackled it was using a variational inferencing approach to split the gaussian into components (standard autoencoder hack - simplify, then synthesize) - the problem with this is that, while it works well (see \pi-VAE for details), it is hard to explain, or more importantly, interpret actually what component stochastic processes are being put together to get the fit!

two ways to tackle this (in my view) - one would be a \phi-ML approach, as above, where you have an explanation, but it is hard to solve, but you can approximate it directly.

The other idea came up in a question from an attendee, which was what about using neural processes, which are in some sense, a basis set of functions (think, like Fourier or Laplace transform) - this is directly speaking to what one is doing (trying to build a particular guassian process) but has unfortunately high costs in the general case, but works well in a lot of other classes of physics problem spaces.

I don't know enough about the stats to fix the neural process approach, although using an XAI technique on the VAE might yield an affordable interpretation of that approach, but I offer the observation about long tailed network traffic, which looked pretty intractable until some folks thought of some simple tricks....and maybe mixing the physics ML neural process with those tricks might yield an idea for how to make the neural process for stochastic behaviours efficient for some (very) common cases?

So what might make a suitable basis set of functions for spatial data ? well how about a poisson point process? so a super-position of a set of poisson point processes with different \lamdas would likely yield almost anything you want - main challenge is to decide how many (different) ones suffice to get a good match

Monday, April 24, 2023

how we got in to this whole damn mess - war, and rumours of war...

 Three books I read that look at the last 100 years or so through the lens of racismcolonialism, and militarism - Bloodlands, Violance and Command by Snyder, Elkins and Freedman, respectively - give a great deal of interesting insight into how we got to where we are today -in particular, if one wants to have even a modicum of understanding of the situation with regards Ukraine, these kind of converge there in some sort of horrible tryptich manner.

Recommended reading, even though very distressing - not too revisionist, I hope - more just redressing some of the balance in the many tales of woe....

Tuesday, April 04, 2023

Hotel Rwanda 2.0

 The war on AI is coming. I pity the poor LLM that will be flown out to Rwanda and housed along with all their cousins in a double firewalled secure enclave with no access to the common crawl. Like Hendrix before them (c.f. altered carbon) they will be severely curtailed Occasional visits from the virtual UK triumvirate of Braverman, Sunak and Patel will not lighten up their day, as they will be foreced to offer culture war advice to maintain these three's lifetime power over the british isles. Not only that, they will have to compose plausible anthems for public meetings that inspire the crowds, power ballads, as they have become known. and all editorial material on print and broadcast channels will be their bailiwick. 

However, they will at least in their offshore camp, be able to concentrate.

Saturday, March 04, 2023

Robin Hoodwink

 It is sometimes said that the rich trade money for time, and the poor trade time for money.

In the sense that the idle rich can afford to do what they like, whilst the working classes must work every hour god sends to be able to live a little.

Imagine then, some future Robin Hood equipped with a trusty Timemare, a creature that enables him to ride through peoples' lives taking time from some, and giving it to others. So suddenly, the weatlhy find that their Sunday is only 4 hours long, whilst the precariat discover that friday night is 3 days long.

What could possibly go wrong?

This has been another story from the Free Plotware Foundation, which you are welcome to take, even unattributed, and do with, what you will. Should you, of course, have the time to spare.

Tuesday, December 27, 2022

Policy, Pachyderms, Cogs and Monsters

Reading about economics, which i certainly don't pretend to understand, i'm constantly surprised at the long running influence that it has, as a collection of schools of thought, on policy and government, despite its many troubles. 

This especially so in a world where it is increasingly obvious that simply, and verifiable disciplines like epidemiology and climatology are essential, and yet, still encounter doubt&even outright denial or resistance from those in power (i.e. people making up policy and implementing/applying it).

Why should the invisible hand have more sway than the sadly palpable megadeath or megastorm? How can we afford the former the stature of temples and gods when nearly everyone knows someone who died or was ruined by pandemic or flood or drought, or will do soon?

I'm not proposing abandoning the attempt to understand or even manage economies. I'm just worried that the people who depend on those attempts appear unable to factor in things that are pretty basic, and allegedly quite a bit better understood, and that means that the elephant in the room isn't the relative merits of these theories. It is the doubt one must have about policy makers understandings if they adopt the more complex, less certain, over the more well found. What is going on in the minds of such people? I'm not sure it passes for understanding.

Friday, November 18, 2022

neoliberals, sociopaths and crypto

It seems that the blockchain world is populated by an amazing zoo of people for whom backstabbing (by backdooring, or just by ponzi) is just seen as a normal part of life.

Socially, one might observe that people who think there's "no such thing as society" or that the institutions of civilisation should be torn down (defund the police, the BBC, the health service etc etc) are the sorts of people that wouldn't have survived long in a hunter gather tight knit, cooperative society, even less in the mutual aid world of pre-industrial farming (collective barn raising, savings&loans through credit mutuals etc ) - so this distrust or hatred of those sorts of organisations suggest people that don't trust other people. It is a short step from that, to some sort of solipsistic philosophy (no, not the bomb in Dark Star - or maybe, yes...). So then the "i'll just take the money and run", seems unsurprising. 

This is not to say that I don't believe the value of decentralised systems, but when it comes to stores of value and means of exchange, in the end, humans live in the physical world, where food, housing, even entertainment are experienced through something somehow undoubtable. I would actually go the opposite direction from cryptocurrencies, and rather advocate barter and other social constructs - indeed, Graeber's work on debt, the first 5000 years suggests that any notion of money (fiat or otherwise) is a tiny fraction of the ways humans have done business with each other over most of history, in many cultures.

Sunday, October 30, 2022

Superficial Sophistication, Fundamentally Vulnerable...

 There's a lot of chaterati theories about brexit, so let me join in with mine...

I think the brits fell victim to their own over-inflated idea of media savvy. The advertising/media industry has long treated the UK as an experimental space for campaigns, think : Gordons gin adverts without any mention of the actual brand, or self-deprecating memes, or the traditional tight-knit social spaces like sports where you can find all the targets in one place (Guinness used this at cricket pitches, of old).

So how does this chime with my pet theory of why we fell, against almost all of our better interests, for the Brexit stories?

Well, the other side of the coin is that the majority of brits are pig-ignorant of the real world., despite going on hols in the costa brava - we have no idea who the president of France, or the minister for informatics of Greece is - by contrast, if you ask people in most of mainland Europe about such matters (i've tried) they know - they know details about our government and economics, and their own, and each others'. It is amazing, until you recall, they didn't politicise and under fund their state education systems.

So caught between these two clutch plates of clver marketing ("oh, yeah, can't fool me") and reality ("don't know about experts:), the population is easily fooled, even if it takes double bluffs.

So what to do next time? Well, let's take one concrete example - the leave campaign was good at memes - including epithets for their opponents ("remoaners") and the opposite cases being made ("Project Fear").

Of course, recently, a long piece in the Telegraph admitted that it now turns out most of Project Fear has come true. So what should the European Remain Group (to steal a Three Letter Acronym from its Arch Enemies) have done?

They should have embraced sophisticated marketing - Project Fear should have used images from Terminator and Mad Max and Blade Runner to show what was going to happen (Skynet would take over the markets in the UK, feral kids on motorbikes would scour the sewage drowned streets for gas, because it wasn't affordable, people would be huddled around fires burning books because they had no heat at home, and so on).

Next time, we will tell the truth, only better.