dpiponi

dpiponi , 1 day ago to random

I just remembered a programming technique from the days when I had access to a Commodore Pet.

I wanted a graph plotting program to allow users to type in a function and plot it. I could have written an expression parser. But far easier was to print out a line of code for the function and a RUN command (or similar), place the cursor in the right place, put a carriage return or two into the keyboard buffer, and stop the program. IIRC the Pet literally read input from the frame buffer (which allowed you to edit code by cursoring into the output of LIST) so this would update your program and continue.

dpiponi , 2 days ago to random

"Autoregressive" seems like a very fancy word for "doing stuff in some order".

dpiponi , 21 days ago to random

Catching up on the last few years of generative ML.

(Pixelwise) average a bunch of pictures of dogs and you get something so blurry it probably doesn't look like a dog. But take lots of random paths from cats to dogs, take the vectors along those paths, average them out to define a vector field in the ambient space, and integrate to get paths, and now you have paths that go from cats to dogs without blur. It's bizarre that something so simple is incredibly powerful.

(It's even weirder than that because this method is sort of self-sacrificing. You can use it to generate pairs that can be used to train a conventional neural net to convert cats to dogs without blur. You don't need to mention the paths or vectors after that.)

(Caveat: I haven't tested specifically on the problem of converting cats to dogs, that's just an illustrative example, but I'm 99% sure it'd work fine.)

https://arxiv.org/abs/2210.02747

dpiponi , 30 days ago to random

I remember mentioning this book [1] on Twitter or G+, maybe 10 years ago. I found it used somewhere. I think I was soliciting recommendations for books like it by which I meant short readable self-contained accounts of some subject, not necessarily completely rigorous, with exercises and solutions.

Anyway, 10 years later I realise this was a really good book to read. I think it gave me the beginnings of good intuition about SDEs [2] and I couldn't have guessed then how useful that would be today.

[1] "An Introduction to Stochastic Processes in Physics" by Don S Lemons
[2] SDE = Stochastic Differential Equation

dpiponi , 1 month ago to random

A recent popular work of sci-fi got me wondering if a biological radio is possible. As cells are basically nano-factories I expect the answer is yes. But what are the most plausible designs for a biological radio receiver and transmitter?

dpiponi , 2 months ago to random

If I was applying for a PhD today I think the topic I'd pick would be related to random matrices. The content of this paper illustrates why in the most readable summary I've come across:

https://www.imaginary.org/sites/default/files/snapshots/snapshots-2018-002.pdf

It (bizarrely!) connects the subject I did study (Riemann surfaces) with physics and random matrices and to keep the CS part of my brain happy there's a recursion scheme. It also hints at why I think String Theory is still an important subject despite me never having "believed" in it.

dpiponi , 2 months ago to random

It makes me happy that there are people in the world who, when I idly wonder about something, have already done the research.

The iridescence you see sometimes see in meat, especially bacon, is likely caused by alternating muscle layers acting as a photonic crystal.

This is different to a more obvious interference related hypothesis that it's just striation on the surface acting a bit like a diffraction grating.

https://academic.oup.com/ijfst/article/56/1/250/7806371

dpiponi , 3 months ago to random

Living in a country where if you ask for a cup of tea in a restaurant they'll bring you a cup of lukewarm water and expect you to dunk a teabag in it I shouldn't be too surprised that I can't buy a decent electric kettle. This time I decided to get an expensive one rather than a cheap one that takes forever to boil and falls apart after a year. Well it seems well made, but produces hot water that tastes of plastic and silicone sealant. Why is this so hard?

(Yes, I could go old school and use a clay pot on open fire like my ancestors did but I'd like modern and convenient.)

dpiponi , 3 months ago to random

A common strategy when implementing arrays that grow at the end is to allocate space for N bytes initially and then when more space is required resize to a block of size AN for some A, typically in the range (1, 2], and then keep reallocating as needed scaling up by A each time.

Let's say A=2. How much memory do we expect to actually occupy?

A good guess might be to say that typically at any moment in time we expect to be half way through the new region so that on average 3/4 of the array is used.

But Benford's law tells us that numbers (for some ill-defined class of number that probably includes array sizes) follow an improper distribution with the property that whatever base you pick, the mantissa M has a cumulative distribution given by log(M).

More precisely, if X is the size of a random array and S is the smallest power of 2 at least X, then we expect P(X/S < z) = 1+log_2(z). Integrating we find

E[X/S] = 1/(2 log(2)) = 0.72...

slightly less than 3/4, but maybe close enough we don't need to worry about it.

https://en.wikipedia.org/wiki/Benford%27s_law

dpiponi , 3 months ago to random

I couldn't get through Wandering Earth. Meanwhile here's how you actually move the Earth:

https://arxiv.org/abs/astro-ph/0102126

"In this paper, we have investigated the feasibility of gradually moving the Earth to a larger orbital radius in order to escape from the increasing radiative flux from the Sun. Our initial analysis shows that the general problem of long-term planetary engineering is almost alarmingly feasible using technologies that are currently under serious discussion. The eventual implementation of such a program, which is moderately beyond current technical capabilities, would profoundly extend the time over which our biosphere remains viable."

Or as Archimedes said: “Give me a firm place to stand and a lever and I can move the Earth.”

dpiponi , 3 months ago to random

I really like Theorem 1 in this paper:

https://arxiv.org/abs/1106.1465

A well known way to compute the determinant of an n x n matrix (aij) is this:

Write 1, 2, ..., n in a row.
Write 1, 2, ..., n in a row below.

Now join each number in the first row to a number in the second with a line.

Each such diagram represents a term like a12a21a33 where we have one factor for each line and the indices are the beginning and end of the line. Sum all of these terms weighted by -1 if the number of intersections of the lines is 1 The total is the determinant. I've sketched the result for a 3 x 3 matrix.

I'm not sure it needs a proof as this is practically one of the standard definitions. But see
https://mathoverflow.net/a/417941/1233

for how the number of intersections arises.

The thing that's new to me is that if you allow not just lines joining top and bottom, but also "cups" and "caps" that double back to the row where they started, you get another method to compute determinants. This time you write the matrix as A+B with symmetric A and antisymmetric B and where the diagonal and above elements of A are aij and the diagonal and above elements of B are bij. The determinant, as a function of (aij, i <= j) and (bij, i <= j) is given in a similar way where we now consider lines to represent aij and cups and caps to represent bij. I have again supplied a sketch but see the paper for details.

There's also a version for matrices written as A+iB.

Sorry: in the diagram I use A=(aij) + (bij) but in the text above it's A=(aij) and B=(bij).

image/jpeg

dpiponi , 3 months ago to random

Prepare to have your mind blown

ALT

dpiponi , 3 months ago to random

Hardly a week seems to go by without someone claiming that they have for the first time achieved true definitely verifiable without a doubt "quantum advantage". Often it's the same people making the same claim each time.

dpiponi , 3 months ago to random

Now I've looked it up on Wikipedia I'm embarrassed by the atrociousness of my pronunciation of Ithell Colquhoun when I bought a ticket for the exhibition of her work at the Tate Britain the other day: /ˈaɪθəl kəˈhuːn/

ALT

dpiponi , 4 months ago to random

Sitting in a coffee bar in Cambridge it seems like a good moment to push out a first rough version of my probabilistic DSL that's good enough to answer actual questions on the RPG stack exchange.

https://github.com/dpiponi/dice-nine

I know it can do this because I pulled down 50 examples from the stack exchange and used them as my tests - this being the version that finally runs all 50 successfully.

https://colab.research.google.com/drive/1sOh3Ie_uD2RXVKGoFZ3MZwXN-t9_5KCQ?usp=sharing

There's not a lot that's original about the code but it performs well compared to similar tools (eg. anydice) despite being 100% Python.

One thing that may be of interest - using Python generators allows you to support a degree of laziness which is really powerful. If you want exact computations for discrete distributions you need to be able to explore the entire probability space which means you want to "materialise" the smallest amount of state you can get away with. For example, lazy_sum(1000 @ d(6)) sums the rolls of 1000 dice but 1000 @ d(6) is a lazy sequence and so only ever materialises one die roll at a time and in the worst case it has thousands of states to manage rather than 6^1000.

Also fun was using Wang's rational reconstruction algorithm to implement fast parallel (in the sense of numpy) high precision rational arithmetic without having to leave Python. (Bit rough around the edges, currently the onus is on the user to detect overflows.)

https://en.wikipedia.org/wiki/Rational_reconstruction_(mathematics)

dpiponi , 4 months ago to random

View from local shopping mall

dpiponi , 4 months ago to random

I find it weird how many physical processes look like extremely unlikely events drawn from some probability distribution. Most of thermodynamics looks like this - the Boltzmann law is what you get if you assume particles have energies drawn from a wide flat probability distribution conditioned on the sums of energies being very low.

Suppose you have a particle doing a random walk on the integers with a 50/50 chance of going left or right at each step. After 100 steps it's very unlikely to have moved 75 steps to the right. Nonetheless, we can still ask about its likely trajectory conditioned on it ending up 75 steps to the right. The first plot is really a stack of 101 plots. Row T of pixels represents the conditional distribution at time T given that the particle walked from 0 at time 0 to 75 at time 100. As you might expect, we get a cloud that expands to maximum uncertainty at the midpoint - and of course the checkerboard pattern due to parity. The key thing is we get a cloud around a straight line...

regehr , 4 months ago to random

30 mil buys you a sauna and a hobbit hole, but it doesn't buy you good taste

https://www.zillow.com/homedetails/13322-N-Grove-Dr-Alpine-UT-84004/2071451490_zpid/

dpiponi , 4 months ago to random

If you say artoo, as an abbreviation for R2D2, which do you say?

dpiponi , 4 months ago to random

My wife, who works in the local libraries, brought this book home one day. Not my usual type of reading material but if you live in the SF Bay Area I think it's a must-read. Seems like all the stereotypes you might have about Marin County were well established 50 years ago. They're all true of course.

Very funny. Made me feel self-conscious for wearing Birkenstock slippers. (Medical reasons...)

https://en.wikipedia.org/wiki/The_Serial

dpiponi , 4 months ago to random

Technician comes to repair dishwasher. Brings wrong parts. Says AI orders parts. Between Republicans and AI I've never before felt so completely surrounded by ignorance and stupidity.

mattblaze , 5 months ago to Photography

"Bowl For Health", Treasure Island, San Francisco, CA, 2007.

All the pixels, shoe rental not included, at https://www.flickr.com/photos/mattblaze/2123765265

#photography

ALT

dpiponi , 6 months ago to random

Back in Oakland having a PhD makes you overeducated and unsuitable for jury duty but that’s not working here in Marin. I’ve a feeling I’m actually going to be doing jury duty for real this time.

dpiponi , 6 months ago to random

I love deleting stuff. Photos that are near duplicates, albums I no longer listen to, blocks of code replaced by a much shorter abstraction, junk email, receipts from ancient purchases, ancient apps.

In some ways it's more of a distillation process, deleting something often raises the average quality of what's left.

dpiponi , 6 months ago to random

I phoned customer service for my phone provider. Impressed with how human the AI sounds. It asks my name. And then makes some joke about pepperoni. Not as bad as praising Hitler but that's still a pretty bad customer service faux pas.

dpiponi , 6 months ago to random

If I ever take up a life of crime I'm gonna buy a load of crime scene evidence markers and leave them around my crime scenes to mess with the detectives.

https://arrowheadforensics.com/products/photo-documentation/evidence-marking-cones-and-tents.html

dpiponi , 6 months ago to random

Before visiting Norway I didn't realise how into weird sculpture Norwegians are. Not just the Vigeland stuff. Sculpture lurking everywhere.

Having just reread Use of Weapons I "enjoyed" the chair.

A gray stone sculpture of a pile of maybe a dozen interlocked naked bodies.
A chair carved into a tree stump with a widely spreading network of creepy roots on a foggy day in a forest.
A wooden sculpture of a two headed seated person sitting on a stool. One head has a shock of yellow hair and wears a shocked expression. The other, looking away. looks like it has a beak. In a foggy forest.

ALT

dpiponi , 6 months ago to random

I have a terrible confession to make. I've been reading the Dragonlance novels. I have an even more terrible confession to make. I'm enjoying them.

dpiponi , 6 months ago to random

Sitting on the sofa at home with some cold-like illness I'm fascinated by the moiré patterns through the insect screen on the windows.

About 7-8m away is some wood. About 3-4m away is a screen. The wood grain is too fine to perceive. The mesh in the screen is also too fine to perceive. But through the magic of aliasing I can see a low frequency version of the wood grain through the screen.

It's a form of heterodyning that makes a high frequency signal (wood grain) perceptible by multiplying by a local oscillator (screen) and filtering (my eyesight).

https://en.wikipedia.org/wiki/Heterodyne

dpiponi , 7 months ago to random

Oh wow! Are there really people who get any answers in the white areas?

dpiponi , 8 months ago to random

Although everyone is into large models these days I like small models that you can understand.

I good example is the Bradley-Terry model of tournaments that I've mention before. If you have historical data on who beat whom in the past then you can build a model by assigning everyone a score s_i and say that the probability that i beats j is

f(s_i-s_j)

where f is the logistic function

f(x)=1/(1+exp(-x)).

The task is then to fit s_i to your historical data. This formula is almost the simplest thing you might make up using the toolkit of machine learning, but it turns out to be a maximum entropy model. (Weird how this happens more often than it should.)

The task of fitting the s_i has a long history and a popular method was developed by Zermelo (yes, that Zermelo) back in 1929.

Anyway, I noticed a very recent paper that does a simple algebraic rearrangement of the underlying mathematics and results in a much faster algorithm to find the globally optimal fit.

https://jmlr.org/papers/volume24/22-1086/22-1086.pdf

Tangential: I'm amused by the fact that this model of game playing was developed by Milton (Terry) and (Ralph) Bradley but has nothing to do with Milton Bradley.

mcc , 8 months ago (edited 8 months ago) to random

For three full years now I've been semi-daily posting a music recommendation in a big long thread, but it turns out 300-post threads kinda break Mastodon, so I have to restart the thread every so often.

This post is a placeholder. I normally make a YouTube playlist for the previous year to put at the top of the thread, but I've had an awful month and haven't finished that yet.

But if you want to see year three's posts, they're here: https://mastodon.social/@mcc/112356066616688565

And here's year four: