Solving physics-based initial value problems with unsupervised machine learning

https://link.aps.org/doi/10.1103/PhysRevE.111.055302

27•opnac•8mo ago

Comments

staunton•8mo ago

This paper is solving (basically) high-school-level problems by training neural networks on the "obvious" cost function. All of those problems can be solved much cheaper by standard numerical solvers for ordinary differential equations. They don't even compare to standard methods.

So what's the point? Riding the neural network hype?

logtempo•8mo ago

It's a paper done by a phd, so it's part of a larger study that is probably more interesting than this paper.https://etheses.dur.ac.uk/15828/ (I'm not the author).

But the results and use cases seems to be legit to me. Agin, I'm not an expert on computer science and quantum physics.

ktallett•8mo ago

I would imagine this research is the starting point to prove viability. It isn't about solving issues that haven't been solved, it is often a good point to find new techniques to improve upon current ones eventually that can then be applied to other issues.

(Plus yes, I expect funding was easy to get because of AI)

gus_massa•8mo ago

It would be nice to know if it can solve some hard equations like https://en.wikipedia.org/wiki/Stiff_equation

tanderson92•8mo ago

This is basically all work in the physics-informed ML literature. (As another commenter points out and links to, more and more people have been increasingly frustrated with the hype of this subcommunity).

What is more amazing is that they have conned their way into the funding agency priorities and have broadly affected hiring at universities.

staunton•8mo ago

Is it? I thought the "neural differential equations" stuff sounded cool, https://en.m.wikipedia.org/wiki/Neural_differential_equation

Don't know if it's been used for anything practical yet but modeling dynamical systems which you only partially know by making use of data sounds useful.

acc_297•8mo ago

Today this front page post has a companion front page post probably worth reading in tandem:

https://news.ycombinator.com/item?id=44037941

CGMthrowaway•8mo ago

One of the points seems really important: Because AI researchers almost never publish negative results, AI-for-science is experiencing survivorship bias.

AI will accelerate greatly the survivorship bias crisis we have already seen. Because there are so many more reasons to reject an AI-driven result.

The distribution chart, which I assume was pre-AI, is really scary. It implies that 85-90% of results are never published. https://substackcdn.com/image/fetch/w_1456,c_limit,f_webp,q_...

Al Lowe on model trains, funny deaths and working with Disney

Hoot: Scheme on WebAssembly

First Proof

OpenCiv3: Open-source, cross-platform reimagining of Civilization III

Software factories and the agentic moment

Stories from 25 Years of Software Development

Reinforcement Learning from Human Feedback

Start all of your commands with a comma (2009)

The Waymo World Model

France's homegrown open source online office suite

The AI boom is causing shortages everywhere else

Vocal Guide – belt sing without killing yourself

Coding agents have replaced every framework I used

A Fresh Look at IBM 3270 Information Display System

What Is Stoicism?

72M Points of Interest

Unseen Footage of Atari Battlezone Arcade Cabinet Production

Where did all the starships go?

Show HN: Look Ma, No Linux: Shell, App Installer, Vi, Cc on ESP32-S3 / BreezyBox

Monty: A minimal, secure Python interpreter written in Rust for use by AI

Learning from context is harder than we thought

Making geo joins faster with H3 indexes

British drivers over 70 to face eye tests every three years

Hackers (1995) Animated Experience

Sheldon Brown's Bicycle Technical Info

Ga68, a GNU Algol 68 Compiler

Show HN: I spent 4 years building a UI design tool with only the features I use

An Update on Heroku

Show HN: If you lose your memory, how to regain access to your computer?

What Is Ruliology?