Hacker News Reader

Comment by nextos | Hacker News Reader

I find the prize a bit odd this time since it focused on Hopfield networks and Boltzmann machines. Picking those two architectures in particular seems a bit arbitrary. Besides, Parisi got the prize last year (edit: actually 2021, time flies) for spin glasses. Hopfield networks are quite related. They could have included Hopfield & Hinton too, and it would have looked more coherent.

It is also concerning that lately the Nobel Committee seems to be ignoring fundamental broad theoretical contributions. In this case, backpropagation, where Seppo Linnainmaa could have been one of the awardees. It is a bit sad he and others who have already passed away get little credit for something so fundamental.

View on HN

The Nobel in Physics only goes with experimental discoveries, Peter Higgs didn't get his (deserved since the 70s) until the LHC directly observed the particle.

I agree that Hopfield networks and Boltzmann machines are a surprisingly arbitrary choices. It is like they wanted to give a prize to someone for neural networks, but had to pick people from inside their own field to represent the development, which limited the range of options. There is also the aspect of the physics community wanting to give somebody that they liked a Nobel, and then trying to fit them in. (The prize isn't handed out by a shadowy committee of Swedes, there's an involved and highly bureaucratic process for nomination that requires your colleagues to take up your case.)

I've never heard that it had to be tied to experimental discoveries. For example, Feynman got the prize for Feynman diagrams, path integrals and QED calculations. None of that directly tied to experimental work.

It has definitely been awarded for both theoretical and experimental contributions throughout its history. Many theoretical physicists have received the prize for their conceptual breakthroughs, even without direct experimental verification at the time.

That was the reasoning given when Einstein won his prize for the photoelectric effect, not relativity (although the reasons were actually fairly complicated: https://www.advancedsciencenews.com/the-dramatic-story-behin...)

That was relatively early and it had clearly changed later when they started giving folks like Feynman the prize for something that wasn't at all experiment related.

Interesting. Those are essentially inventions, not discoveries?

Giorgio Parisi's prize proves the committee gives prizes for theoretical discoveries nowadays. This year's prize is more proof.

But, as you said, Higgs got his prize once theories were tested. Hence, theoretical contributors (still alive as per prize rules) could have been included here as well.

Nor did Georges Lemaître, nor other theorists, and in that case the experimental physicists who (accidentally!) discovered the evidence did win Nobels.

Mmmmhhhhh… What about Penrose? Honest question.

And also why Einstein didnt get Nobel prize for his theories of relativity, but rather the 'discovery' of photoelectric effect.

The photoelectric effect was known before Einstein, so the award was for theoretical achievements. (But only after Millikan had done precision measurements on it)

> Hopfield networks and Boltzmann machines

Think of this as a Nobel prize for systems physics – essentially "creative application of statistical mechanics" – and it makes a lot more sense why you'd pick these two.

(I am a mineral physicist who now works in machine learning, and I absolutely think of the entire field as applied statistical mechanics; is that correct? Yes and no: it's a valid metaphor.)

You ain't wrong.

Lots of ML is heavily influenced by fundamental research done by Physicists (eg. Boltzmann Machines), Linguists (eg. Optimality Theory / Paul Smolensky, Phylogenetic Trees/Stuart Russell+Tandy Warnow), Computational Biologists (eg. Phylogenetic Trees/Stuart Russell+Tandy Warnow), Electrical Engineers (eg. Claude Shannon), etc.

ML (and CS in general) is very interdisciplinary, and it annoys me that a lot of SWEs think they know more than other fields.

Having studied control engineering, it feels like ML is control theory + optimization all the way down :)

I love how folks from different backgrounds can interpret it in so many ways.

ML is remembering that computers can do math.

It’s also signal processing.

Parisi won in 2021, not last year. His work was more about establishing spin glasses as a way to study complex systems. Hopfield definitely built on that, showing how those ideas could be applied to neural networks and info storage in state-space machines.

As for focusing on Hopfield networks and Boltzmann machines, I get where you're coming from. They’re just a couple of architectures among many, but they’re pretty foundational. They’re deeply rooted in statistical mechanics and have had a huge impact, finding applications across a range of fields beyond just machine learning.

Thanks, that's right, 2021 not 2023. Corrected.

> It is also concerning that lately the Nobel Committee seems to be ignoring fundamental broad theoretical contributions.

How lately? Einstein never got one for special and general relativity, for example.

Reply on news.ycombinator.com