2024 Nobel Prize in Physics Awarded for Applications of Statistical Physics to Machine Learning!

The 2024 Nobel Prize in Physics has been awarded to physicist John Hopfield and computer scientist Geoffrey Hinton.

The official prize citation is "for foundational discoveries and inventions that enable machine learning with artificial neural networks". However, I strongly prefer the phrasing along the lines of "for the statistical-physics basis of neural networks", which is how mathematical physicist Barry Simon described it.

Here is the information, press release, and other materials from The Nobel Foundation.

Naturally, I am strongly in favor of more Nobel Prizes being awarded for foundational interdisciplinary work, as that is the world in which I live. Unsurprisingly, many traditional physicists are arguing against and doing their common practice of staking territorial claim (as we also often see on the job market and in other arenas). This is an old battle, and we can look forward to a later phase when somebody gets a physics Nobel Prize for work in networks. That said, I am grateful to see people arguing about science, rather than about other ideological things!

"Detecting Political Biases of Named Entities and Hashtags on Twitter"

One of my papers came out in final form earlier today. Here are some details. (This is in collaboration with computer scientists, and stylistically it is rather different from much of my work. However, you'll still notice my hand in it. :P)

Title: Detecting Political Biases of Named Entities and Hashtags on Twitter

Authors: Zhiping Xiao, Jeffrey Zhu, Yining Wang, Pei Zhou, Wen Hong Lam, Mason A. Porter, and Yizhou Sun

Abstract: Ideological divisions in the United States have become increasingly prominent in daily communication. Accordingly, there has been much research on political polarization, including many recent efforts that take a computational perspective. By detecting political biases in a text document, one can attempt to discern and describe its polarity. Intuitively, the named entities (i.e., the nouns and the phrases that act as nouns) and hashtags in text often carry information about political views. For example, people who use the term “pro-choice” are likely to be liberal and people who use the term “pro-life” are likely to be conservative. In this paper, we seek to reveal political polarities in social-media text data and to quantify these polarities by explicitly assigning a polarity score to entities and hashtags. Although this idea is straightforward, it is difficult to perform such inference in a trustworthy quantitative way. Key challenges include the small number of known labels, the continuous spectrum of political views, and the preservation of both a polarity score and a polarity-neutral semantic meaning in an embedding vector of words. To attempt to overcome these challenges, we propose the Polarity-aware Embedding Multi-task learning (PEM) model. This model consists of (1) a self-supervised context-preservation task, (2) an attention-based tweet-level polarity-inference task, and (3) an adversarial learning task that promotes independence between an embedding’s polarity component and its semantic component. Our experimental results demonstrate that our PEM model can successfully learn polarity-aware embeddings that perform well at tweet-level and account-level classification tasks. We examine a variety of applications—including a study of spatial and temporal distributions of polarities and a comparison between tweets from Twitter and posts from Parler—and we thereby demonstrate the effectiveness of our PEM model. We also discuss important limitations of our work and encourage caution when applying the PEM model to real-world scenarios.

The Two–Infinity Norm and Beyond!

Two–infinity and beyond!!! pic.twitter.com/b2KGnIol35

— Mason Porter (@masonporter) October 21, 2019

Counting Bots with a Math Dracula

I forced a bot to watch over 1,000 hours of Sesame Street and then asked it to write an episode of Sesame Street of its own. Here is the first page. pic.twitter.com/0R2Ww4JkH2

— Keaton Patti (@KeatonPatti) July 2, 2019

(Tip of the cap to Kate Owens.)

Digital Archæoludology: Studying Ancient Games with Machine Learning

Digital Archæoludology is a new field that concerns the computational study of ancient game using tools like machine learning. Very cool!

Here is the arXiv paper that led to the above popular piece.

Game on!

A Robot Powered by "Advanced Machine Learning"

;)

When people say 'we use #machinelearning in the business process.'

Sometimes...I know that...🙄 pic.twitter.com/2XQSO5jsod

— Selin Cetin (@robolegist) February 28, 2019

Tales from the ArXiv: Speech-Balloon Detection in Comic Books

There is an intriguing new paper on the ArXiv called Deep CNN-based Speech Balloon Detection and Segmentation for Comic Books.

Finally, an important paper.

Choice quote from the abstract: "Qualitative results suggest that wiggly tails, curved corners, and even illusory contours do not pose a major problem."

A Paper that Rejects Itself

What a conundrum! :P

Amazing. Train a network to classify papers (accept/reject). Then run the network on the paper describing the network, and it classifies the paper as a strong reject. This is why we can't have nice paper classifiers. (h/t @hardmaru) https://t.co/OC2baCI1Za pic.twitter.com/3N9wsKQ374

— Chethan Pandarinath (@chethan) December 21, 2018

(Tip of the cap to Karen Daniels.)

"Detecting Sarcasm with Deep Convolutional Neural Networks"

I think they found it.

Detecting Sarcasm with Deep Convolutional Neural Networks https://t.co/C05UdMjCoe pic.twitter.com/q50x9QqXhD

— KDnuggets (@kdnuggets) June 24, 2018

(Tip of the cap to Peter Rothman.)

A Big Pile of ... Linear Algebra

That's right. In many cases, machine learning is a big pile of ... linear algebra.

April Fooling Around: 2017

Here's an April Fool's Day paper on the arXiv. The title of the paper is: A Neural Networks Approach to Predicting How Things Might Have Turned Out Had I Mustered the Nerve to Ask Barry Cottonfield to the Junior Prom Back in 1997

(The arXiv doesn't have 1 April as a mailing day this year, so we get it early.)

The last sentence of the abstract is amusing: "Over-training is also discussed, although the linear algebra teacher assures us that in Barry’s case this is not possible."

As things catch my eye, I will post links to more April Fool's Day shenanigans.

Here are various past posts related to April Fool's Day (and a couple of other posts that show up in the search but aren't particularly related).

Update: Here is another joke arXiv paper. It is called: Schrodinger's Cat and World History: The Many Worlds Interpretation of Alternative Facts

Update: This article seems to purposely be dated April 1st, but it has a rather different flavor from the other two.

Update: According to an April 1st article in The Guardian, former British chancellor (more formally, "Chancellor of the Exchequer") George Osborne has become a fashion designer. (Tip of the cap to Dominic Vella.)

Update (4/01/17) Here is a screenshot of my April Fool's Day prank of 2006, for which I was able to convince a member of Caltech's public-relations department to post my article (actual fake news, which of course is an April 1st tradition) on the Caltech web page and include a link to it in an e-mail circular.

Update (4/01/17): Cherwell, a student publication from University of Oxford, published an interesting story about possible cancellation of the Cancer Research UK Boat Races. Here is another one from Cherwell.

Update (4/01/17): There's also the matter of the Spaced X rocket launch out of Santa Monica, California. (Tip of the cap to Andrea Bertozzi.)

Update (4/01/17): In other news, an ancient particle accelerator was discovered on Mars. (Tip of the cap to Jean Bellissard.)

Update (4/01/17): George Takei played an amusing prank.

Update (4/01/17): Also, the American Physical Society is launching a new journal called Physical Review Tweets. Awesome!

Update (4/02/17): Google remixed an old (but awesome) prank by letting people play Ms. Pac-Man on Google Maps. (Tip of the cap to Myah Evers.) Google also played a few other pranks.

Update (4/02/17): And here are various other pranks that you may have encountered yesterday.

Update (4/08/17): Well, the Reddit prank appears to have resulted in a rather interesting example of self-organization (and of astounding art, with some "This is why we can't have nice things." thrown in). (Tip of the cap to Kevin Hickerson, Maria Satterwhite, and others.)

How to Annoy Experts in Machine Learning

Fun fact: If you want to annoy a machine-learning expert, refer to what he/she studies as "data mining". Most of them hate that, and their reactions to it are often pretty damn funny. :P