#AcademicRunPlaylist - 6/17/24

CELTICS WIN! Prior to the great game, I was able to go on a long run up the coast into Maine from NH, and I was accompanied by championship-caliber talks for my #AcademicRunPlaylist!

First was a thought-provoking talk by Roger Ames on Confucian role ethics at Brown University https://www.youtube.com/watch?v=ceWef867eCk

Next was a fantastic talk by Boris Hanin on scaling limits of neural networks at Rutgers University. Hanin methodically lays out how to understand the relationship between data, network depth, and network width and limits to these models, providing much needed rigor for designing large models and evaluating claims made about them. He also delivers quips such as this gem: "Many people at different levels of masochism have worked this out in various levels of generality." Highly recommend https://www.youtube.com/watch?v=L-zMXCjeqmY

Next was an excellent talk by Kerry McInerney on feminist AI at the CPDP Conferences. McInerney goes through a rapid tour of this topic, then goes into more detail on predictive policing technologies https://www.youtube.com/watch?v=d8NeI0IEehs

Next was an interesting talk by Mert Gurbuzbalaban on the heavy tail phenomenon in stochastic gradient descent at Rutgers https://www.youtube.com/watch?v=xI34W1cQhQY

Next was an amazing talk by Iain Macneil on how to quantify legal rules and the issues with simplified indices in corporate law at the UCL Faculty of Laws. Macneil focuses on the problem of interdependence between laws in addition to essentially random heuristics around scoring/measurement, with some suggestions for improving these methodologies. Highly recommend https://www.youtube.com/watch?v=IKptRokXGWE

Next was an intriguing talk by Nhat Pham Minh Ho on neural collapse in deep neural networks, deriving theoretical guarantees for certain setups, at Rutgers https://www.youtube.com/watch?v=FmQfO1p6fZk

Next was an important talk by Amanda Levendowski on feminist cyberlaw at CPDP https://www.youtube.com/watch?v=zqiYQnSwUR0

Next was a nice talk by Bruno Loueiro on understanding how two-layer networks learn features at Rutgers https://www.youtube.com/watch?v=yknXuSzIzFI

Next was a sweeping talk by Diane Coyle on the history of the GDP measure at the LSE. Coyle reveals the shockingly subjective origins of this metric, its slow rate of change, what's not included, and the implications of all of these sources of error and subjectivity for research and policy. Highly recommend https://www.youtube.com/watch?v=EryTjln7l2g

Next was an impressive talk by Adityanarayanan Radhakrishnan on how neural networks learn features from data at Rutgers. Radhakrishnan presents a new statistical method for characterizing how features are learned in neural networks, even empirically demonstrating that this method captures much of what is learned in LLMs. Highly recommend https://www.youtube.com/watch?v=6hNC1WqQGG0

Next was a great talk by Chris Meissner on the past, present, and future of globalization at the Peterson Institute for International Economics https://www.youtube.com/watch?v=Br-uyC-ZHPM

Last was a wide-ranging talk by Dan Levinthal on viewing organizational adaptation through an evolutionary lens at the Strategic Management Society https://www.youtube.com/watch?v=vB27YJHs_ac