'Birdwatching saved me from my gaming addiction' When things were at their worst Edward Bartlett was playing computer games twenty hours a day, sometimes only pausing to eat and sleep. I was addicted to video games, the 28-year-old said. I lived with friends, but for two years I never really saw them in real life. We talked through a gaming microphone. Now the University of Sheffield zoology student has swapped his gaming headset for a pair of binoculars as he's embraced a new passion - birdwatching.

This paper presents a margin-based multiclass generalization bound for neural networks that scales with their margin-normalized spectral complexity: their Lipschitz constant, meaning the product of the spectral norms of the weight matrices, times a certain correction factor. This bound is empirically investigated for a standard AlexNet network trained with SGD on the mnistand cifar10datasets, with both original and random labels; the bound, the Lipschitz constants, and the excess risks are all in direct correlation, suggesting both that SGD selects predictors whose complexity scales with the difficulty of the learning task, and secondly that the presented bound is sensitive to this complexity.

New Scientist recommends Jamie Bartlett's insightful How to Talk to AI I don't use AI chatbots, so you might wonder what use I could make of Jamie Bartlett's book, . Well, this plain-speaking guide makes the compelling case that, despite their popularity, we don't know how to speak to chatbots properly. Few of us have had adequate training on getting the most out of AI - or on how to protect ourselves from it . That's where it can all go very wrong, sending us down misinformation rabbit holes or fostering emotional dependence. Mastering the art of prompting a chatbot is about more than AI, says Bartlett.

The sufficiency result can be extendedtodeepernetworks;weshowthatan L-layernetworkwith W parameters in the hidden layers can memorizeN data points ifW = Ω(N).

This perspectiveparallels an earlier phenomenon inthe much better understood field of optimization where convexity has played a preponderant role for both theoretical and methodological advances [Nes04; Bub15].

The performance of the proposed algorithm isevaluated on awide variety of machine learning tasks.

This profoundly shapes the way we view and understand traditional MCMC sampling algorithms, deviating from the Markovsemigroup path.