Welcome to the 16th edition of MLguru, a bi-weekly Machine Learning newsletter. In this edition you will read about:
Booksby.ai – a bookstore that sells AI-generated novels.
OctConv: An efficient alternative to standard convolution.
And much more!
Are you a Machine Learning pro already? Great! Join us as a Senior Machine Learning Engineer and help us empower international clients like Volkswagen, IKEA, and Keller Williams, as well as startups and industry innovators. Visit our job posting to find out how we can help your career.
Booksby.ai is an online bookstore selling science fiction, AI-generated novels. Through training, the AI has been exposed to a large dataset of science fiction books and has learned to generate new ones that mimic their language, style, and visual appearance. None of the stories, titles, descriptions, book covers, or reviews related to any of the books on Booksby.ai have been written or designed by humans. All items on Booksby.ai are for sale on Amazon.com and can be ordered as printed paperbacks. Read more
The good old days where people played chess against computers are gone. Now, Artificial Intelligence is playing esports. This January, a preliminary version of AlphaStar challenged two of the world's top players in StarCraft II. Since then, its authors took on a greater challenge: playing the full game at a Grandmaster level under professionally approved conditions. How did it do? Read the article.
Octave convolution (OctConv) is an easy-to-implement alternative to standard 2D or 3D convolution. It provides a simple and effective way to reduce computation and memory, allowing the use of larger, more powerful models under the same computational budget.
OctConv replaces a standard convolution in neural networks without requiring any other network architecture adjustments. It also boosts accuracy for image and video recognition tasks, while reducing the memory and computational footprint during both training and inference. Read more about it on Facebook.
Confident learning (CL) is part of supervised learning and weak-supervision. Its aim is to characterize label noise, find label errors, learn with noisy labels, and find ontological issues. Unlike most machine learning approaches, confident learning requires no hyperparameters. Sound interesting? Read about all the benefits it may bring.
I’ve mentioned many versions of GPT-2 in previous MLgurus. Now let’s take a look at the 1.5B release. It’s the largest version along with code and model weights to facilitate detection of outputs of GPT-2 models. What are the findings? Did humans find GPT-2 outputs convincing? Find the answers in the newest openAI post.