The technology inside Amazon’s new convenience store, opening Monday in downtown Seattle, enables a shopping experience like no other — including no checkout lines.

Cloud AutoML is a service that makes it easier to create custom ML models for image recognition using a drag-and-drop interface to upload images and train and manage models.

The round was led by Morningside Venture Capital and Legend Capital, both early-stage, China-focused venture funds. Little information about Pony is available on the website, but its mission is to “build the safest and most reliable autonomous driving technology”.

AI and deep learning have been subject to a huge amount of hype. In a new paper, Gary Marcus argues there’s been an “irrational exuberance” surrounding deep learning.

A two-part (part1, part2)  tutorial on normalizing flows, which can transform simple densities like Gaussians into rich complex distributions that can be used for generative models, RL, and variational inference.

New theoretical insights and into and novel ways to analyze two-population asymmetric games, such as which include Leduc poker and various board games such as Scotland Yard. 

This post goes over a quick and dirty way to deploy a trained machine learning model to production.

A summary of PyTorch progress over the past year: News, and highlights from the community.

Training very deep neural networks requires a lot of memory. Using the tools in this package you can trade off some of this memory usage with computation to make your model fit into memory more easily.

Official implementations for the recent set of Deep Neurevolution paper (see repository for details). A separate respository also contains code for  the Safe Mutations for Deep and Recurrent Neural Networks paper.

Tensorboard is a Tensorflow-specific tool for monitoring model training. This library allows you to write to events to Tensorboard with a simple function call, which means you can now use it from within other libraries.

This project is a faster PyTorch implementation of faster R-CNN, aimed to accelerate the training of faster R-CNN object detection models.

The concept of innateness is rarely discussed in the context of artificial intelligence. In this paper, the author considers as a test case a recent series of papers on AlphaGo and its successors that have been presented the argument that “even in the most challenging of domains: it is possible to train to superhuman level, without human examples or guidance, starting tabula rasa”. The author argues that these claims are overstated and that that artificial intelligence needs greater attention to innateness.

The authors propose Fine-tuned Language Models (FitLaM), an effective transfer learning method that can be applied to any task in NLP, and introduce techniques that are key for fine-tuning a state-of-the-art language model. The method  outperforms the state-of-the-art on five text classification tasks, reducing the error by 18-24% on the majority of datasets. The authors open-source pretrained models and code to enable adoption by the community.

This paper discusses whether computers, using Artifical Intelligence, could create art. The first part concerns AI-based tools for assisting with art making. The history of technologies that automated aspects of art is covered, including photography and animation. In each case, we see initial fears and denial of the technology, followed by acceptance, and a blossoming of new creative and professional opportunities for artists. The hype and reality of Artificial Intelligence (AI) tools for art making is discussed, together with predictions about how AI tools will be used. The second part concerns AI systems that could conceive of artwork, and be credited with authorship of an artwork.

This article provides a very brief introduction to the basic ideas that underlie deep learning from an applied mathematics perspective. The target audience includes postgraduate and final year undergraduate students in mathematics who are keen to learn about the area.