Deep Learning Weekly Issue #108

Facebook F8, Google I/O, new CV and NLP datasets, PyTorch 1.1, and more...

Hi folks,

This week in deep learning we bring you the latest updates from Facebook’s F8 conference, Google’s I/O conference, and Microsoft Azure’s machine learning toolkit.

We’ve also found new datasets spanning famous landmarks and text, a practical guidebook to building human-centered AI products, a larger version of GPT-2, and some new tools for PyTorch from Facebook.

As always happy reading and hacking. If you have something you think should be in next week's issue, find us on Twitter: @dl_weekly.

Until next week!


AI at Facebook’s F8 Conference [Facebook]

Last week at Facebook’s F8 conference, AI featured heavily in the company’s strategy to understand the content uploaded to its platform so it can surface the best content to users and remove the worst.

Facebook launches PyTorch 1.1 with TensorBoard support [VentureBeat]

The latest release of PyTorch brings TensorBoard support, JIT support, and more.

Microsoft launches a drag-and-drop machine learning tool [TechCrunch]

Microsoft created a new interface for Azure’s automated machine learning tool that brings a drag-and-drop visual interface to users. Is this the final form of, the visual deep learning platform Microsoft purchased last year?

The 8 biggest announcements from the Google I/O 2019 keynote [The Verge]

Once again, AI featured heavily at Google’s I/O conference. Android got a number of new capabilities including full automatic speech recognition to caption videos and transcribe audio in real-time and new vision features in Lens to understand and extract information from scenes.

Using AI to predict breast cancer and personalize care [MIT News]

MIT/MGH's image-based deep learning model can predict breast cancer up to five years in advance.


Illustrated Transformers

Fantastic explanation of transformer networks with great visualizations.

Stegastamp: Invisible Hyperlinks in Physical Photographs

Neat method for turning any image into a QR code.

Google’s People + AI Guidebook

A guidebook to help you build human-centered AI products. It’ll enable you to avoid common mistakes, design excellent experiences, and focus on people as you build AI-driven applications.


VisualData: A curated list of computer vision datasets.

An extensive, searchable list of CV data.


“An open source effort to reproduce OpenAI’s WebText dataset. 38GB of text data from 8,013,769 documents.”

Announcing Google-Landmarks-v2: An Improved Dataset for Landmark Recognition & Retrieval

“A completely new, even larger landmark recognition dataset that includes over 5 million images (2x that of the first release) of more than 200 thousand different landmarks (an increase of 7x)”

Libraries & Code

[Github] SeldonIO/alibi: Algorithms for monitoring and explaining machine learning models

Alibi is an open source Python library aimed at machine learning model inspection and interpretation. The initial focus on the library is on black-box, instance based model explanations.

OpenAI 345M GPT-2 Weights

OpenAI has decided to release a larger version of it’s GPT-2 model. The largest version has been provided to a select group of people to determine adverse impacts of release.

Open-sourcing Ax and BoTorch: New AI tools for adaptive experimentation

“Ax is an accessible, general-purpose platform for understanding, managing, deploying, and automating adaptive experiments. BoTorch, built on PyTorch, is a flexible, modern library for Bayesian optimization, a probabilistic method for data-efficient global optimization.”


Google’s latest AI art project turns your face into a ‘poem portrait’ [The Verge]

An Instagram filter with AI-generated poetry.

Papers & Publications

Controllable Artistic Text Style Transfer via Shape-Matching GAN

Abstract: “....In this paper, we present the first text style transfer network that allows for real-time control of the crucial stylistic degree of the glyph through an adjustable parameter. Our key contribution is a novel bidirectional shape matching framework to establish an effective glyph-style mapping at various deformation levels without paired ground truth. Based on this idea, we propose a scale-controllable module to empower a single network to continuously characterize the multi-scale shape features of the style image and transfer these features to the target text….”

NeuronBlocks -- Building Your NLP DNN Models Like Playing Lego

Abstract: “....In this paper, we introduce NeuronBlocks, a deep neural network toolkit for natural language processing tasks. In NeuronBlocks, a suite of neural network layers are encapsulated as building blocks, which can easily be used to build complicated deep neural network models by configuring a simple JSON file. NeuronBlocks empowers engineers to build and train various NLP models in seconds even without a single line of code…”