2019 SO FAR: WHAT HAPPENED IN AI/ML?

June 2019

Semantic Sanity

“Semantic Sanity is an adaptive ArXiv feed inspired by ArXiv Sanity. It uses AI to quickly learn what papers you care about reading and recommends the latest research to help you stay up to date in Computer Science.”

ICML 2019: Thirty-sixth International Conference on Machine Learning

Google, ETH Zurich, MPI-IS, Cambridge & PROWLER.io Share Best Paper Honours

ICML announced the recipients of the Best Paper Awards:

Research: AI is a ‘welcome boost’ that could add millions to GDP

Recent research underlines the importance of AI for business companies. A new study predicts that AI overall could provide a 22 percent GDP boost worth millions to the UK economy. [Read more]

May 2019

Reconstruction of the original MNIST dataset

Read the paper Cold Case: The Lost MNIST Digits

The MNIST dataset is derived from the NIST database. A research team has reconstructed the derivation process that is accurate enough to serve as a replacement for the MNIST dataset. They trace each MNIST digit to its NIST source including its metadata such as writer identifier, partition identifier, … and reconstruct the complete MNIST test set with 60,000 samples instead of 10,000.

 

Tweet source: Yann LeCun

Facebook has upgraded its AI services

Facebook launched PyTorch 1.1 with TensorBoard support and gave its just-in-time (JIT) compiler an upgrade.[Read more]

G-7 Science Academies release statements on AI

The national science academies of the G-7 countries issued joint statements to their respective governments informing ongoing policymaking as well as discussions for the G-7 summit coming up in August 2019. [Read more]

ML algorithm predicts heart attacks with 90% accuracy

A machine learning algorithm claims to predict heart attacks and death from heart disease with 90 percent accuracy which is higher than the accuracy achieve through human practitioners. [Read more]

 

April 2019

Generating adversarial patches against YOLOv2

Fooling automated surveillance cameras: adversarial patches to attack person detection.

[Paper] [YouTube]

Tesla Autonomy Day

Watch the full video:

OpenAI Five defeats Dota 2 world champions @OGesports

In a series of live competitions held in San Francisco the five-bot team OpenAI Five defeated the reigning Dota 2 world champion team OG. The AI won two matches back-to-back, settling the best-of-three tournament.

 

Ian Goodfellow changes from Google Brain to Apple

Goodfellow, who is best known for inventing generative adversarial networks (GANs), is hired to become a director of machine learning for a mysterious “special projects” group. With this employment Apple is continuing to hire key AI talent from its rival Google. [Read more]

 

March 2019

Yann LeCun, Geoffrey Hinton and Yoshua Bengio win Turing Award

Yann LeCun, Geoffrey Hinton and Yoshua Bengio – three Pioneers in Artificial Intelligence won the Turing Award for their work on neural networks. They will share $1 million for what many consider the Nobel Prize of computing.

Source: The New York Times

NVIDIA introduces Jetson Nano – a small AI computer

NVIDIA announced the Jetson Nano, a small AI computer that delivers 472 GFLOPS of compute performance for running modern AI workloads and is highly power-efficient, consuming only 5 watts. It comes in two versions — the $99 devkit for developers, makers and enthusiasts and the $129 production-ready module for companies looking to create mass-market edge systems. [Read more]

OpenAI created OpenAI LP, a for-profit company

OpenAI restructured from non-profit to “capped-profit” and explained its decision in a blog post:

We’ll need to invest billions of dollars in upcoming years into large-scale cloud compute, attracting and retaining talented people, and building AI supercomputers.

We want to increase our ability to raise capital while still serving our mission, and no pre-existing legal structure we know of strikes the right balance. Our solution is to create OpenAI LP as a hybrid of a for-profit and nonprofit—which we are calling a “capped-profit” company.

Read more on TC

  • The company justifies this rather high profit “cap” by saying that if it succeeds in creating a working artificial general intelligence (AGI is a poorly defined concept that is nonetheless perhaps the Holy Grail of current AI research), “we expect to generate orders of magnitude more value than we’d owe to people who invest in or work at OpenAI LP.”

Original blog post on OpenAI

TensorFlow 2.0 Alpha release

The TF 2.0 Alpha release is available now. This is an early version meant to share with users what the TensorFlow 2.0 API will be like, to gather feedback, and to identify and fix issues. Below are some of the key enhancements:

  • Eager execution as a central feature of 2.0. It aligns users’ expectations about the programming model better with TensorFlow practice and should make TensorFlow easier to learn and apply.
  • Keras tightly integrated with the TensorFlow ecosystem, and has support for Eager execution, tf.data API, tf.distribute.MirroredStrategy for multi-GPU training, TensorBoard visualization, and TF Lite and TF.js conversion.
  • Starter list of TF-Hub models loadable in TF 2.0.
  • Autograph making it easier to write models with custom control flow ops and getting graph performance with tf.function.

Read more: TF 2.0 Roadmap

GauGAN Turns Doodles into Photorealistic Landscapes

NVIDIA Research developed a GAN-based tool that turns rough doodles into photorealistic landscapes. The tool leverages generative adversarial networks to convert segmentation maps into lifelike images.

New to Microsoft 365

Nice ML usecase for automatic data entry: You can now take a picture of a printed data table and automatically convert it into a fully editable Excel spreadsheet. Microsoft is rolling out to Android users, coming soon to iOS.

http://msft.social/kH8GuB

Google open sourced GPipe, an Open Source Library for Efficiently Training Large-scale Neural Network Models

GPipe is a distributed machine learning library that uses synchronous stochastic gradient descent and pipeline parallelism for training, applicable to any DNN that consists of multiple sequential layers. GPipe allows researchers to easily deploy more accelerators to train larger models and to scale the performance without tuning hyperparameters.

https://ai.googleblog.com/2019/03/introducing-gpipe-open-source-library.html

OpenAI released Activation Atlases

Activation Atlases: a new technique for visualizing what interactions between neurons can represent. Activation atlases build on feature visualization, a technique for studying what the hidden layers of neural networks can represent.

💻Blog: https://blog.openai.com/introducing-activation-atlases/

📝Paper: https://distill.pub/2019/activation-atlas

🔤Code: https://github.com/tensorflow/lucid/#activation-atlas-notebooks

🗺️Demo: https://distill.pub/2019/activation-atlas/app.html

 

February 2019

Andrew Ng launched his non-technical AI for Everyone course, focusing on:

  • The meaning behind common AI terminology, including neural networks, machine learning, deep learning, and data science
  • What AI realistically can – and cannot – do
  • How to spot opportunities to apply AI to problems in your own organization
  • What it feels like to build machine learning and data science projects
  • How to work with an AI team and build an AI strategy in your company
  • How to navigate ethical and societal discussions surrounding AI

https://www.coursera.org/learn/ai-for-everyone

DeepMind and Google: Machine Learning can increase the value of wind energy

Since last year, a joint DeepMind and Google project to apply ML to 700 MW of wind power in the central US has so far boosted the value of wind energy by ~20%.

A neural net trained on weather forecasts & historical turbine data predicts wind power output 36 hours ahead of actual generation. Based on these, DeepMind’s model recommends optimal hourly delivery commitments to the power grid 24 hours in advance.

Alt Text

https://deepmind.com/blog/machine-learning-can-boost-value-wind-energy/

Lingvo

In support of the research community and to encourage reproducible research efforts, Google AI has open-sourced Lingvo, a general deep learning TensorFlow framework with a focus on language-related tasks like machine translation, speech recognition, and speech synthesis.

Alt Text

https://goo.gl/4da5Zj

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s