The MachineLearning EngineerIssue #7

This week in Issue #7:

AI vs Humans Starcraft II edition, Tensorflow 2.0 features, neural architecture search, distributed ML architectures, the impact of learning rate, R-CNNs for car parking, ML Jobs, Machine Learning Conferences and more!

Support the ML Engineer!

Forward the email, or share the online version on 🐦 Twitter, 💼 Linkedin and 📕 Facebook!

If you would like to suggest articles, ideas, papers, libraries, jobs, events or provide feedback just hit reply or send us an email to a@ethical.institute!

AI vs Human - Starcraft II Edition

The DeepMind team did it again, this time taking the challenge to the game of Starcraft II. DeepMind covered it live in Youtube last Thursday (check the short video here). [SPOLIER ALERT] In summary, DeepMind started by introducing the new AI called AlphaStar (a recurrent-network+attention-like reinforcement learning model). DeepMind then brought in two top players known as “Mana” and “TLO” to discuss the 10 matches where AlphaStar beat them both. “Mana” then played a re-match live, which he won. It was a truly interesting game, primarily as Mana used a lot of the knowledge gathered from his previously lost games. The video is quite long, but there are a lot of interesting parts - one of the most interesting points were the limitations that were introduced to AlphaStar to make sure games against humans were “fair”. For anyone curious to learn more, the core DeepMind team hosted an AMA in Reddit.

From zero to distributed

“If an ML model makes a prediction in Jupyter, is anyone around to hear it?”. Great short tutorial that covers the benefits of horizontally-scalable architectures over flask-wrapped-like model deployments This is a very good (super) small scale version of our conversation last week introducing Uber’s Michaelangelo framework. This tutorial builds a queue-enabled machine learning serving system using MLQ for the queue interface, and the tensorflow.js library to perform the inference on the front-end. Great stuff.

Tensorflow 2.0

An exciting milestone for one of the most widely adopted libraries in deep learning. In this great post released by the core team, they cover some of the high level objectives. The main point they mention is that Tensorflow 2.0 will see all the components that have been build separately brought in together as a comprehensible unified platform. This includes all its flavours, such as Tensorflow Serving, Tensorflow Lite, Keras, tensorflow JS and beyond. In this blog post, Iorni showcases key Tensorflow 2.0 (experimental) features by building a deep reinforcement model.

I heard you like machine learning

That’s why Google put machine learning on their machine learning so you can automatically learn your machine learning. Great post that provides a high level overview of Google Cloud’s AutoML and Neural Architecture Search. This week’s featured libraries focus on Neural Architecture search - check them out below, or online at our Awesome MLOps list. If you are interested to learn more about AutoML on a feature-level you can also check out the libraries we featured on our release #4.

Impact of Learning Rate

Another great post by machine learning mastery diving into the “Impact of Learning Rate on Model Performance With Deep Learning Neural Networks”. In this tutorial, Jason provides a hands-on example creating a keras multi-class classification model.

Monitoring Parking with R-CNN

Adam Geitgey proposes to automate manual annoying tasks with deep learning. In his article he walks us through on how he built an R-CNN to monitor for parking spaces so he can get notified when one is free. Great insight on how image ML tasks are often tackled by using domain knowledge to modularising the problem into multiple “layers” (e.g Detect spaces -> detect cars -> detect parking spaces that are newly empty).

MLOps = Featured OS Libraries

This week’s edition is focused on AutoML and Neural Architecture Search frameworks which fall on our Responsible ML Principle #4. The four featured libraries this week from the Awesome MLOps list are

• Autokeras - AutoML library for Keras based on “Auto-Keras: Efficient Neural Architecture Search with Network Morphism”.
• ENAS-PyTorch - Efficient Neural Architecture Search (ENAS) in PyTorch based on this paper.
• Neural Architecture Search with Controller RNN - Basic implementation of Controller RNN from Neural Architecture Search with Reinforcement Learning and Learning Transferable Architectures for Scalable Image Recognition.
• Neural Network Intelligence - NNI (Neural Network Intelligence) is a toolkit to help users run automated machine learning (AutoML) experiments.

If you know of any libraries that are not in the “Awesome MLOps” list, please do give us a heads up or feel free to add a pull request!

MLJobs = Jobs & Careers

Since last week, we started showcasing Machine Learning Engineering jobs (primarily in London for now) to help our community stay up to date with great opportunities that come up.

Junior Opportunities

• Seldon is hiring for a Machine Learning / Data Engineer in London
• Atlas ML is hiring for a Machine Learning / NLP Engineer in London
• Migacore is hiring for a Machine Learning Engineer in London
• CloudNC is hiring for a Machine Learning Engineer in London
• Babylon Health is hiring for a Machine Learning Engineer in London

Mid-level Opportunities

• QuantumBlack is hiring for a Senior Machine Learning Engineer in London
• Proportunity is hiring for a Senior Machine Learning Engineer in London

Leadership Opportunities

• Fractal Labs is hiring for a VP of Engineering in London
• Distributed is hiring for a VP of Engineering in London
• FactMata is hiring for a Head of Machine Learning in London
• Skyscanner is hiring for a Senior Director of Sciences in London

MLConf = Conferences & Events

From this week on, we will be featuring conferences that have core ML tracks (primarily in Europe for now) to help our community stay up to date with great events coming up.

Technical Conferences

• FOSDEM [02/02/2019] - Europe’s largest open source conference in Brussels, Belgium.

• We are speaking on the state of MLOps in 2019 at the HPC, Big Data & Data Science track - come say hello!

• PyCon Belarus [15/02/2019] - The 5th edition of the Python conference in Minsk, Belarus.

• We’ll be speaking on Machine Learning Explainability and Bias Evaluation.

• Machine Learning Prague [22/02/2019] - A practical conference on machine learning & DL in Prague, CZ.

Business Conferences

• Big Data & AI Tech World [12/03/2019] - AI & Big Data Business conference in London, UK.

• If you are around, do join our talk on AI Explainability.

• AI in Business Ethics [09/02/2019] - Conference on AI business and ethics in London, UK.

• Join our panel on the impact of AI in industry!

• AI Expo Global [19/04/2019] - Global conference on artificial intelligence in London, UK.

• Come join us at our talk on AI orchestration at scale.