News in artificial intelligence and machine learning
From 1st thru 17th March. Referred by a friend? Sign up here. Want to share? Give it a tweet.
As promised, I've produced a comprehensive analysis of the state of the AI market in 2015 for the benefit of entrepreneurs, investors and corporates.
Part 1 focuses on investments and Part 2 focuses on exits. I'm eager to get your thoughts and hear of additional questions that come up!
Technology news, trends and opinions
Much of the AI world has been laser focused on DeepMind's AlphaGo landmark 4-1 victory over Lee Sedol, an elite Go player holding 18 international titles (2nd most in the world). This comes 19 years after IBM's Deep Blue mastered Garry Kasparov in chess and 5 years after IBM's Watson conquered the Jeopardy! quiz show. Indeed, experts didn't believe Go could be tackled by software for another 10 years due to its intrinsic complexity. AlphaGo even earned honorary 9-dan status from the Korean Go Association.
Why the US is buying up so many UK artificial intelligence companies, Vice Motherboard. A profile on M&A activity in the last year (DeepMind, VocalIQ, SwiftKey and other) and what it means for the UK technology ecosystem and academia.
Project AIX: Using Minecraft to build more intelligent technology. Microsoft announce their work using Minecraft, which was developed by Swedish game studio Mojang (acquired by Microsoft in 2014 for $2.5bn), as an environment for advancing reinforcement learning research. For a primer on this space, watch this recent talk video by Pieter Abbeel from UC Berkeley.
Rolling Stone publish a (long) two part special report on The Artificial Intelligence Revolution (part 1 and part 2) exploring robotics, Tesla, DeepMind, Facebook, self driving cars, existential risk and the future of society.
The Director of Applied ML at Facebook runs a fascinating Q&A session on Quora covering ML applications at FB, which of data/infrastructure/algorithms is most important, DL hype/reality, OpenAI, academia vs. industry work and more!
Andrew Ng of Baidu, a strong proponent and developer of driverless cars, makes his case for requisite infrastructure and policy to make this a reality by 2018.
Bill Gates ran an Ask Me Anything on Reddit last week where he sided with the fears expressed by Elon Musk and Stephen Hawking re: AI without regulation. Others have made a similar case following AlphaGo, while this piece portrays the bear case re: tech unemployment.
Research, development and resources
CryptoNets: Applying Neural Networks to Encrypted Data with High Throughput and Accuracy, Microsoft Research and Princeton University. The authors demonstrate that neural networks can be used in tandem with homomorphic encryption (HE) to produce highly accurate, high throughput predictions on optical character recognition tasks while the underlying data remains encrypted. This approach is paving the way for computations on sensitive data to occur in the cloud (vs. the device) given that it remains encrypted in the process. Further explanations here.
One-Shot Generalization in Deep Generative Models, Google DeepMind. Here, the team seeks to recapitulate our ability as humans to encounter a new concept (with one or few examples) and generalise to create new versions of the concept. The core solution is to describe the probabilistic process by which an observed data point (e.g. a handwritten "8") can be generated. The authors use a deep neural network to specify this probabilistic process and show that their models can generate written characters and faces.
Diagnosing Heart Diseases with Deep Neural Networks, a step-by-step guide to solving the Second Data Science Bowl on Kaggle to automatically measure end-systolic and end-diastolic volumes in cardiac MRIs. This process otherwise takes 20 mins to be done manually by a cardiologist and thus an automated solution can empower doctors to save more lives.
Item2Vec: Neural Item Embedding for Collaborative Filtering, Microsoft and Tel Aviv University. The authors extend Word2Vec, which is used to map the semantic relationship between words in a low dimensional vector space, to item-based product recommendations. Specifically, this approach works well when the number of users far outnumbers a product catalogue (e.g. music streaming) or when user-item relations aren't available as users browse e-commerce pages anonymously.
Sequential Short-Text Classification with Recurrent and Convolutional Neural Networks, MIT. The authors use a neural network approach to classify short text that incorporates a knowledge of the sequential context of these sentences. They demonstrate its performance on predicting dialog.
50 machine learning role interview questions and associated resources to brush up your knowledge.
Venture capital financings and exits
$43M of announced deal making in the last two weeks, including:
Mapillary raises a $8M Series A from Atomico, Sequoia, Playfair Capital, Wellington Partners and others to map the world with photos and street level visualisations.
Endgame raises a $6M Series C extension from Bessemer to help detect and prevent malicious activity within critical enterprise infrastructure.
Arterys raises a further $5M to close out a $12M Series A from GE Ventures, Norwich Ventures and Emergent Medical Partners for their MRI-based AI-assisted cardiovascular disease screening product.
Mariana raises a $2M seed round from Blumberg Capital and angels for their B2B lead generation product that leverages deep learning.
Framed Data, a user behaviour analytics product used to predict churn, was acqui-hired by Square. The YC company raised $2M in seed capital from a large number of Silicon Valley angels including Alexis Ohanian, Garry Tan, as well as Google Ventures. Framed will be shut down and the 6 person team will join Square Capital.
Anything else catch your eye? Just hit reply! I’m actively looking for entrepreneurs building companies that build/use AI to rethink the way we live and work.