News in artificial intelligence and machine learning

Revue
 
Week of 16th November 2015. I've added a short synopsis/commentary to each piece, let me know if that
 
November 20 - Issue #2

Nathan Benaich

An analytical digest of artificial intelligence and machine learning news from the technology industry, research lab and venture capital market.

Week of 16th November 2015. I’ve added a short synopsis/commentary to each piece, let me know if that’s helpful :)

Technology news, trends and opinions
TensorFlow and monetising intellectual property. This is well worth a read. Google has three big strengths: data, infrastructure and ML. Open sourcing the latter makes today and tomorrow’s developers adopt, grow up with and improve a Google core competency while gravitating towards using their paid complementary services (hosting, data…). Classic trojan horse approach to winning market share. Thanks @jesolem for sharing!
Manufacturers must learn to behave more like tech firms. A case for industrial manufacturers to adopt IoT/M2M communication to capture valuable information pertaining to their processes and build platforms to extract value from this data. Wot.io (data integration), ThingWorx (app platform), Mnubo and Cubitic are working on this.
How IBM’s Watson went from ‘Jeopardy’ champ to vital startup partner. This piece walks you through the sales Watson sales process and four example use cases: engagement, discovery, policy enforcement, and decision support.
Sentient Technologies partner with Shoes.com for visual search. If shoe shopping isn’t your thing, have a play with Google Photos on iOS/Android to grasp the power of deep learning to enable visual search.
Hipmunk Hello uses artificial intelligence to plan your next vacation for you. Is this the next step onwards from predictive pricing forecasts? In the low margin/high volume online travel agent business, perhaps this could optimise conversions.
a16z Podcast: Artificial Intelligence and the ‘Space of Possible Minds’. A show with Murray Shanahan (Prof. of Cognitive Robotics at Imperial College London), Tom Standage (Deputy Editor of The Economist, responsible for AI issue a few months back) and Azeem Azhar (ex-PeerIndex).
Will artificial intelligence bring us utopia or destruction? Lengthy piece in the New Yorker on work and future outlook of Oxford philosopher and author of Superintelligence, Nick Bostrom.
Research, development and resources
Deep Neural Decision Forests, Microsoft Research and CMU. Here the authors combine random forests (a popular and effective classification method for high dimensional data) with neural networks and use a differential technique to optimise training. They manage to reduce the error rate when testing on ImageNet.
Medical Image Deep Learning with Hospital PACS Dataset, Mass. General Hospital/Harvard Medical School. Using CNNs to classify medical images is a massive opportunity (see: Enlitic, Zebra Medical, MetaMind, IBM/Merge $1bn deal). The authors find the optimal training data set size required to accurately classify computed tomography images. Working on a similar problem too? Let me know!
Conditional Computation in Neural Networks for Faster Models, McGill University. Dropout is a popular method to improve the speed of training deep neural networks by randomly dropping nodes in the network to prevent overfitting. The authors present an approach to learn effective dropout policies instead of conduct them at random.
Transfer Learning for Speech and Language Processing, Tsinghua University. A review of the state of the art for applying auxiliary resources (e.g. labels/data/models from one source) to a target task. For example, one could use reviews for films on Amazon product pages to tune model parameters when the target task is predicting IMDB rankings. Indico.io is a company working in this space.
For any avid Go players, try your hand playing against this CNN trained on a dataset of >80k games by professional players.
Venture capital financings
DeepGenomics, a startup spun out of the University of Toronto that made press headlines a few months ago for their deep learning approach to studying the implications of genetic variants on disease, raised a $3.7m seed round from True Ventures and Bloomberg Beta.
TuringSense, a developer of multi-sensor wearable technology for high-speed fully-body motion analysis, raised a $3m seed round from Angel Plus and others in SF.
—————————————————————————————
Anything else catch your eye? Drop me a line on @nathanbenaich. Playfair Capital is actively seeking out entrepreneurs building companies that change the way we live, work and play using deep learning, machine learning, artificial intelligence, computer vision, speech and NLP.

Do you enjoy receiving this? Please share it:
     
Carefully curated by Nathan Benaich with Revue
If you were forwarded this newsletter and you like it, you can subscribe here.
If you don't want these updates anymore, please unsubscribe here.