A week's news in artificial intelligence and machine learning
From 2nd thru 10th December 2015...Expecting more headlines to trickle in following the Neural Information Processing Systems (NIPS) conference happening this week.
Technology news, trends and opinions
Why 2015 Was a Breakthrough Year in Artificial Intelligence. We've seen reductions in error rates for complex visual recognition tasks, Google adopt deep learning in to close to 2.7k projects internally, speech recognition applications become useable and more...2016 will be big.
Investing in Artificial Intelligence: a VC perspective. Here, I run through the reason why AI companies are investable, how they enter the market, what challenges operators/investors will face, the importance of involving the user in the product, the financing/exit climate and two big opportunities in healthcare and enterprise automation.
Baidu’s Self-Driving Car Takes On Beijing Traffic. Launched in collaboration with BMW, driverless cars are being testing in a top-10 most congested city. I'd love to see the effect of training ground (Beijing vs. SF/Mountain View) on performance! Longer piece in the Atlantic here.
Top VC Says Gene Editing Is Riskier Than Artificial Intelligence. Vinod Khosla makes the point that we case for research into existential risk and regulation, pointing to genome editing as riskier than AI. Let's remember that genome editing of stem cells to create genetically modified animals has been around since the '80s - in fact, this research was awarded the 2007 Nobel Prize for Medicine. Aside from the South Korean research scandal a few years ago, the technology is safely used all the time in the research lab.
Why Memory and Mimicry Are The Next Big Frontiers in Artificial Intelligence. The author makes a case to use reinforcement learning to train real-world policies that machines can action once they've analysed input data. For example, a robot must understand its surroundings, make a decision for what to do next and be rewarded/punished if the action isn't correct. A startup, Osaro, raised $3.3m this week to productise such a solution.
Software Bots: From Do-it-Yourself companion bots to AI powered software. A goodie - given that structured communication is a low hanging fruit for AI, there's a whole load of bots out there. The guys at Point Nine have published a thorough series on this topic.
New AI product launches:
SwiftKey's Symbols: an assistive keyboard for non-verbal communication.
Jukedeck MAKE: create unique, royalty-free music for your videos using AI.
Clarifai's Forevery: an app to search your photo database, similar to Google Photos.
Research, development and resources
Deep Speech 2: End-to-End Speech Recognition in English and Mandarin, Baidu Research AI Lab. Existing automatic speech recognition systems are made of numerous pipeline components that cope with the diversity of input speech and acoustics. This makes them difficult to build, maintain and extend to new languages. Here, the authors present an end-to-end system that captures large training sets (11,940h of English speech and 9,400h of Mandarin speech), trains larger networks in 3-5 days and runs computations at scale to approach or exceed human-level transcription performance.
Software Reconstructs Famous Faces from Still Images, Captures Their Unique Mannerisms, University of Washington. Presenting at the International Conference on Computer Vision next week, the team reconstruct a controllable 3D model of a person from a large photo collection that captures their 'persona' with expression-dependent textures. They can use one person's persona to control another's face (e.g. Barack Obama with George Bush's expressions).
Artists Make DeepDream Dream Deeper With Secret 'DeepUI' Tool. Using Google's DeepDream to create suitably hallucinogenic music videos - AI can certainly be creative!
The annual NIPS conference is on this week and here are notes from the deep learning session. Here are slides from Yann LeCun's vision talk.
Deep learning - a visual introduction: bitesize entry level learnings here.
University of Cambridge receives a £10m grant to set up the Leverhulme Centre for the Future of Intelligence. The proposal was authored along with the Centre for the Study of Existential Risk to examine the technical, practical and philosophical questions artificial intelligence raises for humanity in the coming century.
Venture capital financings and exits
Osaro, developing deep reinforcement learning technology applied to robotics, raised $3.3m from Peter Thiel, AME Cloud Ventures and angels.
Jukedeck, which develops software that writes unique and royalty free music, raised a £2m seed round from Cambridge Innovation Capital, Playfair Capital and Backed.vc, and wins TechCrunch Disrupt London 2015.
Anything else catch your eye? Drop me a line on @nathanbenaich. I’m actively looking for entrepreneurs building companies that build/use AI to rethink the way we live and work.