A week's news in artificial intelligence and machine learning
10th thru 18th December. Have an awesome weekend and happy holidays!
Referred by a friend? Sign up here. Want to share? Give it a tweet.
Technology news, trends and opinions
Inside Deep Dreams: how Google made its computers go crazy. A piece that dives into the story of how Deep Dreams was born as an experiment in the early hours of the morning.
OpenAI launches as a nonprofit to make AI that “benefits humanity”. Backed by several tech billionaires and SV institutions, OpenAI's goal is to advance AI research in a way that is unconstrained by the need to generate financial return. In today's world, the technology incumbents of the 2000's - FB, GOOG, BIDU and others - have the upper hand because they control the largest datasets, have the most sophisticated (and often proprietary) infrastructure, the largest research budgets and access to talent. Given that academia can hardly compete, OpenAI is a worthy shot to equalise the playing field. However, I question who has first dibs over the commercialisation of IP generated and it's definitely an assumption that financial incentive alone is what's stopping AI from being used for good.
The First Person to Hack the iPhone Built a Self-Driving Car. In His Garage, a captivating story of a prodigy technologist who is proving that access to commoditised hardware, compute capacity on GPUs and deep learning can enable innovations that could't possibly be tackled a few years ago.
Facebook Joins Stampede of Tech Giants Giving Away Artificial Intelligence Technology, this time relating to its server design.
The current state of machine intelligence 2.0, an updated landscape
Google Raids Threaten Canada's Lead in Artificial Intelligence, some of the top AI research programs are Canadian and they've been pillaged over the years.
Research, development and resources
MovieQA: Understanding Stories in Movies through Question-Answering, Karlsruhe Institute of Technology, Massachusetts Institute of Technology and University of Toronto. The authors create a dataset of 7,702 questions about 294 movies from full-length movies, subtitles, scripts and plots. This is used to create a question-answering dataset to help push research on automatic understanding of high-level semantics underlying human actions. For example, enabling machines to understand the motivation, intent and emotion behind human communication. Layman write up here.
Visual Genome: Connecting Language and Vision Using Crowdsourced Dense Image Annotations, Stanford University. Paper yet to be released, but you can browse the dataset that connects structured image concepts to language.
Andrew Ng presents at the GPU Technology Conference 2015 on what's next in deep learning.
Deep Active Object Recognition by Joint Label and Action Prediction, UCSD. This paper presents a model for deep active object recognition that jointly predicts the label (to describe image contents) and the next best action on an input image. The system uses reinforcement learning to teach the CNN to output action values with minimal error. The applications pertain to robotics, where an agent analyses its environment and decides what to do.
An Empirical Comparison of Neural Architectures for Reinforcement Learning in Partially Observable Environments, Vrije Universiteit Brussel. Reinforcement learning is all about iterative policy learning to maximise a reward signal in a stochastic environment. In most real-world situations, however, an agent makes decisions with incomplete information of its environment. Here, memory of past observations and actions can help complete the understanding of a current environment. The authors evaluate three ways of incorporating memory into recurrent neural networks and reinforcement learning.
Microsoft researcher leads team to win Marr prize for outstanding computer vision research for their paper, Deep Neural Decision Forests.
Researchers demonstrate how the brain can handle so much data, Georgia Institute of Technology.
Venture capital financings and exits
$80m worth of financing during this period (median deal size of $5m), including the following noteworthy transactions:
Blue River Technology: $17m Series B from Data Collective, Khosla and Innovation Endeavors (Eric Schmidt). The company produces tractor-towed robots that measures plant health and treat weeds to grow crop yields.
WorkFusion: $11.3m Series C for its platform that combines the best of human and machine intelligence to increase workforce productivity and engagement, automate predictable manual work, and improve data quality.
Preferred Networks: $8.2m direct investment from Toyota to further its application of deep learning and edge-heavy computing to autonomous vehicles.
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.