Deep neural networks have been widely adopted in recent years, exhibiting impressive performances in several application domains. It has however been shown that they can be fooled by adversarial examples, i.e., images altered by a barely-perceivable adversarial noise, carefully crafted to mislead classification. In this work, we aim to evaluate the extent to which robot-vision systems embodying deep-learning algorithms are vulnerable to adversarial examples, and propose a computationally efficient countermeasure to mitigate this threat, based on rejecting classification of anomalous inputs. We then provide a clearer understanding of the safety properties of deep networks through an intuitive empirical analysis, showing that the mapping learned by such networks essentially violates the smoothness assumption of learning algorithms. We finally discuss the main limitations of this work, including the creation of real-world adversarial examples, and sketch promising research directions.

In this article and the next, we'll discuss some of the history and basics of Deep Learning to serve as a kind of primer for the coverage of the International Conference on Machine Learning recently held in Sydney. Much of the discussion centres around recent advances in Deep Learning and the purpose here is to give some background, basic terminology and provide some context. Deep Learning is part of Machine Learning which is itself a sub-field of the field of Artificial Intelligence. It's generally accepted that there are three broad categories or types of Machine Learning: Supervised, Unsupervised and Reinforcement (these are fuzzy boundaries, other categories exist and definitions vary but we'll stick with these for now). In Unsupervised Learning we don't make any assumptions about the data we have, the processes that generated it nor any underlying structure/constraints.

A new artificial intelligence system teaches itself to recognize a range of visual and audio concepts by watching short video clips. Researchers at Google's DeepMind unit have developed an artificial intelligence (AI) system that teaches itself to recognize a range of visual and audio concepts by watching short video clips. For example, the new system can understand the concept of lawn mowing, even when it has not learned the words to describe what it is hearing or seeing. "We want to build machines that continuously learn about their environment in an autonomous manner," says University of California, Berkeley researcher Pulkit Agrawal. He notes the DeepMind project brings the field one step closer to the goal of creating AI that can teach itself by watching and listening to the world around it.

Elon Musk and more than 100 leaders and experts in artificial intelligence (AI) have come together urging the UN to commit to an outright ban on killer robot technology. An open letter signed by Musk, Google Deepmind's Mustafa Suleyman, and 114 other AI and robotics specialists urges the UN to prevent "the third revolution in warfare" by banning the development of all lethal autonomous weapon systems. The open letter, released to coincide with the world's largest conference on AI – IJCAI 2017, which is taking place in Melbourne, Australia this week – warns of a near future where independent machines will be able to choose and engage their own targets, including innocent humans in addition to enemy combatants. "Once developed, they will permit armed conflict to be fought at a scale greater than ever, and at timescales faster than humans can comprehend," the consortium writes. "These can be weapons of terror, weapons that despots and terrorists use against innocent populations, and weapons hacked to behave in undesirable ways."

This is a guide to the main differences I've found between PyTorch and TensorFlow. This post is intended to be useful for anyone considering starting a new project or making the switch from one deep learning framework to another. The focus is on programmability and flexibility when setting up the components of the training and deployment deep learning stack. PyTorch is better for rapid prototyping in research, for hobbyists and for small scale projects. TensorFlow is better for large-scale deployments, especially when cross-platform and embedded deployment is a consideration.

Today at Hot Chips 2017, our cross-Microsoft team unveiled a new deep learning acceleration platform, codenamed Project Brainwave. I'm delighted to share more details in this post, since Project Brainwave achieves a major leap forward in both performance and flexibility for cloud-based serving of deep learning models. We designed the system for real-time AI, which means the system processes requests as fast as it receives them, with ultra-low latency. Real-time AI is becoming increasingly important as cloud infrastructures process live data streams, whether they be search queries, videos, sensor streams, or interactions with users. First, Project Brainwave leverages the massive FPGA infrastructure that Microsoft has been deploying over the past few years.

ACKTR can learn continuous control tasks, like moving a robotic arm to a target location, purely from low-resolution pixel inputs (left). ACKTR (pronounced "actor") -- Actor Critic using Kronecker-factored Trust Region -- was developed by researchers at the University of Toronto and New York University, and we at OpenAI have collaborated with them to release a Baselines implementation. The authors use ACKTR to learn control policies for simulated robots (with pixels as input, and continuous action spaces) and Atari agents (with pixels as input and discrete action spaces). ACKTR combines three distinct techniques: actor-critic methods, trust region optimization for more consistent improvement, and distributed Kronecker factorization to improve sample efficiency and scalability. For machine learning algorithms, two costs are important to consider: sample complexity and computational complexity.

I research intelligence to understand what we are, and to harness it wisely.

Over the past few years, much of the progress in deep learning for computer vision can be boiled down to just a handful of neural network architectures. Setting aside all the math, the code, and the implementation details, I wanted to explore one simple question: how and why do these models work? The VGG networks, along with the earlier AlexNet from 2012, follow the now archetypal layout of basic conv nets: a series of convolutional, max-pooling, and activation layers before some fully-connected classification layers at the end. MobileNet is essentially a streamlined version of the Xception architecture optimized for mobile applications. The remaining three, however, truly redefine the way we look at neural networks.