A General Framework for Uncertainty Estimation in Deep Learning

arXiv.org Machine Learning

End-to-end learning has recently emerged as a promising technique to tackle the problem of autonomous driving. Existing works show that learning a navigation policy from raw sensor data may reduce the system's reliance on external sensing systems, (e.g. GPS), and/or outperform traditional methods based on state estimation and planning. However, existing end-to-end methods generally trade off performance for safety, hindering their diffusion to real-life applications. For example, when confronted with an input which is radically different from the training data, end-to-end autonomous driving systems are likely to fail, compromising the safety of the vehicle. To detect such failure cases, this work proposes a general framework for uncertainty estimation which enables a policy trained end-to-end to predict not only action commands, but also a confidence about its own predictions. In contrast to previous works, our framework can be applied to any existing neural network and task, without the need to change the network's architecture or loss, or to train the network. In order to do so, we generate confidence levels by forward propagation of input and model uncertainties using Bayesian inference. We test our framework on the task of steering angle regression for an autonomous car, and compare our approach to existing methods with both qualitative and quantitative results on a real dataset. Finally, we show an interesting by-product of our framework: robustness against adversarial attacks.

Apple is turning Siri into a next-level Artificial Intelligence


Was Siri the secret star of the World Wide Developer's Conference Keynote? At first blush, I'd say no. There was no moment where Apple CEO Tim Cook declared it the most important platform in Apple's domain. Cook and SVP of Software Engineering Craig Federighi never ticked off all the Siri updates at once. There was no "Siri summary" screen.

Microsoft, Nvidia work to speed up AI platform powering Cortana


Thanks to artificial intelligence, we have autonomous cars, chat bots, and speech recognition. Microsoft's CNTK (Cognitive Toolkit) is one among many platforms that trains computers to learn, and it's getting an upgrade. CNTK drives the Microsoft services Cortana and Skype language translation, and it boasts more than 90 percent accuracy in speech recognition tasks. Microsoft will soon release an upgraded CNTK toolkit, and one hardware maker wants to ensure the toolkit works best on its hardware. Nvidia is partnering with Microsoft to optimize its GPU development tools for CNTK.

Is Facebook Building An Autonomous Car?


Today at the Frankfurt motor show, one of the biggest and most prestigious motor shows in the world, Sheryl Sandberg, COO of Facebook, spoke before German Chancellor Angela Merkel. Now what is Facebook and most importantly, Sheryl Sandberg doing at an automotive industry event? The obvious answer that comes to mind when one relates Facebook and the car industry is the billions of advertising dollars the industry spends on marketing and advertising. However, that does not seem to be Facebook's game plan, as highlighted by Sheryl and shown at their pavilion. Facebook seems to have a strategy of leveraging its capabilities in social marketing, AR & VR and interestingly, who would have thought of it, leveraging its advanced AI and deep learning capabilities to support the development of autonomous vehicles.

Overview of Udacity Artificial Intelligence Engineer Nanodegree, Term 1


After finishing Udacity Deep Learning Foundation I felt that I got a good introduction to Deep Learning, but to understand things, I must dig deeper. Besides I had a guaranteed admission to Self-Driving Car Engineer, Artificial Intelligence, or Robotics Nanodegree programs.