For all the previously introduced layers, the same output will be generated if we repeat the same input several times. For instance, if we have a linear layer with f(x) 2.x. Each time we ask to predict f(3) we will get 6. Now imagine we are training an algorithm to detect repetitions, so we want that F(3) 0 for the first time (no repetition detected), then we would like to get F(3) 1 for the second time. We can't achieve this behavior with non-recurrent layers.

In Machine Learning as a Service, a provider trains a deep neural network and provides many users access to it. However, the hosted (source) model is susceptible to model stealing attacks where an adversary derives a surrogate model from API access to the source model. For post hoc detection of such attacks, the provider needs a robust method to determine whether a suspect model is a surrogate of their model or not. We propose a fingerprinting method for deep neural networks that extracts a set of inputs from the source model so that only surrogates agree with the source model on the classification of such inputs. These inputs are a specifically crafted subclass of targeted transferable adversarial examples which we call conferrable adversarial examples that transfer exclusively from a source model to its surrogates.

In connection with the grant of the Canada First Research Excellence Fund (IVADO), pending final budgetary approval, the Department of Computer Science and Operations Research is seeking applications for one full-time tenure-track position (as assistant or associate professor), in areas related to machine learning and its applications (e.g., healthcare, natural language processing, computer vision, robotics). The appointed candidates will have the possibility to join Mila, the largest academic research group in the area of deep learning, worldwide, and become an academic member of the IVADO data science community. Through their membership with Mila, the candidate(s) could also be nominated for a prestigious Canada CIFAR AI (CCAI) Chair. The appointed candidate will be expected to teach undergraduate and graduate courses, supervise graduate students, be active in research, publishing, and the diffusion of knowledge, and contribute to the activities of the University. Mila researchers pioneered deep learning more than a decade ago and remain an academic pillar of this approach to artificial intelligence.

Based on Apache OpenWhisk, IBM Cloud Functions is a Functions as a Service (FaaS) platform that makes it easy to build and deploy serverless applications. In this tutorial, you'll build a serverless application using IBM Cloud Functions that monitors the content of a Cloud Object Storage bucket and analyzes the content of images that are uploaded to the bucket by a human or an automated process. For illustrative purposes, analysis is performed by a deep learning microservice from the Model Asset eXchange and analysis results are stored as JSON files in the same bucket. You can easily adapt the outlined approach to take advantage of hosted cognitive services, such as those provided by IBM Watson, and to store results in a NoSQL datastore like Cloudant or a relational database. By completing this introductory tutorial, you learn how to monitor a Cloud Object Storage bucket for changes (new objects, updated objects, or deleted objects) using Cloud Functions and how to use deep learning microservices from the Model Asset eXchange to automatically analyze those objects in near real time.

To write this script we enter the first sentence into GPT-2 artificial intelligence. Every single word from then on was written by AI. This is how GPT-2's website describes it: While GPT-2 was only trained to predict the next word in a text, it surprisingly learned basic competence in some tasks like translating between languages and answering questions. That's without ever being told that it would be evaluated on those tasks. How do you say "hello" in spanish?

The personnel changes at Alphabet continue, this time with Mustafa Suleyman -- one of the three co-founders of the company's influential AI lab DeepMind -- moving to Google. Suleyman announced the news on Twitter, saying that after a "wonderful decade" at DeepMind, he would be joining Google to work with the company's head of AI Jeff Dean and its chief legal officer Kent Walker. The exact details of Suleyman's new role are unclear but a representative for the company told The Verge it would involve work on AI policy. The move is notable, though, as it was reported earlier this year that Suleyman had been placed on leave from DeepMind. Some speculated that Suleyman's move was the fallout of reported tensions between DeepMind and Google, as the former struggled to commercialize its technology. Although DeepMind has achieved a number of research milestones in the AI world, most notably the success of its AlphaGo program in 2016, the lab has also recorded significant financial losses.

Points clouds of an object, in this case a plane, are broken apart; if a neural network can be trained to reassemble the object, it can develop a capacity to predict the parts of the object without having the point clouds labeled, a form of self-supervised learning. This Sunday begins the annual NeurIPS conference on artificial intelligence, one of the most prominent gatherings in the field. It's being held this year in Vancouver, British Columbia. As always, the first thing you want to do, even before the conference starts, is to look over the accepted research papers. All the papers are posted on the NeurIPS Web site, so you can take a tour of this year's research before going, or even without going at all.

Earlier this year, I posted about our project KataGo and research to improve self-play learning in Go, with an initial one-week run showing highly promising results. Several months later in June, KataGo performed a second, longer 19-day run with some major bugfixes and minor optimizations. Starting from scratch and with slightly less hardware than before, up to 28 V100 GPUs, it reached and surpassed the earlier one-week run in barely more than the first three days. By the end of the 19 days, it had reached the strength of ELF OpenGo, Facebook AI Research's multi-thousand-GPU replication of one of AlphaZero's runs - equating to roughly a factor of 50 reduction in computation required. This version of KataGo has also been released to the Go player community for several months now.

Deep learning, an advanced form of artificial intelligence (A.I.), is all around us, and it's growing increasingly ingrained in how we live and work. Deep learning tech is "the brains" behind the automated traffic control on our city streets, our sophisticated translation systems, and the fast, accurate facial recognition at our airports. Additionally, deep learning can also vastly improve both shopper experience and retailers' sales, on and offline. Technically speaking, deep learning uses artificial neural networks, software constructs that are inspired by the biological structure of the human brain. Neural networks are excellent at rapidly processing and understanding unstructured data such as video, images, audio or large amounts of text without requiring intervention from human engineers, who could not possibly keep pace with the A.I. Neural network tech particularly shines at visual perception, natural language understanding and the ability to predict behaviors such as online shoppers' desires and purchase intent.

The findings, published in Scientific Reports, mark the first time scientists have used machine learning tools for rapid quantitative and qualitative cell analysis in basic science. "This new test will allow investigators to measure NETosis in different diseases and to test drugs that may inhibit or promote the process," said senior author Leslie Parise, PhD, professor and chair of the UNC Department of Biochemistry and Biophysics. When foreign invaders such as viruses or bacteria enter our bodies, white blood cells rush in to fight the invaders in various ways. One type of white cell, the neutrophil, expels its DNA into the bloodstream to trap bacteria and viruses and aid in their killing to prevent infections. This neutrophil DNA has a net-like appearance and is called Neutrophil Extracellular Traps, or NETs.