The year 2019 was an excellent year for the developers, as almost all industry leaders open-sourced their machine learning tool kits. Open-sourcing not only help the users but also helps the tool itself as developers can contribute and add customisations that serve few complex applications. The benefit is mutual and also helps in accelerating the democratisation of ML. LIGHT (Learning in Interactive Games with Humans and Text) -- a large-scale fantasy text adventure game and research platform for training agents that can both talk and act, interacting either with other models or humans. The game uses natural language that's entirely written by the people who are playing the game.

Every week we bring to you best research papers, articles and videos that we have found interesting that week. Rubik's Code is a boutique data science and software service company with more than 10 years of experience in Machine Learning, Artificial Intelligence & Software development. Check out the services we provide. Eager to learn how to build Deep Learning systems using Tensorflow 2 and Python? Get our'Deep Learning for Programmers' ebook here!

Current rapid growth in artificial intelligence (AI) is fuelled predominantly by the rediscovery of deep learning. When it first surfaced 15 years ago, the concept was met with unfavourable conditions and largely abandoned in the aftermath. Nowadays, deep learning is back, circumstances are right and the field flourishes. The described third boom in artificial AI and subsequent tightening technological and economic competition sent ripples through various aspects of the social realm, including policymaking. Many countries began working on national AI strategies, including current leaders – China and U.S. – and Japan also followed suit.

There is one single thing that every Deep Learning practitioner agrees. Deep learning models are data inefficient. Let's start by considering the popular task of classification in Computer Vision. Take the ImageNet database as an example. It contains 1.3 million images from 1000 different classes.

Quantum physics phenomena is perhaps the most smoking subject in contemporary physical science. It takes a look at how particles in nature "meet up" and bring along their interesting properties, for example, electrical conductivity or magnetism. Nonetheless, it has been practically incomprehensible for even the most seasoned researchers to get more than a look at these unpredictable phenomena. This is a direct result of the colossal number of particles these phenomena contain (more than one billion billion in every gram) and the tremendous number of interactions between them. In the fast-paced, confusing universe of quantum science, AI's are utilized to assist scientific experts with ascertaining significant substance properties and make predictions about experimental results.

The recent deep learning hype aims to reach the Artificial General Intelligence (AGI): an AI that would express (supra-)human-like intelligence. Unfortunately current deep learning models are flawed in many ways: one of them is that they are unable to learn continuously as human does through years of schooling, and so on. Regardless of the far away goal of AGI, there are several practicals reasons why we want our model to learn continuously. Before describing a few of them, I'll mention two constraints: A real applications of these two constraints is robotics: a robot in the wild should learn continuously its environment. Furthermore due to hardware limitation, it may neither store all previous data nor spend too much computational resource.

Daniel Kahneman is winner of the Nobel Prize in economics for his integration of economic science with the psychology of human behavior, judgment and decision-making. He is the author of the popular book "Thinking, Fast and Slow" that summarizes in an accessible way his research of several decades, often in collaboration with Amos Tversky, on cognitive biases, prospect theory, and happiness. The central thesis of this work is a dichotomy between two modes of thought: "System 1" is fast, instinctive and emotional; "System 2" is slower, more deliberative, and more logical. The book delineates cognitive biases associated with each type of thinking. Subscribe to this YouTube channel or connect on: - Twitter: https://twitter.com/lexfridman

This article is featured in the new DZone Guide to Artificial Intelligence. Get your free copy for more insightful articles, industry statistics, and more! This article will illustrate the transition of the NLP landscape from a machine learning paradigm to the realm of machine intelligence and walk the readers through a few critical applications along with their underlying algorithms. Nav Gill's blog on the stages of AI and their role in NLP presents a good overview of the subject. A number of research papers have also been published to explain how to take traditional ML algorithms to the next level.

Computer simulations are invaluable tools for scientific discovery. However, accurate simulations are often slow to execute, which limits their applicability to extensive parameter exploration, large-scale data analysis, and uncertainty quantification. A promising route to accelerate simulations by building fast emulators with machine learning requires large training datasets, which can be prohibitively expensive to obtain with slow simulations. Here we present a method based on neural architecture search to build accurate emulators even with a limited number of training data. The method successfully accelerates simulations by up to 2 billion times in 10 scientific cases including astrophysics, climate science, biogeochemistry, high energy density physics, fusion energy, and seismology, using the same super-architecture, algorithm, and hyperparameters. Our approach also inherently provides emulator uncertainty estimation, adding further confidence in their use. We anticipate this work will accelerate research involving expensive simulations, allow more extensive parameters exploration, and enable new, previously unfeasible computational discovery.

Recent years have witnessed the trend that machine learning, especially deep learning, is being increasingly used in the application of PET imaging. Various t ypes of machine learning networks have been borrowed from computer vision field and adapted to speci fic clinical tasks for PET quantitative imaging. As reviewed in this paper, the most common applicat ions are PET AC and low-count PET reconstruction. It is also an emerging field since all of thes e reviewed studies were published within five years. With the development in both machine learning alg orithm and computing hardware, more learning-based methods are expected to facilitate the clin ical workflow of PET imaging with more potential quantification application. In addition to PET AC and low-count reconstruction, there ar e other topics in PET imaging where machine learning can be exploited. For example, high resolu tion PET has great potential in visualizing and accurately measuring the radiotracer concentrat ion in structures with dimensions of millimeter, while it is subject to the partial volume e ffect due to the limited spatial discriminating ability of scanner.[