It's been 7 years since my first interview with Gary Marcus and I felt it's time to catch up. Gary is the youngest Professor Emeritus at NYU and I wanted to get his contrarian views on the major things that have happened in AI as well as those that haven't happened. Prof. Marcus is an interesting interviewee not only because he is an expert in the field but also because he is a skeptic on the current approaches and progress towards Artificial General Intelligence but an optimist that we will eventually figure it all out. I can honestly say that I have learned a lot from Gary and hope that you will too. During this 90 min interview with Gary Marcus we cover a variety of interesting topics such as: Gary's interest in the human mind, natural and artificial intelligence; Deep Mind's victory in Go and what it does and doesn't mean for AGI; the need for Rebooting AI; trusting AI and the AI chasms; Asimov's Laws and Bostrom's paper-clip-maximizing AI; the Turing Test and Ray Kurzweil's singularity timeline; Mastering Go Without Human Knowledge; closed vs open systems; Chomsky, Minsky and Ferrucci on AGI; the limits of deep learning and the myth of the master algorithm; the problem of defining (artificial) intelligence; human and machine consciousness; the team behind and the mission of Robust AI.

Heather Gorr holds a Ph.D. in Materials Science Engineering from the University of Pittsburgh and a Masters and Bachelors of Science in Physics from Penn State University. Since 2013, she has supported MATLAB users in the areas of mathematics, data science, deep learning, and application deployment. Prior to joining MathWorks, she was a Research Fellow, focused on machine learning for prediction of fluid concentrations. Johanna Pingel joined the MathWorks team in 2013, specializing in Image Processing and Computer Vision applications with MATLAB. She has a M.S. degree from Rensselaer Polytechnic Institute and a B.A. degree from Carnegie Mellon University.

A convex optimization problem is a problem where all of the constraints are convex functions, and the objective is a convex function if minimizing, or a concave function if maximizing. A non-convex function "curves up and down" -- it is neither convex nor concave. However note that this function is convex from -pi to 0, and concave from 0 to pi. If the bounds on the variables restrict the domain of the objective and constraints to a region where the functions are convex, then the overall problem is convex. Linear functions are convex, so linear programming problems are convex problems.

In this article, we list down 6 trending jobs in machine learning one can apply. Responsibilities: The responsibilities include developing highly scalable classifiers and tools leveraging machine learning, data regression and rule-based models, deep learning, create language models from petabytes of text data in different languages, suggest, collect and synthesize requirements and innovate to create next-generation feature sets. The candidate will work as part of the product team to implement algorithms that power user and developer-facing products reaching out to millions of users, adapt standard machine learning methods to best exploit modern parallel environments. Prerequisites: The candidate must have strong background in one or more of Machine Learning, Artificial Intelligence, Pattern Recognition, Natural Language, Deep Learning, DNNs, large scale Data Mining, experience with scripting languages such as Perl, Python, PHP, and shell scripts, experience with recommendation systems, targeting systems, ranking systems or similar systems, experience with any of Hadoop/Hbase/Pig or MapReduce/Bigtable or R/Matlab/AzureML or similar technologies. Responsibilities: The responsibilities for a Machine Learning Engineer – Lead include building common ML capabilities used across Corporate based on machine learning models, automate and streamline existing processes, procedures, and toolsets.

The concept of recognising gestures using hands and/or other body parts is based on three layers: Detection, Tracking and Recognition. We use special interfaces that can capture these movements, and later use computer vision technology & deep learning algorithms to understand the underlying pattern. Today, they are several gesture-interface products in the market made by Big Tech giants like Intel, Apple & Google for applications in home automation, shopping, virtual/augmented reality gaming, consumer electronics and navigation among others.

Feature visualisation is an active area of research which aims to understand how neural networks perceive images, by exploring ways in which we can look "through their eyes". It has emerged and evolved in response to an increasing desire to make neural networks more interpretable to humans. Earliest works include analysing what neural networks are paying attention to within input images. For example, image-specific class saliency maps visualise regions within an input image that contribute the most to the corresponding output, by calculating the gradient of a class output with respect to the input image via backpropagation (more on saliency maps later in the post). Another strand of technique in feature visualisation is activation maximisation.

Memory management is now a really important topic in Machine Learning. Because of memory constraints, it is becoming quite common to train Deep Learning models using cloud tools such as Kaggle and Google Colab thanks to their free NVIDIA Graphical Processing Unit (GPU) support. Nonetheless, memory can still be a huge constraint in the cloud when working with large amounts of data. In my last article, I explained how to speed up Machine Learning workflow execution. This article aims instead to explain to you how to efficiently reduce memory usage when implementing Deep Learning models.

Combining deep neural networks with structured logic rules is desirable to harness flexibility and reduce uninterpretability of the neural models. We propose a general framework capable of enhancing various types of neural networks (e.g., CNNs and RNNs) with declarative first-order logic rules. Specifically, we develop an iterative distillation method that transfers the structured information of logic rules into the weights of neural networks.

New Delhi: Two startups incubated at the Indian Institute of Technology, Madras, have joined hands with a mission to create 1,00,000 experts in artificial intelligence (AI) and deep learning by 2020. GUVI, which offers a platform for students in Tier 2,3 cities, to learn in vernacular languages, is now collaborating with One Fourth Labs, a startup founded by a IIT Madras faculty, which offers advanced AI courses at nominal fees, starting at ₹1,000. AI is one of the dominant technologies of this generation, which has helped machines reach human-level performance on specific tasks such as identifying faces, classifying images, playing complex strategy games, detecting anomalies in medical images and so on. There is a huge demand for AI talent in India, but the supply is limited due to a shortage of affordable courses, which take students from basics to advanced topics. GUVI will be the platform partner and One Fourth Labs will be the content partner for the joint initiative.

Usually, the task of model building gets reduced to trying all sorts of fancy algorithms - from standard machine algorithm to Deep learning models. But, if we are going to feed garbage to our machine learning algorithm, garbage is going to come out of it (GIGO). In model building, Feature selection/creation is a step where maximum time should be spent. Feature selection is somewhat easier than feature creation. Feature selection is a well-researched area and most of the Data Science algorithms offered under Python or R have automated this process. Feature creation is a bigger dragon to slay.