Convolutional Neural Networks (CNNs) have achieved state-of-the-art performance on a variety of computer vision tasks, particularly visual classification problems, where new algorithms reported to achieve or even surpass the human performance. In this paper, we examine whether CNNs are capable of learning the semantics of training data. To this end, we evaluate CNNs on negative images, since they share the same structure and semantics as regular images and humans can classify them correctly. Our experimental results indicate that when training on regular images and testing on negative images, the model accuracy is significantly lower than when it is tested on regular images. This leads us to the conjecture that current training methods do not effectively train models to generalize the concepts. We then introduce the notion of semantic adversarial examples - transformed inputs that semantically represent the same objects, but the model does not classify them correctly - and present negative images as one class of such inputs.

Our goal is to design architectures that retain the groundbreaking performance of CNNs for landmark localization and at the same time are lightweight, compact and suitable for applications with limited computational resources. To this end, we make the following contributions: (a) we are the first to study the effect of neural network binarization on localization tasks, namely human pose estimation and face alignment. We exhaustively evaluate various design choices, identify performance bottlenecks, and more importantly propose multiple orthogonal ways to boost performance.

Machine Learning and Deep Learning has brought the future here. Predicting the future has always been the most sought after skill in this world. How much money could you make if you could predict the price of a stock or if you could predict which color will be in fashion six months later? You can predict almost anything that you wish. The future will be in your own palms.

University of Washington researchers at the UW Graphics and Image Laboratory have developed new algorithms that turn audio clips into a realistic, lip-synced video, starting with an existing video of that person speaking on a different topic. As detailed in a paper to be presented Aug. 2 at SIGGRAPH 2017, the team successfully generated a highly realistic video of former president Barack Obama talking about terrorism, fatherhood, job creation and other topics, using audio clips of those speeches and existing weekly video addresses in which he originally spoke on a different topic decades ago. Realistic audio-to-video conversion has practical applications like improving video conferencing for meetings (streaming audio over the internet takes up far less bandwidth than video, reducing video glitches), or holding a conversation with a historical figure in virtual reality, said Ira Kemelmacher-Shlizerman, an assistant professor at the UW's Paul G. Allen School of Computer Science & Engineering. This beats previous audio-to-video conversion processes, which have involved filming multiple people in a studio saying the same sentences over and over to try to capture how a particular sound correlates to different mouth shapes, which is expensive, tedious and time-consuming. The new machine learning tool may also help overcome the "uncanny valley" problem, which has dogged efforts to create realistic video from audio.

UPDATE: Mar 20, 2016 - Added my new follow-up course on Deep Learning, which covers ways to speed up and improve vanilla backpropagation: momentum and Nesterov momentum, adaptive learning rate algorithms like AdaGrad and RMSProp, utilizing the GPU on AWS EC2, and stochastic batch gradient descent. We look at TensorFlow and Theano starting from the basics - variables, functions, expressions, and simple optimizations - from there, building a neural network seems simple! Deep learning is all the rage these days. What exactly is deep learning? Well, it all boils down to neural networks.

First some general, higher-level thoughts and comments before I dive into specifics: MY BACKGROUND: Data Scientist; B.S. in Economics and M.S. in Business Analytics; experienced (though by no means expert) user of Scikit-learn OVERALL THOUGHTS: I've purchased and read (virtually) every Machine Learning book that aims to teach the reader the basics of ML using the Scikit-learn library as the main focus. I've found them to be...less than satisfactory. The examples in other books often use ML techniques in contexts for which they are not intended to be used and/or contexts they are not used in out in the real world (among other issues I have found within them). In stark contrast, Python Machine Learning by Sebastian Raschka is stunningly-impressive, not only for the breadth and depth of coverage, but also in the manner the information is presented to the reader. To date, I have not encountered a book on ML that incorporates multiple levels of learning in a manner such as this.

Every industry is inching towards automation and the insurance industry is no exception. According to a white paper titled "2017 Future of Claims Study" published by the legal research firm Lexis Nexis, insurance claims are being increasingly processed using drones, artificial intelligence and app-based interfaces as opposed to sending field agents to examine such claims. The study was conducted using a sample size of 24 insurance executives and their opinions on automation in insurance. The push towards automation is largely driven by customers' need for faster and more convenient processing of claims. " While there hasn't yet been a complete shift to Virtual Claims handling, carriers who want to remain competitive will need to make the move to virtual and consider touchless processing if customer preferences are any indication," the study says.

One of the most important tasks in Machine Learning are the Classification tasks (a.k.a. Classification is used to make an accurate prediction of the class of entries in the test set (a dataset of which the entries have not been labelled yet) with the model which was constructed from a training set. You could think of classifying crime in the field of Pre-Policing, classifying patients in the Health sector, classifying houses in the Real-Estate sector. Another field in which classification is big, is Natural Lanuage Processing (NLP). This is the field of science with the goal to makes machines (computers) understand (written) human language.

Google parent Alphabet Inc. GOOGL 0.58% is pitching an idea to store power from renewable energy in tanks of molten salt and cold liquid, an example of the tech giant trying to marry its far-reaching ambitions with business demand. Alphabet's research lab, dubbed X, said Monday that it has developed plans to store electricity generated from solar panels or wind turbines as thermal energy in hot salt and cold liquids, such as antifreeze. The lab is seeking partners in the energy industry, including power-plant developers and utilities, to build a prototype to plug into the electrical grid. Whether the project, called Malta, ever comes to market depends as much on a sound business model as it does on science. Academics said the technology is likely years away from market, if it ever makes it.

Distributed learning is an effective way to analyze big data. In distributed regression, a typical approach is to divide the big data into multiple blocks, apply a base regression algorithm on each of them, and then simply average the output functions learnt from these blocks. Since the average process will decrease the variance, not the bias, bias correction is expected to improve the learning performance if the base regression algorithm is a biased one. Regularization kernel network is an effective and widely used method for nonlinear regression analysis. In this paper we will investigate a bias corrected version of regularization kernel network. We derive the error bounds when it is applied to a single data set and when it is applied as a base algorithm in distributed regression. We show that, under certain appropriate conditions, the optimal learning rates can be reached in both situations.