Which technologies have the greatest potential to transform the world over the next decade? The National Football League is teaming up with Sleep Number to help its players use big data and machine learning to improve their sleep and boost performance. Research and advisory firm Lux Research set out to find the answer, applying its in-house data analysis platform and the expertise of its global technical team to identify and rank the 18 most transformative technologies. The firm's newly released "18 for 2018" report covers everything "from current rock stars of innovation to hidden gem technologies." At the top of the list of potentially transformative technologies is machine learning and deep neural networks.

ZeitgeistMinds 1,617 views How we teach computers to understand pictures Fei Fei Li - Duration: 18:03.

Mobile technology provides unprecedented reach, connecting people on the move no matter where they are in the world. The cloud delivers capacity on a previously unimaginable scale, enabling organisations to add or remove various components to their infrastructure quickly and as needed. Deep-learning AI enables us to mine massive amounts of data in real-time and to use those insights to dramatically accelerate academic discovery and create entirely new business models. And the IoT connects the physical and digital worlds, bringing technology into every dimension of human progress. As these innovations quickly mature and build on one another, they are reshaping every aspect of society, from healthcare to education to transport and financial inclusion.

A translation widget is provided for your convenience to facilitate translation of the English language version of this blog into several languages. If you choose to utilize this automated translation facility, please understand there may be deviations between the automated translation and the original English version. IBM is not responsible for any such automated translation deviations and offers the translated version "AS IS" without warranties of any kind. This is an exciting time for IBM Business Partners and their customers, as IBM unveils its next-generation Power Systems servers incorporating its newly designed POWER9 processor. Built specifically for compute-intensive AI workloads, the new POWER9 systems are capable of improving the training times of deep learning frameworks by nearly 4x,[i] allowing enterprises to build more accurate AI applications, faster.

Thanks to Google and artificial intelligence (AI) research company, DeepMind, your phone will no longer sound like a robot when reading out or dictating requested information. Google Assistant is using an improved version of DeepMind's WaveNet, a deep neural network that can synthesize realistic human speech. WaveNet uses an improved system of speech synthesis or text-to-speech (TTS). In order to closely mimic human speech, concatenative TTS juxtaposes different parts of a voice actor's recordings to construct the desired sentence. Upgrading concatenative TTS is cumbersome as it involves replacing the audio libraries.

"Once upon a time, I, Chuang Tzu, dreamt I was a butterfly, fluttering hither and thither, to all intents and purposes a butterfly. I was conscious only of following my fancies as a butterfly, and was unconscious of my individuality as a man. Suddenly, I awoke, and there I lay, myself again. Now I do not know whether I was then a man dreaming I was a butterfly, or whether I am now a butterfly dreaming that I am a man." from The Brain: The Story of you – David Eagleman "Thought is a great big vector of neural activity" Prof Geoffrey Hinton This is the third part in my series on Deep Learning from first principles in Python, R and Octave. In the first part Deep Learning from first principles in Python, R and Octave-Part 1, I implemented logistic regression as a 2 layer neural network.

A lot has happened since Part III. While the last couple of articles went in-depth into exactly I was learning, this one will be a little different. Rather than break it down week by week, I'll cover the major milestones. I graduated from the Udacity Deep Learning Nanodegree (DLND) in August last year. Thinking about how I emailed the support team asking what the refund policy was before starting the course makes me laugh.

We study the impact of regularization for learning neural networks. Our goal is speeding up training, improving generalization performance, and training compact models that are cost efficient. Our results apply to weight-sharing (e.g.~convolutional), sparsity (i.e.~pruning), and low-rank constraints among others. We first introduce covering dimension of the constraint set and provide a Rademacher complexity bound providing insights on generalization properties. Then, we propose and analyze regularized gradient descent algorithms for learning shallow networks. We show that problem becomes well conditioned and local linear convergence occurs once the amount of data exceeds covering dimension (e.g.~\# of nonzero weights). Finally, we provide insights on layerwise training of deep models by studying a random activation model. Our results show how regularization can be beneficial to overcome overparametrization.

We consider a class of a nested optimization problems involving inner and outer objectives. We observe that by taking into explicit account the optimization dynamics for the inner objective it is possible to derive a general framework that unifies gradient-based hyperparameter optimization and meta-learning (or learning-to-learn). Depending on the specific setting, the variables of the outer objective take either the meaning of hyperparameters in a supervised learning problem or parameters of a meta-learner. We show that some recently proposed methods in the latter setting can be instantiated in our framework and tackled with the same gradient-based algorithms. Finally, we discuss possible design patterns for learning-to-learn and present encouraging preliminary experiments for few-shot learning.

We present graph attention networks (GATs), novel neural network architectures that operate on graph-structured data, leveraging masked self-attentional layers to address the shortcomings of prior methods based on graph convolutions or their approximations. By stacking layers in which nodes are able to attend over their neighborhoods' features, we enable (implicitly) specifying different weights to different nodes in a neighborhood, without requiring any kind of costly matrix operation (such as inversion) or depending on knowing the graph structure upfront. In this way, we address several key challenges of spectral-based graph neural networks simultaneously, and make our model readily applicable to inductive as well as transductive problems. Our GAT models have achieved or matched state-of-the-art results across four established transductive and inductive graph benchmarks: the Cora, Citeseer and Pubmed citation network datasets, as well as a protein-protein interaction dataset (wherein test graphs remain unseen during training).