One way to interpret neural model predictions is to highlight the most important input features---for example, a heatmap visualization over the words in an input sentence. In existing interpretation methods for NLP, a word's importance is determined by either input perturbation---measuring the decrease in model confidence when that word is removed---or by the gradient with respect to that word. To understand the limitations of these methods, we use input reduction, which iteratively removes the least important word from the input. This exposes pathological behaviors of neural models: the remaining words appear nonsensical to humans and are not the ones determined as important by interpretation methods. As we confirm with human experiments, the reduced examples lack information to support the prediction of any label, but models still make the same predictions with high confidence. To explain these counterintuitive results, we draw connections to adversarial examples and confidence calibration: pathological behaviors reveal difficulties in interpreting neural models trained with maximum likelihood. To mitigate their deficiencies, we fine-tune the models by encouraging high entropy outputs on reduced examples. Fine-tuned models become more interpretable under input reduction without accuracy loss on regular examples.

Deep learning approaches have achieved state-of-the-art performance in cardiac magnetic resonance (CMR) image segmentation. However, most approaches have focused on learning image intensity features for segmentation, whereas the incorporation of anatomical shape priors has received less attention. In this paper, we combine a multi-task deep learning approach with atlas propagation to develop a shape-constrained bi-ventricular segmentation pipeline for short-axis CMR volumetric images. The pipeline first employs a fully convolutional network (FCN) that learns segmentation and landmark localisation tasks simultaneously. The architecture of the proposed FCN uses a 2.5D representation, thus combining the computational advantage of 2D FCNs networks and the capability of addressing 3D spatial consistency without compromising segmentation accuracy. Moreover, the refinement step is designed to explicitly enforce a shape constraint and improve segmentation quality. This step is effective for overcoming image artefacts (e.g. due to different breath-hold positions and large slice thickness), which preclude the creation of anatomically meaningful 3D cardiac shapes. The proposed pipeline is fully automated, due to network's ability to infer landmarks, which are then used downstream in the pipeline to initialise atlas propagation. We validate the pipeline on 1831 healthy subjects and 649 subjects with pulmonary hypertension. Extensive numerical experiments on the two datasets demonstrate that our proposed method is robust and capable of producing accurate, high-resolution and anatomically smooth bi-ventricular 3D models, despite the artefacts in input CMR volumes.

What inspired our team to build the DGX Station? Get an inside look into how this deep learning workstation was designed for developers and researchers to effortlessly bring their deep-learning initiatives to the office. We packed this portable and whisper-quiet system with twice the performance of the most powerful workstations, so AI researchers can carry out their life's work from their desks.

How to develop and evaluate a baseline neural network model for the MNIST problem. How to implement and evaluate a simple Convolutional Neural Network for MNIST. How to implement a close to state-of-the-art deep learning model for MNIST. How to develop and evaluate a baseline neural network model for the MNIST problem. How to implement and evaluate a simple Convolutional Neural Network for MNIST.

You're going to hear many discussions about AI and Deep Learning at VMworld, along with other associated technologies. What is this all about? To get you ready for VMworld, let's talk definitions. AI is the capability for a machine to learn to act as a human (to see, to hear, to respond, to decide, etc). This is accomplished by teaching the machine to learn, either through machine learning or deep learning.

DeepMind Technologies, an artificial intelligence (AI) company based in London (U.K.) that was purchased by Google in 2014, recently announced that it is applying its machine learning algorithms to Google data centers to reduce the amount of energy dedicated to cooling by as much as 40 percent.

AI may have beaten the world's best Go player, but Dota 2 pros have shown that in their game, humans are still top of the food chain -- for now, at least. Last week, Dota 2 players from around the world clashed at the biggest tournament of the year, The International, with team OG taking the title and over $11 million in prize money. Arguably more important, though, was the contest of man versus machine(-learning) in a best-of-three exhibition series. OpenAI, the research group co-founded by Elon Musk, took its team of five bots to The International to square up against professional players in their toughest test yet. Earlier in August, OpenAI wiped the floor with a squad of Dota 2 casters and ex-pro players in a warm-up match.

Today, deep neural networks with different architectures, such as convolutional, recurrent and autoencoder networks, are becoming an increasingly popular area of research. A number of high-tech companies, including Microsoft and Google, are using deep neural networks to design intelligent systems. In deep learning systems, the processes of feature selection and configuration are automated, which means that the networks can choose between the most effective algorithms for hierarchal feature extraction on their own. Deep learning is characterized by learning with the help of large samples using a single optimization algorithm. Typical optimization algorithms configure the parameters of all operations simultaneously, and effectively estimate every neural network parameter's effect on error with the help of the so-called backpropagation method.

In Plato's Republic, Socrates challenges the idea that justice should boil down to telling the truth and returning things that were taken. He raises an interesting dilemma: what if the goods in question would harm other individuals if returned? In other words, giving a weapon back to its owner constitutes a moral dilemma as it could lead to people being harmed by the weapon's use. For Socrates, there was no doubt that protecting people was what should always prevail. Centuries later, Socrates's view is still very relevant in light of recent developments regarding the use of artificial intelligence in warfare.

Pardon me for overgeneralizing, but I also see of folks who very loosely use the word A.I and have absolutely no clue/idea about what they are talking about. If you try to avoid them and try to seek the answer for "What is AI?", you are bound to get flooded with conflicting views and very obfuscated terms and definitions. Just because someone is using a deep learning library/package, that doesn't mean their system is intelligent. There is more to it. So here is my yet another modest attempt to convey via a picture -- "Demystifying AI".