This article focuses on the political and geopolitical consequences of the feedback relationship linking Artificial Intelligence (AI) in its Deep Learning component and computing power – hardware – or rather high performance computing power (HPC). It builds on a first part where we explained and detailed this connection. There we underlined notably three typical phases where computation is required: creation of the AI program, training, and inference or production (usage). We showed that a quest for improvement across phases, and the overwhelming and determining importance of architecture design – which takes place during the creation phase – generates a crucial need for ever more powerful computing power. Meanwhile, we identified a feedback spiral between AI-DL and computing power, where more computing power allows for advances in terms of AI and where new AI and the need to optimize it demand more computing power.

Artificial intelligence has an unimaginable potential. Within the next couple of years, it will revolutionize every area of our life, including medicine. I am fully convinced that it will redesign healthcare completely – and for the better. Let's take a look at the promising solutions it offers. There are various thought leaders who believe that we are experiencing the Fourth Industrial Revolution, which is characterized by a range of new technologies that are fusing the physical, digital and biological worlds, impacting all disciplines, economies and industries, and even challenging ideas about what it means to be human.

This paper addresses a key NLP problem known as sarcasm detection using a combination of models based on convolutional neural networks (CNNs). Detection of sarcasm is important in other areas such as affective computing and sentiment analysis because such expressions can flip the polarity of a sentence. Sarcasm can be considered as expressing a bitter gibe or taunt. Examples include statements such as "Is it time for your medication or mine?" and "I work 40 hours a week to be this poor". To understand and detect sarcasm it is important to understand the factsrelated to an event.

This post is authored by Viral B. Shah, co-creator of the Julia language and co-founder and CEO at Julia Computing, and Avik Sengupta, head of engineering at Julia Computing. The Julia language provides a fresh new approach to numerical computing, where there is no longer a compromise between performance and productivity. A high-level language that makes writing natural mathematical code easy, with runtime speeds approaching raw C, Julia has been used to model economic systems at the Federal Reserve, drive autonomous cars at University of California Berkeley, optimize the power grid, calculate solvency requirements for large insurance firms, model the US mortgage markets and map all the stars in the sky. It would be no surprise then that Julia is a natural fit in many areas of machine learning. And the powers of Julia make it a perfect language to implement these algorithms.

For the past five years, the hottest thing in artificial intelligence has been a branch known as deep learning. The grandly named statistical technique, put simply, gives computers a way to learn by processing massive amounts of data. Thanks to deep learning, computers can easily identify faces and recognize spoken words, making other forms of humanlike intelligence suddenly seem within reach. Companies like Google, Facebook and Microsoft have poured money into deep learning. And the technology's perception and pattern-matching abilities are being applied to improve progress in fields such as drug discovery and self-driving cars. But now some scientists are asking whether deep learning is really so deep after all.

Big data, artificial intelligence and machine learning are ruling the tech structure of most industries. We all know how Amazon combines a customer's historical data and other customers' data to power recommendations. Likewise, for Google, it's not difficult to predict our preferences and interests. They make use of big data, analytics and machine learning to be able to process huge amounts of data, identify patterns, analyze them and consequently indulge in predictive analysis. The most complicated disease of the most important organ of the body – the brain, is a clear beneficiary of this AI approach.

Chemical autoencoders are attractive models as they combine chemical space navigation with possibilities for de novo molecule generation in areas of interest. This enables them to produce focused chemical libraries around a single lead compound so they can be employed early in a drug discovery project. Here it is shown that the choice of chemical representation, such as SMILES strings, has a large influence on the properties of the latent space. It is further explored to what extent translating between different chemical representations influences the latent space similarity to the SMILES strings or circular fingerprints. By employing SMILES enumeration for both the encoder and decoder, it is found that the decoder has the largest influence on the properties of the latent space. Training a sequence to sequence heteroencoder based on recurrent neural networks(RNNs) with long short-term memory cells (LSTM) to predicts different enumerated SMILES strings from the same canonical SMILES string gives the largest similarity between latent space distance and molecular similarity measured as circular fingerprints similarity.

Given independent samples generated from the joint distribution $p(\mathbf{x},\mathbf{y},\mathbf{z})$, we study the problem of Conditional Independence (CI-Testing), i.e., whether the joint equals the CI distribution $p^{CI}(\mathbf{x},\mathbf{y},\mathbf{z})= p(\mathbf{z}) p(\mathbf{y}|\mathbf{z})p(\mathbf{x}|\mathbf{z})$ or not. We cast this problem under the purview of the proposed, provable meta-algorithm, "Mimic and Classify", which is realized in two-steps: (a) Mimic the CI distribution close enough to recover the support, and (b) Classify to distinguish the joint and the CI distribution. Thus, as long as we have a good generative model and a good classifier, we potentially have a sound CI Tester. With this modular paradigm, CI Testing becomes amiable to be handled by state-of-the-art, both generative and classification methods from the modern advances in Deep Learning, which in general can handle issues related to curse of dimensionality and operation in small sample regime. We show intensive numerical experiments on synthetic and real datasets where new mimic methods such conditional GANs, Regression with Neural Nets, outperform the current best CI Testing performance in the literature. Our theoretical results provide analysis on the estimation of null distribution as well as allow for general measures, i.e., when either some of the random variables are discrete and some are continuous or when one or more of them are discrete-continuous mixtures.

Recently, much progress has been made building systems that can capture static image properties, but natural environments are intrinsically dynamic. For an intelligent agent, perception is responsible not only for capturing features of scene content, but also capturing its \textit{affordances}: how the state of things can change, especially as the result of the agent's actions. We propose an unsupervised method to learn representations of the sensorimotor affordances of an environment. We do so by learning an embedding for stochastic future prediction that is (i) sensitive to scene dynamics and minimally sensitive to static scene content and (ii) compositional in nature, capturing the fact that changes in the environment can be composed to produce a cumulative change. We show that these two properties are sufficient to induce representations that are reusable across visually distinct scenes that share degrees of freedom. We show the applicability of our method to synthetic settings and its potential for understanding more complex, realistic visual settings.

The broad set of deep generative models (DGMs) has achieved remarkable advances. However, it is often difficult to incorporate rich structured domain knowledge with the end-to-end DGMs. Posterior regularization (PR) offers a principled framework to impose structured constraints on probabilistic models, but has limited applicability to the diverse DGMs that can lack a Bayesian formulation or even explicit density evaluation. PR also requires constraints to be fully specified {\it a priori}, which is impractical or suboptimal for complex knowledge with learnable uncertain parts. In this paper, we establish mathematical correspondence between PR and reinforcement learning (RL), and, based on the connection, expand PR to learn constraints as the extrinsic reward in RL. The resulting algorithm is model-agnostic to apply to any DGMs, and is flexible to adapt arbitrary constraints with the model jointly. Experiments on human image generation and templated sentence generation show models with learned knowledge constraints by our algorithm greatly improve over base generative models.