IT4Innovations and M Computers would like to invite you to three full day NVIDIA Deep Learning Institute certified training courses to learn more about Artificial Intelligence (AI) and High Performance Computing (HPC) development for NVIDIA GPUs. The first half day is an introduction by IT4Innovations and M Computers about the latest state of the art NVIDIA technologies. We also explain our services offered for AI and HPC, for industrial and academic users. The introduction will include a tour though IT4Innovations' computing center, which hosts an NVIDIA DGX-2 system and the new Barbora cluster with V100 GPUs. The first full day training course, Fundamentals of Deep Learning for Computer Vision, is provided by IT4Innovations and gives you an introduction to AI development for NVIDIA GPUs.

At the 2019 AWS re:Invent conference, Amazon released Deep Java Library (DJL), an open-source library with Java APIs to simplify training, testing, deploying, and making predictions with deep-learning models. While Java remains the first or second most popular programming language since the late 90s, Python is the most used language for machine learning, with numerous resources and deep-learning frameworks. DJL aims to make deep-learning open-source tools accessible to Java developers, using familiar concepts and intuitive APIs. Java developers can use their favorite IDE with DJL or Jupyter Notebook-based code execution for Java. DJL is framework agnostic; it abstracts away commonly used deep-learning functions, using Java Native Access (JNA) on top of existing deep-learning frameworks, currently providing implementations for Apache MXNet and TensorFlow.

Reinforcement learning (RL) tasks are challenging to implement, execute and test due to algorithmic instability, hyper-parameter sensitivity, and heterogeneous distributed communication patterns. We argue for the separation of logical component composition, backend graph definition, and distributed execution. To this end, we introduce RLgraph, a library for designing and executing reinforcement learning tasks in both static graph and define-by-run paradigms. The resulting implementations are robust, incrementally testable, and yield high performance across different deep learning frameworks and distributed backends.

Machine learning Machine learning is a branch of AI focused around the idea that computer algorithms can recognize patterns and "learn," continuing to improve the more data they are fed. Whereas the broader term "artificial intelligence" refers to systems that can produce smart results in a variety of situations, "machine learning" refers to systems that can infer from experience and adapt. Neural networks Neural networks are algorithms of many densely interconnected processing nodes that have been inspired by the human brain. They are defined initially without any specific operating rules, instead inferring rules and connections through pattern recognition. Deep learning Deep learning is a subcategory of machine learning that uses neural networks to analyze a set of data along a wide variety of different dimensions, identifying patterns, and stacking these patterns on top of one another to create categories that can be used for classification.

Artificial intelligence is gaining traction in areas of social responsibility. From climate change to social polarization to epidemics, humankind has been seeking new solutions to old but persistent problems. From a technological point of view, the amount of daily data produced in the digital universe now allows for state-of-the-art approaches, which may lead to innovative solutions in these underserved areas. AI for social good turned into a reality for us at Tryolabs after we collaborated with an NGO to improve upon how African lions are tracked, which helps with species preservation. We will go into more detail on that timely case, especially as wildlife conservation faces the immense challenges posed by devastating megafires threatening the lives of millions of animals in historic ways.

A variety of computationally challenging constrained optimization problems in several engineering disciplines are solved repeatedly under different scenarios. In many cases, they would benefit from fast and accurate approximations, either to support real-time operations or large-scale simulation studies. This paper aims at exploring how to leverage the substantial data being accumulated by repeatedly solving instances of these applications over time. It introduces a deep learning model that exploits Lagrangian duality to encourage the satisfaction of hard constraints. The proposed method is evaluated on a collection of realistic energy networks, by enforcing non-discriminatory decisions on a variety of datasets, and on a transprecision computing application. The results illustrate the effectiveness of the proposed method that dramatically decreases constraint violations by the predictors and, in some applications, increases the prediction accuracy.

Over the past year, the emergence of transfer learning with large-scale language models (LM) has led to dramatic performance improvements across a broad range of natural language understanding tasks. However, the size and memory footprint of these large LMs makes them difficult to deploy in many scenarios (e.g. on mobile phones). Recent research points to knowledge distillation as a potential solution, showing that when training data for a given task is abundant, it is possible to distill a large (teacher) LM into a small task-specific (student) network with minimal loss of performance. However, when such data is scarce, there remains a significant performance gap between large pretrained LMs and smaller task-specific models, even when training via distillation. In this paper, we bridge this gap with a novel training approach, called generation-distillation, that leverages large finetuned LMs in two ways: (1) to generate new (unlabeled) training examples, and (2) to distill their knowledge into a small network using these examples. Across three low-resource text classification datsets, we achieve comparable performance to BERT while using 300 fewer parameters, and we outperform prior approaches to distillation for text classification while using 3 fewer parameters. 1 Introduction Over the past year, rapid progress in unsupervised language representation learning has led to the development of increasingly powerful and gener-alizable language models (Radford et al., 2019; Devlin et al., 2018).

Molecular graph generation is a fundamental problem for drug discovery and has been attracting growing attention. The problem is challenging since it requires not only generating chemically valid molecular structures but also optimizing their chemical properties in the meantime. Inspired by the recent progress in deep generative models, in this paper we propose a flow-based autoregressive model for graph generation called GraphAF. GraphAF combines the advantages of both autoregressive and flow-based approaches and enjoys: (1) high model flexibility for data density estimation; (2) efficient parallel computation for training; (3) an iterative sampling process, which allows leveraging chemical domain knowledge for valency checking. Experimental results show that GraphAF is able to generate 68% chemically valid molecules even without chemical knowledge rules and 100% valid molecules with chemical rules. The training process of GraphAF is two times faster than the existing state-of-the-art approach GCPN. After fine-tuning the model for goal-directed property optimization with reinforcement learning, GraphAF achieves state-of-the-art performance on both chemical property optimization and constrained property optimization.

Deep learning models have demonstrated high-quality performance in areas such as image classification and speech processing. However, creating a deep learning model using electronic health record (EHR) data, requires addressing particular privacy challenges that are unique to researchers in this domain. This matter focuses attention on generating realistic synthetic data while ensuring privacy. In this paper, we propose a novel framework called correlation-capturing Generative Adversarial Network (corGAN), to generate synthetic healthcare records. In corGAN we utilize Convolutional Neural Networks to capture the correlations between adjacent medical features in the data representation space by combining Convolutional Generative Adversarial Networks and Convolutional Autoencoders. To demonstrate the model fidelity, we show that corGAN generates synthetic data with performance similar to that of real data in various Machine Learning settings such as classification and prediction. We also give a privacy assessment and report on statistical analysis regarding realistic characteristics of the synthetic data. The software of this work is open-source and is available at: https://github.com/astorfi/cor-gan.

Question-answering (QA) is certainly the best known and probably also one of the most complex problem within Natural Language Processing (NLP) and artificial intelligence (AI). Since the complete solution to the problem of finding a generic answer still seems far away, the wisest thing to do is to break down the problem by solving single simpler parts. Assuming a modular approach to the problem, we confine our research to intent classification for an answer, given a question. Through the use of an LSTM network, we show how this type of classification can be approached effectively and efficiently, and how it can be properly used within a basic prototype responder. Keywords: Deep Learning, LSTM, Intent classification, Question-Answering 1 Introduction Despite the remarkable results obtained in the different areas of Natural Language Processing, the solution to the Question-Answering problem, in its general sense, still seems far away [1].