The availability of large amounts of time series data, paired with the performance of deep-learning algorithms on a broad class of problems, has recently led to significant interest in the use of sequence-to-sequence models for time series forecasting. We provide the first theoretical analysis of this time series forecasting framework. We include a comparison of sequence-to-sequence modeling to classical time series models, and as such our theory can serve as a quantitative guide for practitioners choosing between different modeling methodologies.

Policy optimization on high-dimensional continuous control tasks exhibits its difficulty caused by the large variance of the policy gradient estimators. We present the action subspace dependent gradient (ASDG) estimator which incorporates the Rao-Blackwell theorem (RB) and Control Variates (CV) into a unified framework to reduce the variance. To invoke RB, our proposed algorithm (POSA) learns the underlying factorization structure among the action space based on the second-order advantage information. POSA captures the quadratic information explicitly and efficiently by utilizing the wide & deep architecture. Empirical studies show that our proposed approach demonstrates the performance improvements on high-dimensional synthetic settings and OpenAI Gym's MuJoCo continuous control tasks.

Experimental evidence indicates that simple models outperform complex deep networks on many unsupervised similarity tasks. We provide a simple yet rigorous explanation for this behaviour by introducing the concept of an optimal representation space, in which semantically close symbols are mapped to representations that are close under a similarity measure induced by the model's objective function. In addition, we present a straightforward procedure that, without any retraining or architectural modifications, allows deep recurrent models to perform equally well (and sometimes better) when compared to shallow models. To validate our analysis, we conduct a set of consistent empirical evaluations and introduce several new sentence embedding models in the process. Even though this work is presented within the context of natural language processing, the insights are readily applicable to other domains that rely on distributed representations for transfer tasks.

Overview Deep Reinforcement Learning and GANs LiveLessons is an introduction to two of the most exciting topics in Deep Learning today.

Machine learning is a subset of artificial intelligence that permits the computer with the ability to learn things on the go. The current level of Artificial Intelligence is achieved through years of research in Machine Learning, Deep Learning and other related fields. With a lot of hype and investments around, Deep Learning technology – a subdivision of Machine Learning is now successfully applied in our daily life from speech recognition apps in smartphones to YouTube recommendations. The machine learning is mainly used for the advancement of computer programs that can change when the new data is introduced to the picture. The factors that promote the growth of machine learning are its diverse application and its ability to learn and solve real life problems from data.

Fear mongering about killer robots and the recent deaths connected with Uber and Tesla autonomous vehicles have rekindled concerns about artificial intelligence in the machines around us. We are well beyond answering Alan Turing's question, "can machines think?" There is now good reason to ask how we should think of AI, and what we should expect from it. There have been phenomenal advances in AI in just the past few years. They are due in part to advances in processor technology that have increased exponentially the compute performance for artificial neural networks, the development of deep learning software frameworks, and the massive amounts of data mined directly from the Internet and the world around us.

Machines are getting better and better at analyzing complex health data to help physicians better understand their patients' future needs. In a study out today in Nature Digital Medicine, an advanced algorithm evaluated de-identified electronic health records of more than 216,000 adult patient hospitalizations to predict unexpected readmissions, long hospital stays, and in-hospital deaths more accurately than previous approaches. I caught up with one of the authors, Nigam Shah, MBBS, PhD, an associate professor at Stanford, to learn about the new study and discuss the implications for artificial intelligence in medicine. What is deep learning and how does it fit in the larger universe of artificial intelligence? Deep learning is one of several machine learning techniques that can be used to build intelligent systems.

In January 2018, I entered a Kaggle competition called the Mercari Price Suggestion. The competition lasted three months and ended a few weeks ago. The challenge was to build an algorithm that automatically suggests product prices to online sellers, based on free-text descriptions, product categories and a few other attributes. I joined the competition a month before it ended, eager to explore how to use Deep Natural Language Processing (NLP) techniques for this problem.Those four weeks competing were a sequence of curiosity, engagement, and excitement. I ended up with a single end-to-end Deep Learning model (no ensembles) implemented in Tensorflow, which landed me in the the 35th position (out of 2,384 teams) in the Private Leaderboard.

Drag in your training data and Lobe automatically builds you a custom deep learning model. Then refine your model by adjusting settings and connecting pre-trained building blocks. Monitor training progress in real-time with interactive charts and test results that update live as your model improves. Cloud training lets you get results quickly, without slowing down your computer. Export your trained model to CoreML or TensorFlow and run it directly in your app on iOS and Android.

As the war for creating customized AI hardware heats up, Google announced at Google I/O 2018 that is rolling out out its third generation of silicon, the Tensor Processor Unit 3.0. Google CEO Sundar Pichai said the new TPU is eight times more powerful than last year, with up to 100 petaflops in performance. Google joins pretty much every other major company in looking to create custom silicon in order to handle its machine operations. And while multiple frameworks for developing machine learning tools have emerged, including PyTorch and Caffe2, this one is optimized for Google's TensorFlow. Google is looking to make Google Cloud an omnipresent platform at the scale of Amazon, and offering better machine learning tools is quickly becoming table stakes.