AWS User Group Frankfurt will kick off 2017 with another Meetup about an interesting and quite hot topic: Machine Learning and Deep Learning. We are currently still planning the details of the Meetup topics and talks, but the date and the location is already fixed. The Meetup will take place on the 25.01.2017 in the office of codecentric Frankfurt. There will be probably two talks. One of which is already confirmed and will be held by Vahe Andonians of SCDM.

Advancements in artificial intelligence and natural language processing are allowing for more realistic and humanlike chatbot experiences. With a combination of greater computing power and enhanced deep learning algorithms, researchers have extended the layers of abstraction that can be processed by artificial neural networks. These networks are capable of allowing software to identify and understand and process patterns in data such as image, sound and text. For users of'griefbots', not only does this allow for greater understanding of their commands into the application, but also the output through the ability to imitate someone's personality through their digital legacy.

The key takeaways from the AI Frontiers Conference, held on January 11–12, include "the central role that AI/ML [Artificial Intelligence/Machine Learning] capabilities are playing within Internet/technology companies, emergence of voice/virtual assistants" and the significant capabilities of Alphabet Inc (NASDAQ: GOOG) (NASDAQ: GOOGL), Amazon.com, Inc. (NASDAQ: AMZN) and Facebook Inc (NASDAQ: FB) in these domains, Baird's Colin Sebastian said in a report. The AI Frontiers Conference focused on applications of deep learning, analyst Sebastian mentioned. Presentations by Alphabet highlighted how Google was using neural networks to "innovate and improve products, including Search, Ads, Photos, Translate, Gmail, Maps, Cloud, voice recognition, self-driving cars and robotics, among others." The teams or projects using deep learning have increased to 4,000, from 200 in early 2014, Sebastian noted.

Facebook's AI research team has released a Python package for GPU-accelerated deep neural network programming that can complement or partly replace existing Python packages for math and stats, such as NumPy. A Python implementation of the Torch machine learning framework, PyTorch has enjoyed broad uptake at Twitter, Carnegie Mellon University, Salesforce, and Facebook. Torch was originally implemented in C with a wrapper in the Lua scripting language, but PyTorch wraps the core Torch binaries in Python and provides GPU acceleration for many functions. Torch is a tensor library for manipulating multidimensional matrices of data employed in machine learning and many other math-intensive applications. PyTorch provides libraries for basic tensor manipulation on CPUs or GPUs, a built-in neural network library, model training utilities, and a multiprocessing library that can work with shared memory, "useful for data loading and hogwild training," as PyTorch's developers put it.

It may sound like something out of a dystopian novel, but scientists are confident about a machine learning technology that can recognize and replicate human activities like seeing and thinking. Leading artificial intelligence experts are investigating ways to commercialize a rapidly emerging sub-field of research known as "deep learning." This month, a research team under renowned scientist Geoffrey E. Hinton's tutelage won a prize sponsored by Merck to design software to uncover molecules that are most likely to be good candidates for new drugs. The win was a particularly impressive feat given that the team entered at the last minute and was working with relatively small data-sets. Click here to read more about "how they did it."

What was your background prior to entering this challenge? We are a team of computer science and statistics academics. Ruslan Salakhutdinov and Geoff Hinton are professors at the University of Toronto. George Dahl and Navdeep Jaitly are Ph.D. students working with Professor Hinton. Christopher "Gomez" Jordan-Squire is in the mathematics Ph.D. program at the University of Washington, studying (constrained) optimization applied to statistics and machine learning.

I ask 67 questions to OpenAI Engineer Catherine Olsson as we take a stroll around OpenAI HQ in San Francisco. Catherine graciously agreed to an interview right after the release of OpenAI's Universe. I ask her questions that range from her deepest fears to her favorite Operating System. Please hit that subscribe button if you liked this interview! That's what keeps me going.

In this post, I show a network analysis of the R and Python ecosystems in terms of their competitors. To identify the typical substitutes/ competitors of a tool, I use the Google search autofill recommendations. Google search prompts identify the most frequently searched terms which occur after a given string and automatically provides a list of suggestions. Thus, this may be treated as a proxy for the common substitutes people search for against a particular tool. In Fig 1 when I start typing "R vs " in the Google Search bar, Google provides a list of suggestions based on their'autocomplete' feature.

Progress in artificial intelligence causes some people to worry that software will take jobs such as driving trucks away from humans. Now leading researchers are finding that they can make software that can learn to do one of the trickiest parts of their own jobs--the task of designing machine-learning software. In one experiment, researchers at the Google Brain artificial intelligence research group had software design a machine-learning system to take a test used to benchmark software that processes language. What it came up with surpassed previously published results from software designed by humans. In recent months several other groups have also reported progress on getting learning software to make learning software.

This lecture will cover recurrent neural networks, the key ingredient in the deep learning toolbox for handling sequential computation and modelling sequences. It will start by explaining how gradients can be computed (by considering the time-unfolded graph) and how different architectures can be designed to summarize a sequence, generate a sequence by ancestral sampling in a fully-observed directed model, or learn to map a vector to a sequence, a sequence to a sequence (of the same or different length) or a sequence to a vector. The issue of long-term dependencies, why it arises, and what has been proposed to alleviate it will be core subject of the discussion in this lecture. This includes changes in the architecture and initialization, as well as how to properly characterize the architecture in terms of recurrent or feedforward depth and its ability to create shortcuts or fast propagation of gradients in the unfolded graph. Open questions regarding the limitations of training by maximum likelihood (teacher forcing) and ideas towards towards making learning online (not requiring backprop through time) will also be discussed.