Molecular "fingerprints" encoding structural information are the workhorse of cheminformatics and machine learning in drug discovery applications. However, fingerprint representations necessarily emphasize particular aspects of the molecular structure while ignoring others, rather than allowing the model to make data-driven decisions. We describe molecular "graph convolutions", a machine learning architecture for learning from undirected graphs, specifically small molecules. Graph convolutions use a simple encoding of the molecular graph---atoms, bonds, distances, etc.---which allows the model to take greater advantage of information in the graph structure. Although graph convolutions do not outperform all fingerprint-based methods, they (along with other graph-based methods) represent a new paradigm in ligand-based virtual screening with exciting opportunities for future improvement.

There was news recently in bloomberg about how google was able to cut electricity usage in its datacenter by using an AI scheme made by DeepMind (of AlphaGo fame). Earlier this week, i decided to make a quick-and-dirty implemetation in python and share it here for anyone interested in a practical example of what exactly they did. First lets take a quick look at why one would want to make such a thing... Datacenters (and indeed any other large scale structures that use a lot of energy) need to be carefully optimized for efficiency as even a 10% - 15% saving on the electricity bill can add up to millions of dollars a year. The biggest challenge here is that even though there are certain simple steps that anyone can take to reduce energy use (don't use a very low server room set-point, use free-cooling when possible, etc…) one can never actually predict quantitatively what the effect of changing variable x by z% will have on total consumption. This is because there simply are too many variables that affect the net consumption of a datacenter (chillers, AHUs, compressors, condensers, fans, outside conditions, latitude, etc…) and its impossible to actually write down a formula that can quantify all these relationships.

In a brief announcement as part of today's Day 2 ketnote for IDF 2016, Intel has announced a new member of the Xeon Phi family. The new part, currently under the codename of Knights Mill, is being aimed at the deep learning market and is scheduled for release in 2017. At this point there are more unknowns than knowns about Knights Mill, in part because Intel has not offered much detail on how it fits into the larger Xeon Phi brand. The company had previously announced in 2014 that the successor to the current Knights Landing design would be Knights Hill, a true 3rd gen Xeon Phi built on Intel's 10nm process. However this week there has been no mention of Knights Hill, whether Knights Mill is Knights Hill renamed, or what the manufacturing process Knights Mill is being made on.

Intel Corp. INTC -0.54 % signaled it wants a bigger role in artificial intelligence, revealing plans to modify a line of chips to target a fast-growing market turning into a battleground for technology suppliers. The company told technology developers Wednesday that it plans next year to deliver a new version of the Xeon Phi processor--a product line previously targeted at scientific applications--with added features designed to accelerate tasks associated with what Silicon Valley calls artificial intelligence. Intel said the technology will help accelerate a technique called deep learning, increasingly used for tasks such as interpreting speech, identifying objects in photos and piloting autonomous vehicles. Intel's Xeon processors already are a fixture in data centers, and have a role in nearly all deep-learning tasks carried out there. But some users also install auxiliary processors for artificial-intelligence tasks, notably chips called GPUs that rival Nvidia Corp. NVDA -2.32 % has long sold for videogames.

At Harvard Business School (HBS), MBA students are pondering a future when robots rule the road. The pioneers of the driverless car movement -- such as Google and Tesla -- are mapping the MBAs a future in which artificial intelligence and robotics will likely impact the entire job market and global economy. David Yoffie, professor of international business administration at HBS, believes such disruptive technologies are now an "essential" part of the b-school landscape. "What I'm trying to teach students is: What can these technologies deliver? And what are the challenges and opportunities for a company that does AI?" he says. David's offered his MBAs two cases on artificial intelligence (or AI) and deep learning, and reckons that many of his colleagues at HBS are bringing robots into the curriculum too: "It's a capability that MBAs need to know about," he says.

Machine intelligence capable of learning complex procedural behavior, inducing (latent) programs, and reasoning with these programs is a key to solving artificial intelligence. The problems of learning procedural behavior and program induction have been studied from different perspectives in many computer science fields such as program synthesis [1], probabilistic programming [2], inductive logic programming [3], reinforcement learning [4], and recently in deep learning. However, despite the common goal, there seems to be little communication and collaboration between the different fields focused on this problem. Recently, there have been many success stories in the deep learning community related to learning neural networks capable of using trainable memory abstractions. This has led to the development of neural networks with differentiable data structures such as Neural Turing Machines [5], Memory Networks [6], Neural Stacks [7, 8], and Hierarchical Attentive Memory [11], among others. Simultaneously, neural program induction models like Neural Program-Interpreters [9] and the Neural Programmer [10] have created much excitement in the field, promising induction of algorithmic behavior, and enabling inclusion of programming languages in the processes of execution and induction, while remaining trainable end-to-end. Trainable program induction models have the potential to make a substantial impact on many problems involving long-term memory, reasoning, and procedural execution, such as question answering, dialog, and robotics. The aim of the NAMPI workshop is to bring together researchers and practitioners from both academia and industry, in the areas of deep learning, program synthesis, probabilistic programming, inductive programming and reinforcement learning, to exchange ideas on the future of program induction with a special focus on neural network models and abstract machines. Through this workshop we look to identify common challenges, exchange ideas and lessons learned from the different fields, as well as establish a (set of) standard evaluation benchmark(s) for approaches that learn with abstraction and/or reason with induced programs.

IDF16 Intel is working on a powerful Xeon Phi processor for servers and workstations that is "optimized" for artificial-intelligence software – and it's codenamed Knights Mill. Chipzilla's data center group boss Diane Bryant flashed up this slide during this morning's Intel Developer Forum keynote in San Francisco: The chip is geared towards deep-learning applications, and is expected to be available in 2017, we're told. It will use RAM stacked into the top of its die, feature many, many cores, and have a focus on high-performance floating-point calculations – all of which should help it perform the operations necessary for high-throughput machine learning. Crucially, the Mill is not an accelerator or coprocessor: it can run x86 code and can boot and run operating systems and apps without the need of a host CPU. This sets it apart from rival chips, like Nvidia's GPUs, which need a host processor to direct them.

Deep learning is an emerging topic in artificial intelligence (AI). A subcategory of machine learning, deep learning deals with the use of neural networks to improve things like speech recognition, computer vision, and natural language processing. In the last few years, deep learning has helped forge advances in areas as diverse as object perception, machine translation, and voice recognition-all research topics that have long been difficult for AI researchers to crack. In information technology, a neural network is a system of programs and data structures that approximates the operation of the human brain. A neural network usually involves a large number of processors operating in parallel, each with its own small sphere of knowledge and access to data in its local memory.

It may not spring to mind as the first place you would turn to develop the art of conversation, but an AI firm backed by Elon Musk is hoping Reddit will help it's machine learn to converse. The OpenAI project is using a supercomputer known as a DGX-1 for its experiment, feeding popular Reddit threads to the AI for analysis, in the hope it will one day hold a conversation itself. However, the project follows a similar idea from Microsoft - which has to pull its learning chatbot after it began posting racist and offensive messages after just hours. OpenAI will feed popular Reddit threads to algorithms that have a probabilistic understanding of dialogue with the hopes it will one day hold a conversation itself. By feeding Reddit threads to DGX-1, it will hopefully read and learn a range of conversations faster than any other system has done before it, as conversations are filled with natural human language and commonly used slang.

Started exercising in the Gym 2016-04-21 17:56:02.139388 .