As you contemplate the introduction of artificial intelligence, you should articulate what mix of three approaches works best for you. Jeff Heepke knows where to plant corn on his 4,500-acre farm in Illinois because of artificial intelligence (AI). He uses a smartphone app called Climate Basic, which divides Heepke's farmland (and, in fact, the entire continental U.S.) into plots that are 10 meters square. The app draws on local temperature and erosion records, expected precipitation, soil quality and other agricultural data to determine how to maximize yields for each plot. If a rainy cold front is expected to pass by, Heepke knows which areas to avoid watering or irrigating that afternoon. As the U.S. Department of Agriculture noted, this use of artificial intelligence across the industry has produced the largest crops in the country's history. Climate Corp., the Silicon Valley–based developer of Climate Basic, also offers a more advanced AI app that operates autonomously. If a storm hits a region, or a drought occurs, it lowers local yield numbers.

Motherboard has called Ray Kurzweil "a prophet of both techno-doom and techno-salvation." With a little wiggle room given to the timelines the author, inventor, computer scientist, futurist, and director of engineering at Google provides, a full 86 percent of his predictions -- including the fall of the Soviet Union, the growth of the internet, and the ability of computers to beat humans at chess -- have come to fruition. Kurzweil continues to share his visions for the future, and his latest prediction was made at the most recent SXSW Conference, where he claimed that the Singularity -- the moment when technology becomes smarter than humans -- will happen by 2045. Sixteen years prior to that, it will be just as smart as us. As he told Futurism, "2029 is the consistent date I have predicted for when an AI will pass a valid Turing test and therefore achieve human levels of intelligence."

In San Francisco, a young engineer hopes to "optimise" his life through sensors that track his heart rate, respiration and sleep cycle. In Copenhagen, a bus running two minutes behind schedule transmits its location and passenger count to the municipal traffic signal network, which extends the time of the green light at each of the next three intersections long enough for its driver to make up some time. In Davao City in the Philippines, an unsecured webcam overlooks the storeroom of a fast food stand, allowing anyone to peer in on all its comings and goings. What links these wildly different circumstances is a vision of connected devices now being sold to us as the "internet of things". The technologist Mike Kuniavsky, a pioneer of this idea, characterises it as a state of being in which "computation and data communication [are] embedded in, and distributed through, our entire environment".

Outside of its robotic remote control cars, last year Anki released a palm-sized robot companion called Cozmo. As adorable as it is intelligent, it rolls around on its tank treads using its arm to interact with the environment and a built-in camera to recognize you or even your pets. Since its debut last fall it has received a few software updates, but was only on sale in the US. Starting next month that will change when it goes on sale in Canada for $250 CA, before it expands to the UK, France, Germany and Nordic countries in September. Even if you live in the US, starting today you can order a $180 Collector's Edition of the bot that comes in a new Liquid Metal finish.

In the United States, job demographic data from censuses since the 1900s reveal a startling fact. Despite the two post-Industrial revolutions of electricity and computers, the occupations with the largest employment numbers are still jobs for drivers, retail, cashiers, secretaries, janitors etc, i.e. old professions needing simple skills and mostly repetitive work. This lack of transition to "newer" jobs is a global phenomenon, especially in the global south. India, for example, has half of the working population doing agriculture. One must grasp the significance of Artificial Intelligence (AI) in this context.

As deep learning is gaining in popularity, creative applications are gaining traction as well. Looking at music generation through deep learning, new algorithms and songs are popping up on a weekly basis. In this post we will go over six major players in the field, and point out some difficult challenges these systems still face. GitHub links are provided for those who are interested in the technical details (or if you're looking to generate some music of your own). Magenta is Google's open source deep learning music project.

AI is slowly seeping into our lives in profound ways. The way we think about AI is coloured by popular culture and science fiction. Many feel that if they could make that science fiction real then we wouldn't have to worry about messy kinds of stuff in society. My concern is not about the singularity that experts like Ray Kurzweil have talked about in his book Singularity Is Near nor Hollywood movies like Transcendence, which paint a dystopian view of a future transformed by AI. The human brain is a remarkable piece of engineering and fears that AI might take over human brain is not imminent.

Chemical space is so large that brute force searches for new interesting molecules are infeasible. High-throughput virtual screening via computer cluster simulations can speed up the discovery process by collecting very large amounts of data in parallel, e.g., up to hundreds or thousands of parallel measurements. Bayesian optimization (BO) can produce additional acceleration by sequentially identifying the most useful simulations or experiments to be performed next. However, current BO methods cannot scale to the large numbers of parallel measurements and the massive libraries of molecules currently used in high-throughput screening. Here, we propose a scalable solution based on a parallel and distributed implementation of Thompson sampling (PDTS). We show that, in small scale problems, PDTS performs similarly as parallel expected improvement (EI), a batch version of the most widely used BO heuristic. Additionally, in settings where parallel EI does not scale, PDTS outperforms other scalable baselines such as a greedy search, $\epsilon$-greedy approaches and a random search method. These results show that PDTS is a successful solution for large-scale parallel BO.

It is challenging to develop stochastic gradient based scalable inference for deep discrete latent variable models (LVMs), due to the difficulties in not only computing the gradients, but also adapting the step sizes to different latent factors and hidden layers. For the Poisson gamma belief network (PGBN), a recently proposed deep discrete LVM, we derive an alternative representation that is referred to as deep latent Dirichlet allocation (DLDA). Exploiting data augmentation and marginalization techniques, we derive a block-diagonal Fisher information matrix and its inverse for the simplex-constrained global model parameters of DLDA. Exploiting that Fisher information matrix with stochastic gradient MCMC, we present topic-layer-adaptive stochastic gradient Riemannian (TLASGR) MCMC that jointly learns simplex-constrained global parameters across all layers and topics, with topic and layer specific learning rates. State-of-the-art results are demonstrated on big data sets.

We develop a family of reformulations of an arbitrary consistent linear system into a stochastic problem. The reformulations are governed by two user-defined parameters: a positive definite matrix defining a norm, and an arbitrary discrete or continuous distribution over random matrices. Our reformulation has several equivalent interpretations, allowing for researchers from various communities to leverage their domain specific insights. In particular, our reformulation can be equivalently seen as a stochastic optimization problem, stochastic linear system, stochastic fixed point problem and a probabilistic intersection problem. We prove sufficient, and necessary and sufficient conditions for the reformulation to be exact. Further, we propose and analyze three stochastic algorithms for solving the reformulated problem---basic, parallel and accelerated methods---with global linear convergence rates. The rates can be interpreted as condition numbers of a matrix which depends on the system matrix and on the reformulation parameters. This gives rise to a new phenomenon which we call stochastic preconditioning, and which refers to the problem of finding parameters (matrix and distribution) leading to a sufficiently small condition number. Our basic method can be equivalently interpreted as stochastic gradient descent, stochastic Newton method, stochastic proximal point method, stochastic fixed point method, and stochastic projection method, with fixed stepsize (relaxation parameter), applied to the reformulations.