Algorithms are increasingly powerful, and researchers have recently been grappling with how they can operate ethically, and effectively, in society. I wrote earlier this year about a fascinating project where researchers had developed an AI capable of explaining its own workings. The researchers developed an algorithm that is not only capable of performing its task, but also translates how it achieved it into reasonably understandable English via a documentation process that is performed at each stage of its work. Despite official attempts to build this into our rules and regulations however, researchers suggest there is still a way to go. A paper, from a team from The Alan Turing Institute, suggests that the EU's General Data Protection Regulation does little to legally mandate tech companies to explain their algorithms. What's more, there are also doubts raised as to just what kind of information may be included when explanations are provided.

A groundbreaking project led by a Florida State University researcher makes an exponential advance in suicide prediction, potentially giving clinicians the ability to predict who will attempt suicide up to two years in advance with 80 percent accuracy. FSU Psychology researcher Jessica Ribeiro feels an urgency to confront this relentless problem. Shadowing her research is the ever-present awareness that 120 Americans take their lives every day, nearly 45,000 a year. Ribeiro's paper, titled "Predicting Risk of Suicide Attempts over Time through Machine Learning," will be published by the journal Clinical Psychological Science. The study offers a fascinating finding: machine learning -- a future frontier for artificial intelligence -- can predict with 80-90 percent accuracy whether someone will attempt suicide as far off as two years into the future.

Nearly two-thirds of consumers are worried about home IoT devices listening in on their conversations, according to a Gartner survey released Monday. Those jitters aren't too surprising after recent news items about TV announcers inadvertently activating viewers' Amazon Echos, or about data from digital assistants being used as evidence in criminal trials. But privacy concerns are just one hurdle smart homes still have to overcome, according to the survey. In fact, Gartner found that most consumers don't feel they need what smart homes offer. Consumer IoT is still in an early-adopter phase, Gartner concluded from the online survey, which was conducted in the second half of last year in the U.S., U.K., and Australia.

This is a single day course from 9:00am to 2:00pm. You will need to bring your laptop and have python, TensoFflow 1.0 and pandas installed before the class. You can find the instructions here. If you have any difficulties let us know before the day of the training and we will provide you with support. We will be running two parallel sessions, one for new users who have minimal or no experience with TensorFlow and another one for advanced users.

We present an algorithm for model-based reinforcement learning that combines Bayesian neural networks (BNNs) with random roll-outs and stochastic optimization for policy learning. The BNNs are trained by minimizing $\alpha$-divergences, allowing us to capture complicated statistical patterns in the transition dynamics, e.g. multi-modality and heteroskedasticity, which are usually missed by other common modeling approaches. We illustrate the performance of our method by solving a challenging benchmark where model-based approaches usually fail and by obtaining promising results in a real-world scenario for controlling a gas turbine.

We propose Edward, a Turing-complete probabilistic programming language. Edward defines two compositional representations---random variables and inference. By treating inference as a first class citizen, on a par with modeling, we show that probabilistic programming can be as flexible and computationally efficient as traditional deep learning. For flexibility, Edward makes it easy to fit the same model using a variety of composable inference methods, ranging from point estimation to variational inference to MCMC. In addition, Edward can reuse the modeling representation as part of inference, facilitating the design of rich variational models and generative adversarial networks. For efficiency, Edward is integrated into TensorFlow, providing significant speedups over existing probabilistic systems. For example, we show on a benchmark logistic regression task that Edward is at least 35x faster than Stan and 6x faster than PyMC3. Further, Edward incurs no runtime overhead: it is as fast as handwritten TensorFlow.

Given a graphical model, one essential problem is MAP inference, that is, finding the most likely configuration of states according to the model. Although this problem is NP-hard, large instances can be solved in practice. A major open question is to explain why this is true. We give a natural condition under which we can provably perform MAP inference in polynomial time. We require that the number of fractional vertices in the LP relaxation exceeding the optimal solution is bounded by a polynomial in the problem size. This resolves an open question by Dimakis, Gohari, and Wainwright. In contrast, for general LP relaxations of integer programs, known techniques can only handle a constant number of fractional vertices whose value exceeds the optimal solution. We experimentally verify this condition and demonstrate how efficient various integer programming methods are at removing fractional solutions.

We propose the spatiotemporal estimation of the demand that is a function of variable effects related to traffic, pricing and weather conditions. With respect to the methodology, a single decision tree, bootstrap-aggregated (bagged) decision trees, random forest, boosted decision trees, and artificial neural network for regression have been adapted and systematically compared using various statistics, e.g. R-square, Root Mean Square Error (RMSE), and slope. To better assess the quality of the models, they have been tested on a real case study using the data of DiDi Chuxing, the main on-demand ride-hailing service provider in China. In the current study, 199,584 time-slots describing the spatiotemporal ride-hailing demand has been extracted with an aggregated-time interval of 10 mins. All the methods are trained and validated on the basis of two independent samples from this dataset. The results revealed that boosted decision trees provide the best prediction accuracy (RMSE 16.41), while avoiding the risk of over-fitting, followed by artificial neural network (20.09), random forest (23.50), bagged decision trees (24.29) and single decision tree (33.55).

The facility location problem is widely used for summarizing large datasets and has additional applications in sensor placement, image retrieval, and clustering. One difficulty of this problem is that submodular optimization algorithms require the calculation of pairwise benefits for all items in the dataset. This is infeasible for large problems, so recent work proposed to only calculate nearest neighbor benefits. One limitation is that several strong assumptions were invoked to obtain provable approximation guarantees. In this paper we establish that these extra assumptions are not necessary---solving the sparsified problem will be almost optimal under the standard assumptions of the problem. We then analyze a different method of sparsification that is a better model for methods such as Locality Sensitive Hashing to accelerate the nearest neighbor computations and extend the use of the problem to a broader family of similarities. We validate our approach by demonstrating that it rapidly generates interpretable summaries.

We consider support recovery in the quadratic logistic regression setting - where the target depends on both p linear terms $x_i$ and up to $p^2$ quadratic terms $x_i x_j$. Quadratic terms enable prediction/modeling of higher-order effects between features and the target, but when incorporated naively may involve solving a very large regression problem. We consider the sparse case, where at most $s$ terms (linear or quadratic) are non-zero, and provide a new faster algorithm. It involves (a) identifying the weak support (i.e. all relevant variables) and (b) standard logistic regression optimization only on these chosen variables. The first step relies on a novel insight about correlation tests in the presence of non-linearity, and takes $O(pn)$ time for $n$ samples - giving potentially huge computational gains over the naive approach. Motivated by insights from the boolean case, we propose a non-linear correlation test for non-binary finite support case that involves hashing a variable and then correlating with the output variable. We also provide experimental results to demonstrate the effectiveness of our methods.