If you work on IT security in a large enterprise, chances are you're overwhelmed. Just as the capacity of your systems and networks grows and the amount of data going through them mushrooms, the number and complexity of threats grows as well. And if things weren't hard enough, government regulations put time pressure on security personnel to identify breaches and report them promptly. Even if throwing people at the problem were the answer--and it might not help at all--you probably don't have the budget for it. To deal with a large number of potential real-time threats, what you need is a lot of real-time analytical power.

Most studies about the impact of artificial intelligence (AI) on jobs and the economy have focused on developed countries such as the United States and Britain. Through my work as a scientist, technology executive and venture capitalist in the US and China, I have come to believe that the gravest threat AI poses is to emerging economies. In recent decades, China and India have presented the world with two different models on how countries can climb the development ladder. In the China model, the nation leveraged its large population and low costs to build a base of blue-collar manufacturing. The country then steadily worked its way up the value chain by producing better and more technology-intensive goods.

Google's secret prototype search engine for China reportedly links users' mobile phone numbers to what search terms they've used. This feature would allow the Chinese government to simply associate searches with individuals, thereby putting Chinese citizens at increased risk of government repression if they search for topics that their government deems politically sensitive, according to the Intercept. The mobile-focused search engine prototype, code-named Dragonfly, was first revealed last month by the Intercept. Dragonfly is said to have been conceptualized as a joint venture between Google and a Chinese-based company. Both would have the ability to update a list of verboten search words, which could include expected subjects like "human rights" and "student protest" but could also extend to search terms such as "Nobel Prize", according to the Intercept's story.

There aren't many Echo-oriented accessories beyond Amazon's own Echo Button, but that's about to change very shortly. Amazon has released a beta Alexa Gadgets Toolkit that lets hardware brands make Echo-focused Bluetooth accessories that respond to Alexa commands. You can have a cuckoo clock that responds to your Echo's wake word or a notification, a switch that releases dog food after an alarm, or a chime that sounds when time's up. A future update will even allow visual interaction with music -- it's easy to see a lamp that pulses in sync with Amazon Music tracks. Child-oriented updates will also let developers build gadgets that include compatible kid-friendly skills.

We introduce a dynamic artificial neural network-based (ANN) adaptive inference process, which learns temporal predictive models of dynamical systems. We term the process REPRISE, a REtrospective and PRospective Inference SchEme. REPRISE infers the unobservable contextual state that best explains its recently encountered sensorimotor experiences as well as accompanying, context-dependent temporal predictive models retrospectively. Meanwhile, it executes prospective inference, optimizing upcoming motor activities in a goal-directed manner. In a first implementation, a recurrent neural network (RNN) is trained to learn a temporal forward model, which predicts the sensorimotor contingencies of different simulated dynamic vehicles. The RNN is augmented with contextual neurons, which enable the compact encoding of distinct, but related sensorimotor dynamics. We show that REPRISE is able to concurrently learn to separate and approximate the encountered sensorimotor dynamics. Moreover, we show that REPRISE can exploit the learned model to induce goal-directed, model-predictive control, that is, approximate active inference: Given a goal state, the system imagines a motor command sequence optimizing it with the prospective objective to minimize the distance to a given goal. Meanwhile, the system evaluates the encountered sensorimotor contingencies retrospectively, adapting its neural hidden states for maintaining model coherence. The RNN activities thus continuously imagine the upcoming future and reflect on the recent past, optimizing both, hidden state and motor activities. In conclusion, the combination of temporal predictive structures with modulatory, generative encodings offers a way to develop compact event codes, which selectively activate particular types of sensorimotor event-specific dynamics.

Abstract--The ability to estimate joint, conditional and marginal probability distributions over some set of variables is of great utility for many common machine learning tasks. However, estimating these distributions can be challenging, particularly in the case of data containing a mix of discrete and continuous variables. This paper presents a nonparametric method for estimating these distributions directly from a dataset. The data are first represented as a graph consisting of object nodes and attribute value nodes. Depending on the distribution to be estimated, an appropriate eigenvector equation is then constructed. This equation is then solved to find the corresponding stationary distribution of the graph, from which the required distributions can then be estimated and sampled from. The paper demonstrates how the method can be applied to many common machine learning tasks including classification, regression, missing value imputation, outlier detection, random vector generation, and clustering. Being able to estimate joint, conditional and marginal probabilities from some dataset allows a broad range of useful tasks to be performed. For example, classification and regression involve predicting the value of some target variable conditional on the values of the other variables. If we can sample values from the estimated distributions, we could perform random vector generation by generating full random vectors that display the same correlations as the vectors (i.e., data points) in the original data [4], [5]. If we can estimate the joint distribution for the full dataset, then we should also be able to do this for subsets of data, leading to the use of Expectation-Maximization [6] to cluster the data [7]. Taken together, these activities form a large chunk of the tasks commonly used in machine learning. All of this depends, of course, on being able to estimate the various probabilities, and this is particularly challenging on datasets containing a complex mix of continuous and discrete variables.

Service robots need to show appropriate social behavior in order to deploy in social environments such as healthcare, education, retail, etc. Some of the main capabilities that robots should have are navigation and conversational skill. If the person is impatient, he might want a robot to navigate faster and vice versa. Linguistic features that derive politeness can provide social cues about person's patient and impatient behavior. The novelty presented in this paper is to dynamically incorporate politeness in robotic dialogue systems for navigation. Understanding the politeness in users' speech can be used to modulate the robot behavior and responses. Therefore, we developed a dialogue system to navigate in an indoor environment, which produces different robot behaviors and responses based on users' intention and degree of politeness. We deploy and test our system with the Pepper robot that adapts to the changes in user's politeness.

The score function estimator is widely used for estimating gradients of stochastic objectives in stochastic computation graphs (SCG), eg, in reinforcement learning and meta-learning. While deriving the first-order gradient estimators by differentiating a surrogate loss (SL) objective is computationally and conceptually simple, using the same approach for higher-order derivatives is more challenging. Firstly, analytically deriving and implementing such estimators is laborious and not compliant with automatic differentiation. Secondly, repeatedly applying SL to construct new objectives for each order derivative involves increasingly cumbersome graph manipulations. Lastly, to match the first-order gradient under differentiation, SL treats part of the cost as a fixed sample, which we show leads to missing and wrong terms for estimators of higher-order derivatives. To address all these shortcomings in a unified way, we introduce DiCE, which provides a single objective that can be differentiated repeatedly, generating correct estimators of derivatives of any order in SCGs. Unlike SL, DiCE relies on automatic differentiation for performing the requisite graph manipulations. We verify the correctness of DiCE both through a proof and numerical evaluation of the DiCE derivative estimates. We also use DiCE to propose and evaluate a novel approach for multi-agent learning. Our code is available at https://www.github.com/alshedivat/lola.

The computation of convolution layers in deep neural networks typically rely on high performance routines that trade space for time by using additional memory (either for packing purposes or required as part of the algorithm) to improve performance. The problems with such an approach are two-fold. First, these routines incur additional memory overhead which reduces the overall size of the network that can fit on embedded devices with limited memory capacity. Second, these high performance routines were not optimized for performing convolution, which means that the performance obtained is usually less than conventionally expected. In this paper, we demonstrate that direct convolution, when implemented correctly, eliminates all memory overhead, and yields performance that is between 10% to 400% times better than existing high performance implementations of convolution layers on conventional and embedded CPU architectures. We also show that a high performance direct convolution exhibits better scaling performance, i.e. suffers less performance drop, when increasing the number of threads.

Forecasting the movements of stock prices is one the most challenging problems in financial markets analysis. In this paper, we use Machine Learning (ML) algorithms for the prediction of future price movements using limit order book data. Two different sets of features are combined and evaluated: handcrafted features based on the raw order book data and features extracted by ML algorithms, resulting in feature vectors with highly variant dimensionalities. Three classifiers are evaluated using combinations of these sets of features on two different evaluation setups and three prediction scenarios. Even though the large scale and high frequency nature of the limit order book poses several challenges, the scope of the conducted experiments and the significance of the experimental results indicate that Machine Learning highly befits this task carving the path towards future research in this field. Keywords: Machine Learning, limit order book, feature extraction, mid price forecasting 1. Introduction Forecasting of financial time series is a very challenging problem and has attracted scientific interest in the past few decades. Due to the inherently noisy and non-stationary nature of financial time series, statistical models are unsuitable for the task of modeling and forecasting such data. However, the lack of appropriate training and regularization algorithms for Neural Networks at the time, such as the dropout technique [6], rendered them susceptible to over fitting the training data. Support Vector Machines were deemed as better candidates for this task, as their solution implicitly involves the generalization error. The development of effective and efficient training algorithms for deeper architectures [7], in conjunction with the improved results such models presented, steered scientific interests towards Deep Learning techniques in many domains. Deep Learning methods are capable of modeling highly nonlinear, very complex data, making them suitable for application to financial data [8], as well as time series forecasting [9]. Furthermore, ML techniques which perform feature extraction may uncover robust features, better-suited to the specific task at hand. Autoencoders [10], are Neural Networks which learn new features extracted from the original input space, which can be used to enhance the performance of various tasks, such as classification or regression. Bag-of-Features (BoF) models comprise another feature extraction method that can be used to extract representations of objects described by multiple feature vectors, such as time-series [11].