The drone attack by a little-known armed group in Iraq on the United Arab Emirates (UAE) this week has raised questions about Baghdad's involvement in regional tensions between Iran-backed Houthi rebels in Yemen and the Saudi-led coalition. Alwiyat al-Waad al-Haq (AWH), or the True Promise Brigades, claimed responsibility for the strike on the UAE on Wednesday, saying in a statement it launched "four drones targeting vital facilities in Abu Dhabi" in retaliation for the Emirates' policies in Iraq and Yemen. Several analysts linked the strikes to a shadowy militia Kataib Hezbollah (KH), a powerful Iran-backed Shia armed group in Iraq that has been listed by the United States as a "terrorist organisation". The incident brought to light that the UAE was now being targeted from its north and south, after three recent attacks launched by Houthi rebels in Yemen. Following the drone strikes, Iraqi Shia leader Muqtada al-Sadr condemned the attack in a statement, saying some "terrorist outlaws" have dragged Iraq into a "dangerous regional war" by targeting a Gulf state.

Over the last few years, neural networks have started penetrating safety critical systems to take decisions in robots, rockets, autonomous driving car, etc. A problem is that these critical systems often have limited computing resources. Often, they use the fixed-point arithmetic for its many advantages (rapidity, compatibility with small memory devices.) In this article, a new technique is introduced to tune the formats (precision) of already trained neural networks using fixed-point arithmetic, which can be implemented using integer operations only. The new optimized neural network computes the output with fixed-point numbers without modifying the accuracy up to a threshold fixed by the user. A fixed-point code is synthesized for the new optimized neural network ensuring the respect of the threshold for any input vector belonging the range [xmin, xmax] determined during the analysis. From a technical point of view, we do a preliminary analysis of our floating neural network to determine the worst cases, then we generate a system of linear constraints among integer variables that we can solve by linear programming. The solution of this system is the new fixed-point format of each neuron. The experimental results obtained show the efficiency of our method which can ensure that the new fixed-point neural network has the same behavior as the initial floating-point neural network.

The impact of Artificial Intelligence does not depend only on fundamental research and technological developments, but for a large part on how these systems are introduced into society and used in everyday situations. Even though AI is traditionally associated with rational decision making, understanding and shaping the societal impact of AI in all its facets requires a relational perspective. A rational approach to AI, where computational algorithms drive decision making independent of human intervention, insights and emotions, has shown to result in bias and exclusion, laying bare societal vulnerabilities and insecurities. A relational approach, that focus on the relational nature of things, is needed to deal with the ethical, legal, societal, cultural, and environmental implications of AI. A relational approach to AI recognises that objective and rational reasoning cannot does not always result in the 'right' way to proceed because what is 'right' depends on the dynamics of the situation in which the decision is taken, and that rather than solving ethical problems the focus of design and use of AI must be on asking the ethical question. In this position paper, I start with a general discussion of current conceptualisations of AI followed by an overview of existing approaches to governance and responsible development and use of AI. Then, I reflect over what should be the bases of a social paradigm for AI and how this should be embedded in relational, feminist and non-Western philosophies, in particular the Ubuntu philosophy.

Stock Market can be easily seen as one of the most attractive places for investors, but it is also very complex in terms of making trading decisions. Predicting the market is a risky venture because of the uncertainties and nonlinear nature of the market. Deciding on the right time to trade is key to every successful trader as it can lead to either a huge gain of money or totally a loss in investment that will be recorded as a careless trade. The aim of this research is to develop a prediction system for stock market using Fuzzy Logic Type2 which will handle these uncertainties and complexities of human behaviour in general when it comes to buy, hold or sell decision making in stock trading. The proposed system was developed using VB.NET programming language as frontend and Microsoft SQL Server as backend. A total of four different technical indicators were selected for this research. The selected indicators are the Relative Strength Index, William Average, Moving Average Convergence and Divergence, and Stochastic Oscillator. These indicators serve as input variable to the Fuzzy System. The MACD and SO are deployed as primary indicators, while the RSI and WA are used as secondary indicators. Fibonacci retracement ratio was adopted for the secondary indicators to determine their support and resistance level in terms of making trading decisions. The input variables to the Fuzzy System is fuzzified to Low, Medium, and High using the Triangular and Gaussian Membership Function. The Mamdani Type Fuzzy Inference rules were used for combining the trading rules for each input variable to the fuzzy system. The developed system was tested using sample data collected from ten different companies listed on the Nigerian Stock Exchange for a total of fifty two periods. The dataset collected are Opening, High, Low, and Closing prices of each security.

The study of generalisation properties of deep neural networks is arguably one of the topics gaining most traction in deep learning theory (see, e.g., the recent surveys Kawaguchi et al., 2020; Jiang et al., 2020b). In particular, a characterisation of out-of-sample generalisation is essential to understand where trained neural networks are likely to succeed or to fail, as evidenced by the recent NeurIPS 2020 competition "Predicting Generalization in Deep Learning" (Jiang et al., 2020a). One stream of this joint effort, which the present paper contributes to, is dedicated to the study of shallow neural networks, potentially paving the way to insights on deeper architectures.

In addition to the growing use of AI within surveillance systems across Africa, AI has also been integrated into weapon systems. Most prominently, lethal autonomous weapons systems use real-time sensor data coupled with AI and machine learning algorithms to "select and engage targets without further intervention by a human operator." Depending on how that definition is interpreted, the first use of a lethal autonomous weapon system in combat may have taken place on African soil in March 2020. That month, logistics units belonging to the armed forces of the Libyan warlord Khalifa Haftar came under attack by Turkish-made STM Kargu-2 drones as they fled Tripoli. According to a United Nations report, the Kargu-2 represented a lethal autonomous weapons system because it had been "programmed to attack targets without requiring data connectivity between the operator and munition."

In selection processes such as hiring, promotion, and college admissions, implicit bias toward socially-salient attributes such as race, gender, or sexual orientation of candidates is known to produce persistent inequality and reduce aggregate utility for the decision maker. Interventions such as the Rooney Rule and its generalizations, which require the decision maker to select at least a specified number of individuals from each affected group, have been proposed to mitigate the adverse effects of implicit bias in selection. Recent works have established that such lower-bound constraints can be very effective in improving aggregate utility in the case when each individual belongs to at most one affected group. However, in several settings, individuals may belong to multiple affected groups and, consequently, face more extreme implicit bias due to this intersectionality. We consider independently drawn utilities and show that, in the intersectional case, the aforementioned non-intersectional constraints can only recover part of the total utility achievable in the absence of implicit bias. On the other hand, we show that if one includes appropriate lower-bound constraints on the intersections, almost all the utility achievable in the absence of implicit bias can be recovered. Thus, intersectional constraints can offer a significant advantage over a reductionist dimension-by-dimension non-intersectional approach to reducing inequality.

We study algorithms for learning low-rank neural networks -- networks where the weight parameters are re-parameterized by products of two low-rank matrices. First, we present a provably efficient algorithm which learns an optimal low-rank approximation to a single-hidden-layer ReLU network up to additive error $\epsilon$ with probability $\ge 1 - \delta$, given access to noiseless samples with Gaussian marginals in polynomial time and samples. Thus, we provide the first example of an algorithm which can efficiently learn a neural network up to additive error without assuming the ground truth is realizable. To solve this problem, we introduce an efficient SVD-based $\textit{Nonlinear Kernel Projection}$ algorithm for solving a nonlinear low-rank approximation problem over Gaussian space. Inspired by the efficiency of our algorithm, we propose a novel low-rank initialization framework for training low-rank $\textit{deep}$ networks, and prove that for ReLU networks, the gap between our method and existing schemes widens as the desired rank of the approximating weights decreases, or as the dimension of the inputs increases (the latter point holds when network width is superlinear in dimension). Finally, we validate our theory by training ResNet and EfficientNet models on ImageNet.

Writing about A.I. has given me the opportunity to think more about datascience, programming and technology trends for new professionals. Picture this, a young woman in Africa starts her her own startup to meet unmet needs, the year is 2030. Will she need to even hire developers or can her small IT team handle the engineering of the product? Are we heading to a world where the software revolution will truly be democratized? This is an Op-ed about an emerging and somewhat speculative field that is showing real signs of potential.

Keyword localisation is the task of finding where in a speech utterance a given query keyword occurs. We investigate to what extent keyword localisation is possible using a visually grounded speech (VGS) model. VGS models are trained on unlabelled images paired with spoken captions. These models are therefore self-supervised -- trained without any explicit textual label or location information. To obtain training targets, we first tag training images with soft text labels using a pretrained visual classifier with a fixed vocabulary. This enables a VGS model to predict the presence of a written keyword in an utterance, but not its location. We consider four ways to equip VGS models with localisations capabilities. Two of these -- a saliency approach and input masking -- can be applied to an arbitrary prediction model after training, while the other two -- attention and a score aggregation approach -- are incorporated directly into the structure of the model. Masked-based localisation gives some of the best reported localisation scores from a VGS model, with an accuracy of 57% when the system knows that a keyword occurs in an utterance and need to predict its location. In a setting where localisation is performed after detection, an $F_1$ of 25% is achieved, and in a setting where a keyword spotting ranking pass is first performed, we get a localisation P@10 of 32%. While these scores are modest compared to the idealised setting with unordered bag-of-word-supervision (from transcriptions), these models do not receive any textual or location supervision. Further analyses show that these models are limited by the first detection or ranking pass. Moreover, individual keyword localisation performance is correlated with the tagging performance from the visual classifier. We also show qualitatively how and where semantic mistakes occur, e.g. that the model locates surfer when queried with ocean.