Over-parametrization of deep neural networks has recently been shown to be key to their successful training. However, it also renders them prone to overfitting and makes them expensive to store and train. Tensor regression networks significantly reduce the number of effective parameters in deep neural networks while retaining accuracy and the ease of training. They replace the flattening and fully-connected layers with a tensor regression layer, where the regression weights are expressed through the factors of a low-rank tensor decomposition. In this paper, to further improve tensor regression networks, we propose a novel stochastic rank-regularization. It consists of a novel randomized tensor sketching method to approximate the weights of tensor regression layers. We theoretically and empirically establish the link between our proposed stochastic rank-regularization and the dropout on low-rank tensor regression. Extensive experimental results with both synthetic data and real world datasets (i.e., CIFAR-100 and the UK Biobank brain MRI dataset) support that the proposed approach i) improves performance in both classification and regression tasks, ii) decreases overfitting, iii) leads to more stable training and iv) improves robustness to adversarial attacks and random noise.

Education was traditionally seen as an enabler of social mobility. In other words, if you were from a low-income family, you could improve your financial and social standing by getting an education. And for a while, it worked, but inequality is on the rise again. These days, college and university degrees are, at least in developed countries, a dime a dozen and you need a postgraduate qualification to get an entry-level position. The advantage of education to boost social mobility is more noticeable in developing countries where the demand for highly educated individuals outstrips the supply.

While artificial intelligence (AI) has the potential to solve an incredible spectrum of problems and challenges in our lives, our work and our world, there is a widening disconnect between the people who are introducing and deploying AI-based solutions and those who set policies for when and how these solutions are used. Much has been written about one consequence of this disconnect--algorithmic bias in AI systems, in which machine learning algorithms trained on data that reflects historical discrimination replicate and even magnify it. But there's another pressing issue: There are many missed opportunities to use AI for the good of many. Just as AI systems susceptible to bias are a problem, so too is inadequate focus on contributions that improve the lives of marginalized communities, such as Black and brown individuals, economically vulnerable populations and many other groups whose interests are underserved in society. If teams that set research directions, write algorithms or deploy them are made up of individuals with similar backgrounds and experiences, then we will end up with research that is to the benefit of a similarly narrow and already privileged subset of society.

The year 2018 has been an inflection point for machine learning models handling text (or more accurately, Natural Language Processing or NLP for short). Our conceptual understanding of how best to represent words and sentences in a way that best captures underlying meanings and relationships is rapidly evolving. Moreover, the NLP community has been putting forward incredibly powerful components that you can freely download and use in your own models and pipelines (It's been referred to as NLP's ImageNet moment, referencing how years ago similar developments accelerated the development of machine learning in Computer Vision tasks). But I couldn't think of anything else..) One of the latest milestones in this development is the release of BERT, an event described as marking the beginning of a new era in NLP. BERT is a model that broke several records for how well models can handle language-based tasks.

Factorization Machine (FM) is a widely used supervised learning approach by effectively modeling of feature interactions. Despite the successful application of FM and its many deep learning variants, treating every feature interaction fairly may degrade the performance. For example, the interactions of a useless feature may introduce noises; the importance of a feature may also differ when interacting with different features. In this work, we propose a novel model named \emph{Interaction-aware Factorization Machine} (IFM) by introducing Interaction-Aware Mechanism (IAM), which comprises the \emph{feature aspect} and the \emph{field aspect}, to learn flexible interactions on two levels. The feature aspect learns feature interaction importance via an attention network while the field aspect learns the feature interaction effect as a parametric similarity of the feature interaction vector and the corresponding field interaction prototype. IFM introduces more structured control and learns feature interaction importance in a stratified manner, which allows for more leverage in tweaking the interactions on both feature-wise and field-wise levels. Besides, we give a more generalized architecture and propose Interaction-aware Neural Network (INN) and DeepIFM to capture higher-order interactions. To further improve both the performance and efficiency of IFM, a sampling scheme is developed to select interactions based on the field aspect importance. The experimental results from two well-known datasets show the superiority of the proposed models over the state-of-the-art methods.

We propose a technique to develop (and localize in) topological maps from light detection and ranging (Lidar) data. Localizing an autonomous vehicle with respect to a reference map in real-time is crucial for its safe operation. Owing to the rich information provided by Lidar sensors, these are emerging as a promising choice for this task. However, since a Lidar outputs a large amount of data every fraction of a second, it is progressively harder to process the information in real-time. Consequently, current systems have migrated towards faster alternatives at the expense of accuracy. To overcome this inherent trade-off between latency and accuracy, we propose a technique to develop topological maps from Lidar data using the orthogonal Tucker3 tensor decomposition. Our experimental evaluations demonstrate that in addition to achieving a high compression ratio as compared to full data, the proposed technique, $\textit{TensorMap}$, also accurately detects the position of the vehicle in a graph-based representation of a map. We also analyze the robustness of the proposed technique to Gaussian and translational noise, thus initiating explorations into potential applications of tensor decompositions in Lidar data analysis.

Global enterprise software giant Oracle will up the delivery of emerging technologies in the South African market as most local businesses are proceeding with their digital transformation initiatives. So says Arun Khehar, Oracle's senior vice-president for applications sales for the Eastern Central Europe, Middle East and Africa region. In an e-mail interview with ITWeb, Khehar said the world of business is changing at an unimaginable rate. Emerging technologies are changing the way companies do business at every level, across every function, he notes. These technologies have the ability to learn and adapt and change, giving the organisation the space it needs to focus on its core business and growth." According to Khehar, the adoption of cloud technology has come around far quicker than anyone anticipated. "In South Africa, we are seeing the transformation at CEO level in favour of cloud adoption, in particular hybrid solutions.

Multilateral solutions remain the best way to address threats to international peace and security. This is true when we talk about the situation in Syria, Libya, Yemen or the Middle East peace process.

Big data and artificial intelligence offer new ways to solve age-old problems, such as how to feed a growing population. Under our Grand Challenge program, RTI is funding research and analysis based on these technologies to promote agricultural resilience and food security in Rwanda. Most of Rwanda's crop production comes from smallholder farms. The country's agriculture officials have historically had insufficient data on where crops are cultivated or how much yield to expect--a hindrance when planning for the future of their growing country. The government is looking to technology for a solution. Building on our previous work with emerging technologies, machine learning, economics, and agriculture, we are developing a new decision-making tool that can be applied in Rwanda and beyond.

ONCE THOUGHT OF AS A NICHE TOY for early adopters, drones can now be found buzzing over parks, in select cities, and are even being increasingly used for video production as the popularity of aerial photography soars. However, drones aren't only for fun and entertainment, and the high-pitched hum of their spinning propellers could replace the wail of ambulance sirens for global citizens as drones are put to work for humanitarian purposes. In March of 2017, DJI, the manufacturers of the most popular commercial drones, published a report about drones' life-saving capabilities, citing cases in which drones manned by volunteers or bystanders were used in emergency situations like floods and avalanches, resulting in 59 life-saving rescues in China, Canada, the U.S., and Turkey. Given that it takes 25 people 35 hours to search one square mile for missing persons, compared to the 30 minutes it takes a drone to cover the same area, regardless of treacherous conditions on the ground, drones are uniquely suited for search and rescue, even when piloted by hobbyists. Based on the increasing trend of drone use in the last 10 months covered by the report, DJI estimated that drones would be directly responsible for saving at least one person per week in the future.