The dark web is a difficult place to police. Comprising millions of websites shrouded in anonymity, it is an online playground for dangerous criminals and political plotters - anyone trying to evade authority or do the wrong thing. But one Australian Federal Police officer hopes to develop an artificially intelligent (AI) "crawler" that could scan the dark web for illegal activity and alert authorities to anything suspicious. If successful, the AI crawler would - for example - make it easier and faster to track down paedophiles, a task that at present is time-consuming and requires investigators to look at thousands of confronting images and material. The officer, Janis Dalins, has been developing the crawler as part of his PhD.

Ongoing developments in neural network models are continually advancing the state-of-the-art in terms of system accuracy. However, the predicted labels should not be regarded as the only core output; also important is a well calibrated estimate of the prediction uncertainty. Such estimates and their calibration is critical in relation to robust handling of out of distribution events not observed in training data. Despite their obvious aforementioned advantage in relation to accuracy, contemporary neural networks can, generally, be regarded as poorly calibrated and as such do not produce reliable output probability estimates. Further, while post-processing calibration solutions can be found in the relevant literature, these tend to be for systems performing classification. In this regard, we herein present a method for acquiring calibrated predictions intervals for neural network regressors by posing the regression task as a multi-class classification problem and applying one of three proposed calibration methods on the classifiers' output. Testing our method on two exemplar tasks - speaker age prediction and signal-to-noise ratio estimation - indicates both the suitability of the classification-based regression models and that post-processing by our proposed empirical calibration or temperature scaling methods yields well calibrated prediction intervals. The code for computing calibrated predicted intervals is publicly available.

Domestic violence (DV) is a global social and public health issue that is highly gendered. Being able to accurately predict DV recidivism, i.e., re-offending of a previously convicted offender, can speed up and improve risk assessment procedures for police and front-line agencies, better protect victims of DV, and potentially prevent future re-occurrences of DV. Previous work in DV recidivism has employed different classification techniques, including decision tree (DT) induction and logistic regression, where the main focus was on achieving high prediction accuracy. As a result, even the diagrams of trained DTs were often too difficult to interpret due to their size and complexity, making decision-making challenging. Given there is often a trade-off between model accuracy and interpretability, in this work our aim is to employ DT induction to obtain both interpretable trees as well as high prediction accuracy. Specifically, we implement and evaluate different approaches to deal with class imbalance as well as feature selection. Compared to previous work in DV recidivism prediction that employed logistic regression, our approach can achieve comparable area under the ROC curve results by using only 3 of 11 available features and generating understandable decision trees that contain only 4 leaf nodes.

Learning the latent representation of data in unsupervised fashion is a very interesting process that provides relevant features for enhancing the performance of a classifier. For speech emotion recognition tasks, generating effective features is crucial. Currently, handcrafted features are mostly used for speech emotion recognition, however, features learned automatically using deep learning have shown strong success in many problems, especially in image processing. In particular, deep generative models such as Variational Autoencoders (VAEs) have gained enormous success for generating features for natural images. Inspired by this, we propose VAEs for deriving the latent representation of speech signals and use this representation to classify emotions. To the best of our knowledge, we are the first to propose VAEs for speech emotion classification. Evaluations on the IEMOCAP dataset demonstrate that features learned by VAEs can produce state-of-the-art results for speech emotion classification.

With an operating footprint of up to 50km from the mining pit to iron ore carriers, it was easy for Citic Pacific Mining, Australia's largest magnetite mining company, to lose track of its assets, such as light vehicles, buses and service trucks. Find out how to draw up a battle plan for securing connected devices and the key areas to target. You forgot to provide an Email Address. This email address doesn't appear to be valid. This email address is already registered.

New Zealanders are used to the idea of automation and industrial robots in manufacturing and some homes have those disc shaped vacuum cleaners roaming the house of their own volition in order to keep everything spick and span. Many people have exchanged messages with chat bots online in order to get a few questions answered. But a New Zealand company, Soul Machines, has taken the chat bot idea to the next level and developed so called "digital humans." Just over a month ago, the Natwest Bank in the UK started testing an artificial intelligence-powered "digital human" called Cora who will converse with customers from a terminal in bank branches, with the aim of cutting down on waiting times. The bank hopes Cora's artificial intelligence will eventually expand to answering hundreds of different questions, but at the same time insists the avatar is there to complement, not replace humans.

The ethics of artificial intelligence will be critical to the success of AI going forward, a Microsoft leader and a keynote speaker at the AI Day event in Auckland next week says. Steve Guggenheimer, corporate vice president of Microsoft's AI Business, says that given AI has the potential to reshape not just industries and governments, but society as a whole. "Working on the ethics of the use of AI, from the beginning, in key areas like transparency, accountability, privacy and bias will be crucial to the success of AI going forward. "There is a strong focus on the ethical implications of the AI systems that are being built and deployed." The European Commission's group on ethics in science and new technologies recently warned that existing efforts to develop solutions to the ethical, societal and legal challenges AI presents are a'patchwork of disparate initiatives'. It added that uncoordinated, unbalanced approaches in the regulation of AI risked ethics shopping, resulting in the relocation of AI development and use to regions with lower ethical standards. AI Day on March 28 is being organised by NewZealand.AI and the AI Forum NZ, which is part of the NZTech Alliance, bringing together 14 national tech communities, more than 500 organisations and more than 100,000 employees to help create a more prosperous New Zealand underpinned by technology. Guggenheimer says one important element around the adoption of AI is the focus on having AI help to amplify human capabilities and allow them to do more versus simply replacing people and functions. "As AI is adopted by various organisations we are starting to see a few trends occurring.

Fish, like most animals, have a pretty good idea of which other animals they're cool with, and which animals they're not. Very few animals are cool with humans, and fish are no exception--maybe they're afraid, maybe they're curious, and maybe they'll pretend to ignore you until you get too close, but in any of these cases, your presence is affecting their behavior. We've seen many clever examples of animal behavior researchers using robots to study their subjects up close with minimal disruption, and in a paper published in Science Robotics today, roboticists at MIT's Computer Science and Artificial Intelligence Laboratory describe a new kind of soft robotic spy fish that can more or less blend right in with everything else living on a coral reef. SoFi, MIT's soft robotic fish, is designed to provide close-range, minimally disruptive observations of all the fascinating and adorable animals that live underwater. The MIT roboticists (Robert K. Katzschmann, Joseph DelPreto, Robert MacCurdy, and Professor Daniela Rus) were careful to make SoFi as similar in size and behavior to a real fish as was possible, but they also had to make it completely self-contained and actually useful--SoFi isn't just a proof-of-concept for the design of a biomimetic robotic fish, it's a real research tool, with a friendly control system, and practical battery life.

A robotic fish might be able to unlock secrets about marine life that is hard for researchers to access, according to a new report. Scientists from the Massachusetts Institute of Technology (MIT) have created a robotic fish called SoFi that was tested in Fiji. SoFi was able to swim more than 50 feet below the surface of the water and for 40 minutes nonstop. The researchers behind the new study, published in Science Robotics, say robotic fish technology could help scientists learn more about organisms that are hard for humans to get to to study. MIT researchers developed a robotic fish that can swim alongside real fish and take photographs of marine life that is hard for humans to access.

It looks and moves like a real fish, flapping its tail from side to side. But this fish is controlled by a human diver via a waterproofed Super Nintendo controller and an ultrasound transmitter. SoFi, the soft robotic fish, has been designed to let researchers study marine life up close. Remotely-operated or autonomous submersibles are usually propeller-driven, which tends to disturb wildlife. Videos of test dives in Fiji's Rainbow Reef show SoFi skirting over coral alongside real fish, which seem unfazed by the mechanical interloper.