A fall is an abnormal activity that occurs rarely, so it is hard to collect real data for falls. It is, therefore, difficult to use supervised learning methods to automatically detect falls. Another challenge in using machine learning methods to automatically detect falls is the choice of engineered features. In this paper, we propose to use an ensemble of autoencoders to extract features from different channels of wearable sensor data trained only on normal activities. We show that the traditional approach of choosing a threshold as the maximum of the reconstruction error on the training normal data is not the right way to identify unseen falls. We propose two methods for automatic tightening of reconstruction error from only the normal activities for better identification of unseen falls. We present our results on two activity recognition datasets and show the efficacy of our proposed method against traditional autoencoder models and two standard one-class classification methods. Keywords: detection 1. Introduction fall detection, one-class classification, autoencoders, anomaly Falls are a major cause of both fatal and nonfatal injury and a hindrance in living independently. Each year an estimated 424, 000 individuals die from falls globally and 37.3 million falls require medical attention [23]. Experiencing a fall may lead to a fear of falling [6], which in turn can result in lack of mobility, less productivity and reduced quality of life. There exist several commercial wearable devices to detect falls [24]; most of them use accelerometers to capture motion information. They normally come with an alarm button to manually contact a caregiver if the fall is not detected by the device.

But deciding to build an app in the first place was driven by a broader purpose. As one doctor put it, "If you want to do world-class sociology research, build Facebook."

It has long been suspected that using satnavs can make drivers abandon their senses and do silly things. The persuasive voice of the satnav has led to countless cases of big lorries stuck in narrow country lanes, or motorists who have ignored their own eyes and driven into rivers or the sea. Now research has shown that when we use the handy devices, it'switches off' the parts of our brain we normally use to navigate. But don't panic if you're a Google Maps junkie - although your brain might be in sleep mode now, using real maps will wake it up again. UCL researchers studied the brains of 24 volunteers.

European adtech company ADYOULIKE, which offers a AI-driven native advertising platform, today launched a US operations in New York, along with debuting a new server-to-server (S2S) header bidding solution. The company said the new solution assures native ads are placed with the maximum contextual relevance available and is the first to incorporate semantic targeting powered by IBM Watson. Among the first companies to offer S2S for native ads, and the first to leverage AI to enhance their solution, Adyoulike said the development of the new S2S solution is a continuation of the company's ongoing commitment to using emerging technologies to drive advances in programmatic advertising. Publishers using the solution to manage their inventory will have the ability to create private marketplaces and package deals for buyers that include more precise targeting based on context and sentiment. This added relevance, says Adyoulike, increases value both for the publisher, and for the brand.

The Future of Humanity Institute researcher Anders Sandberg has said that we talk about'artificial intelligence' only until it works; thereafter we call it'automation'. How smart, for example, is a computer that can win at chess, Jeopardy, or even Go when it can't extrapolate from its knowledge of those games to tackle something else? Our inner biological supremacists can smugly dismiss those computers as automation. At the beginning of Luke Dormehl's Thinking Machines: The Quest for Artificial Intelligence and Where It's Taking Us Next, 'computers' are people whose actuarial jobs require them to perform complex calculations. By the end, the scientists he interviews are discussing a future in which computers may be a lot like people.

A park in Beijing has installed toilet paper dispensers with facial recognition to stop visitors from taking too much loo roll, media reports say. Machines at the Temple of Heaven park scan visitors' faces before dispensing a fixed length strip of paper. The tourist attraction is reportedly frequented by visitors who take large amounts of loo roll home. It has reignited debate over the lack of social graces among some Chinese. Park officials have installed six machines at its public bathrooms in a half-month trial, with staff on standby to explain the technology to visitors.

Samsung has finally announced its artificial intelligence-based voice assistant, Bixby on Monday. The assistant is expected to be one of the features of the Samsung Galaxy S8, when it is announced on March 29. "Bixby will be our first step on a journey to completely open up new ways of interacting with your phone. At the launch of the Galaxy S8, a subset of preinstalled applications will be Bixby-enabled. This set will continue to expand over time. Our plan is to eventually release a tool (in SDK) to enable third-party developers to make their applications and services Bixby-enabled easily," InJong Rhee, executive vice-president, head of research and development, software and services said in the official press release.

At Adobe Summit, Adobe (Nasdaq:ADBE) introduced significant enhancements to the Adobe Cloud Platform, the underlying, cross-cloud architecture that unifies content and data and leverages Adobe Sensei, Adobe's AI and machine learning framework. Advancements announced today include new Sensei capabilities for enterprise customers as well as new tools and partner integrations to help developers reduce time to market and better integrate Creative Cloud assets into enterprise workflows. Adobe also announced Adobe Experience Cloud (see separate press release) as well as the availability of the first set of solutions that integrate with Microsoft's Azure, Dynamics 365 and PowerBI offerings (see separate press release). Advancements in cloud services have fundamentally altered the computing landscape. The next decade will bring even more disruptive changes in how brands create, immerse and engage their customers in their experiences.

The 2017 New Rules for the Digital Age report from Deloitte found that only 5 percent of the companies surveyed said they have strong digital leadership development programs and a clear majority (65 percent) said they have no significant program to drive digital leadership skills. Josh Bersin, a principal at the Bersin by Deloitte research group, says the challenge is that companies don't realize how much more complicated digital transformation is than simply acquiring new technology. "Digital technology is easy to buy, but once you turn it on it changes the way you work and how you deliver products and services," Bersin told CIO.com. "From the CIO's perspective, it may seem relatively easy to implement artificial intelligence (AI), social media and other new technology, but these things have a disruptive impact on the workplace." For example, the study found that companies feel 31 percent "less ready" to redesign their organization around digital business models than they did last year.

A few months ago I wrote a tutorial on how to classify images using Convolutional Neural Networks (specifically, VGG16) pre-trained on the ImageNet dataset with Python and the Keras deep learning library. The pre-trained networks inside of Keras are capable of recognizing 1,000 different object categories, similar to objects we encounter in our day-to-day lives with high accuracy. Back then, the pre-trained ImageNet models were separate from the core Keras library, requiring us to clone a free-standing GitHub repo and then manually copy the code into our projects. This solution worked well enough; however, since my original blog post was published, the pre-trained networks (VGG16, VGG19, ResNet50, Inception V3, and Xception) have been fully integrated into the Keras core (no need to clone down a separate repo anymore) -- these implementations can be found inside the applications sub-module. Because of this, I've decided to create a new, updated tutorial that demonstrates how to utilize these state-of-the-art networks in your own classification projects.