Results


White Paper Machine Learning in Certified Systems

arXiv.org Artificial Intelligence

Machine Learning (ML) seems to be one of the most promising solution to automate partially or completely some of the complex tasks currently realized by humans, such as driving vehicles, recognizing voice, etc. It is also an opportunity to implement and embed new capabilities out of the reach of classical implementation techniques. However, ML techniques introduce new potential risks. Therefore, they have only been applied in systems where their benefits are considered worth the increase of risk. In practice, ML techniques raise multiple challenges that could prevent their use in systems submitted to certification constraints. But what are the actual challenges? Can they be overcome by selecting appropriate ML techniques, or by adopting new engineering or certification practices? These are some of the questions addressed by the ML Certification 3 Workgroup (WG) set-up by the Institut de Recherche Technologique Saint Exup\'ery de Toulouse (IRT), as part of the DEEL Project.


A new interpretable unsupervised anomaly detection method based on residual explanation

arXiv.org Artificial Intelligence

Despite the superior performance in modeling complex patterns to address challenging problems, the black-box nature of Deep Learning (DL) methods impose limitations to their application in real-world critical domains. The lack of a smooth manner for enabling human reasoning about the black-box decisions hinder any preventive action to unexpected events, in which may lead to catastrophic consequences. To tackle the unclearness from black-box models, interpretability became a fundamental requirement in DL-based systems, leveraging trust and knowledge by providing ways to understand the model's behavior. Although a current hot topic, further advances are still needed to overcome the existing limitations of the current interpretability methods in unsupervised DL-based models for Anomaly Detection (AD). Autoencoders (AE) are the core of unsupervised DL-based for AD applications, achieving best-in-class performance. However, due to their hybrid aspect to obtain the results (by requiring additional calculations out of network), only agnostic interpretable methods can be applied to AE-based AD. These agnostic methods are computationally expensive to process a large number of parameters. In this paper we present the RXP (Residual eXPlainer), a new interpretability method to deal with the limitations for AE-based AD in large-scale systems. It stands out for its implementation simplicity, low computational cost and deterministic behavior, in which explanations are obtained through the deviation analysis of reconstructed input features. In an experiment using data from a real heavy-haul railway line, the proposed method achieved superior performance compared to SHAP, demonstrating its potential to support decision making in large scale critical systems.


FAA Approves Fully Automated Commercial Drone Flights

WSJ.com: WSJD - Technology

U.S. aviation regulators have approved the first fully automated commercial drone flights, granting a small Massachusetts-based company permission to operate drones without hands-on piloting or direct observation by human controllers or observers. The decision by the Federal Aviation Administration limits operation of automated drones to rural areas and altitudes below 400 feet, but is a potentially significant step in expanding commercial applications of drones for farmers, utilities, mining companies and other customers. It also represents another step in the FAA's broader effort to authorize widespread flights by shifting away from case-by-case exemptions for specific vehicles performing specific tasks. In approval documents posted on a government website Thursday, the FAA said that once such automated drone operations are conducted on a wider scale, they could mean "efficiencies to many of the industries that fuel our economy such as agriculture, mining, transportation" and certain manufacturing segments. The FAA previously allowed drones to inspect railroad tracks, pipelines and some industrial sites beyond the sight of pilots or spotters on the ground as long as such individuals were located relatively close by.


AI-equipped guide panels make Tokyo area train station debuts

The Japan Times

Electronic panels equipped with artificial intelligence debuted Tuesday at major train stations in the Tokyo area to provide tourist and transfer information for a trial period, with the railway operator aiming to use them to make up for a future labor shortage. East Japan Railway Co. set up 30 panels at six stations in Tokyo and neighboring Chiba Prefecture for the demonstration, which lasts through late January. As a measure against the coronavirus, users do not have to touch the panels directly to operate them and some can automatically measure a passenger's temperature. Available in Japanese, English, Chinese and Korean, the displays can respond to voice questions and finger movements. They are installed at Shinjuku, Shinagawa, Ikebukuro and Takanawa Gateway stations in Tokyo as well as at two locations in Chiba, Kaihinmakuhari and the Airport Terminal 2 station at Narita Airport.


Towards a Framework for Certification of Reliable Autonomous Systems

arXiv.org Artificial Intelligence

The capability and spread of such systems have reached the point where they are beginning to touch much of everyday life. However, regulators grapple with how to deal with autonomous systems, for example how could we certify an Unmanned Aerial System for autonomous use in civilian airspace? We here analyse what is needed in order to provide verified reliable behaviour of an autonomous system, analyse what can be done as the state-of-the-art in automated verification, and propose a roadmap towards developing regulatory guidelines, including articulating challenges to researchers, to engineers, and to regulators. Case studies in seven distinct domains illustrate the article. Keywords: autonomous systems; certification; verification; Artificial Intelligence 1 Introduction Since the dawn of human history, humans have designed, implemented and adopted tools to make it easier to perform tasks, often improving efficiency, safety, or security.


The Future of Transportation

#artificialintelligence

Sengupta: Thank you so much for having me today. I'm really excited to be in San Francisco. I don't get to come here that often, which is strange because I live in Los Angeles, but I do like to come whenever I can. For my talk today, I'm going to talk about the future of transportation, specifically on the things that I worked on that I think are kind of the up and coming thing, the thing that I'm working on now and what's going to happen in the future. I think part of my career has always been about just doing fun and exciting new things and all my degrees are in aerospace engineering, ever since I was a little kid, I loved science fiction. I actually am a Star Trek person versus a Star Wars person, but I knew since I was a little kid that I wanted to be involved in the space program, so that's why I decided to go the aerospace engineering route and I wanted to build technology. I got my Ph.D. in plasma propulsion systems. Has anyone heard of the mission called Dawn that's out in the main asteroid belt? My Ph.D. research actually was developing the ion engine technology for that mission. It actually flew and got it to a pretty cool place out in the main asteroid belt looking at Vesta and Ceres. I did that for about five years and then I kind of felt like I had done everything I could possibly do on that front, from a research perspective. My management asked me if I wanted to work on the next mission to Mars. There's very few engineers in the space program who'd be like, "No, I'm just not interested in that." And they're like, "We want you to do the supersonic parachute for it."


Tackling Climate Change with Machine Learning

arXiv.org Artificial Intelligence

Climate change is one of the greatest challenges facing humanity, and we, as machine learning experts, may wonder how we can help. Here we describe how machine learning can be a powerful tool in reducing greenhouse gas emissions and helping society adapt to a changing climate. From smart grids to disaster management, we identify high impact problems where existing gaps can be filled by machine learning, in collaboration with other fields. Our recommendations encompass exciting research questions as well as promising business opportunities. We call on the machine learning community to join the global effort against climate change.


Commuters to pay for train fare with their face in China: Report

ZDNet

Those catching a train in Shenzhen may soon be able to pay for their fare through facial recognition, with a trial of the technology reportedly under way. It is one of the various technologies backed by the ultra-fast 5G network being tested by the local Shenzhen subway operator, according to the South China Morning Post. The initiative under way at Futian Station sees commuters scan their faces on a tablet-sized screen mounted on the entrance gate. The fare is then automatically deducted from a linked account. According to the report, there are currently 5 million rides per day on the city's network.


Revisiting the Importance of Individual Units in CNNs via Ablation

arXiv.org Artificial Intelligence

We revisit the importance of the individual units in Convolutional Neural Networks (CNNs) for visual recognition. By conducting unit ablation experiments on CNNs trained on large scale image datasets, we demonstrate that, though ablating any individual unit does not hurt overall classification accuracy, it does lead to significant damage on the accuracy of specific classes. This result shows that an individual unit is specialized to encode information relevant to a subset of classes. We compute the correlation between the accuracy drop under unit ablation and various attributes of an individual unit such as class selectivity and weight L1 norm. We confirm that unit attributes such as class selectivity are a poor predictor for impact on overall accuracy as found previously in recent work \cite{morcos2018importance}. However, our results show that class selectivity along with other attributes are good predictors of the importance of one unit to individual classes. We evaluate the impact of random rotation, batch normalization, and dropout to the importance of units to specific classes. Our results show that units with high selectivity play an important role in network classification power at the individual class level. Understanding and interpreting the behavior of these units is necessary and meaningful.


Airbus Aerial Provides a Whole New View of the World

WIRED

You may know Airbus as that Boeing competitor that also makes planes, but the European company is in fact an defense and aerospace giant that makes helicopters, satellites, and drones, and now it's using its aircraft not just to move people, but to give those on the ground a whole new view from the skies. A year-old effort called Airbus Aerial will seek to serve climate modelers, farmers, city planners, engineers, first responders, and anybody else who needs a a particular view of the world. The company combines data from observation satellites (of which Airbus is the largest global operator), manned planes with cameras slung underneath, and drones, to get to the places others can't reach. Airbus Aerial packages it all up, and presents it neatly to the customer, via a cloud-based interface. "It's a very complex thing to just say'I need satellite data'," says Jesse Kallman, president of the company.