Events: Data- och informationsteknik events at Chalmers University of TechnologyFri, 21 Jan 2022 14:13:18 +0100 Johansson, Computer Science and Engineering<p>Online seminar</p><p>​Machine Learning Based Methods for Virtual Validation of Autonomous Driving</p>During the last decade, automotive manufacturers have introduced increasingly capable driving automation functions in consumer vehicles. As the functionality becomes more advanced, the task of driving moves from the human to the car. Hence, making sure that autonomous driving (AD) functions are reliable and safe is of high importance. Often, increased levels of automation result in more complex safety validation procedures, that may be both expensive, time consuming, and dangerous to perform. One way to address these problems is to move parts of the validation to the virtual domain.<br /><br /> In this thesis, we investigate methods for validating AD functionality in virtual simulation environments, using methods from machine learning and statistics. The main focus is on how to make virtual simulations resemble real-world conditions as closely as possible. We tackle this with an approach based on sensor error modeling. Specifically, we develop a statistical sensor error model that can be used to make ideal object measurements from simulations resemble measurements obtained from the perception system of a real-world vehicle. The model, which is based on autoregressive recurrent mixture density networks, was trained on sensor error data collected on European roads.<br /><br /> The second part considers system falsification using reinforcement learning (RL); a flexible framework for validation of system safety, which naturally allows for the integration of, e.g., sensor error models. We compare results of system falsification using RL to an exact approach based on reachability analysis.<br /><br /> With this thesis, we take steps towards more realistic statistical sensor error models for virtual simulation environments. We also demonstrate that approximate methods based on reinforcement learning may serve as an alternative to reachability analysis for validation of high-dimensional systems. Finally, we connect the RL falsification application to sensor error modeling as a possible direction for future research. Carlsson, Computer Science and Engineering<p>Online</p><p>​Efficient Communication via Reinforcement Learning</p>Why do languages partition mental concepts into words the way the do? Recent works have taken a information-theoretic view on human language and suggested that it is shaped by the need for efficient communication. This means that human language is shaped by a simultaneous pressure for being informative, while also being simple in order to minimize the cognitive load. <br /><br />In this thesis we combine the information-theoretic perspective on language with recent advances in deep multi-agent reinforcement learning. We explore how efficient communication emerges between two artificial agents in a signaling game as a by-product of them maximizing a shared reward signal. This is tested in the domain of colors and numeral systems, two domains in which human languages tends to support efficient communication. We find that the communication developed by the artificial agents in these domains shares characteristics with human languages when it comes to efficiency and structure of semantic partitions. even though the agents lack the full perceptual and linguistic architecture of humans.<br /><br /> Our results offer a computational learning perspective that may complement the information-theoretic view on the structure of human languages. The results also suggests that reinforcement learning is a powerful and flexible framework that can be used to test and generate hypotheses in silico. scale battery production - environmental benefits and new challenges<p>OOTO Café, café, Sven Hultins Plats 2, Johanneberg Science Park 2</p><p>NOTE! Due to the pandemic, we are holding the seminar on a new date and time. We are back, after the Christmas holidays, with the new year&#39;s first Friday seminar. Welcome!Time and place: Café Ooto, Guldhuset, Johanneberg Science Park, 28 January, 12:30-14:00. In this seminar Anders Nordelöf, leader of the SEC theme Electromobility in society and researcher at Technology Management and Economics at Chalmers University of Technology, will give us different perspectives and an exciting discussion on a hot topic: Large scale battery production - environmental benefits and new challenges.</p><div><br /></div> <div><a href=""><img class="ms-asset-icon ms-rtePosition-4" src="/_layouts/images/icgen.gif" alt="" /> <span style="background-color:initial">R</span><span style="background-color:initial">egister to the seminar here</span></a></div> <div><br /></div> <div><strong> </strong><span style="background-color:initial"><strong>Program</strong></span></div> <div><br /></div> <div><ul><li>12:30, Welcome Tomas Kåberger, Chalmers Energy Area of Advance</li> <li>L<span style="background-color:initial">arge scale battery production - environmental benefits and new challenges, Anders Nordelöf.</span></li> <li>13<span style="background-color:initial">: 05-14:00, Mingle.</span></li></ul></div> <div><span style="background-color:initial"><br /></span></div> <div><span style="background-color:initial">Related:<br /></span><a href="/sv/personal/redigera/Sidor/anders-nordelof.aspx"><img class="ms-asset-icon ms-rtePosition-4" src="/_layouts/images/ichtm.gif" alt="" />Read more about Anders Nordelöf.</a><span style="background-color:initial"><br /><div><a href="/en/areas-of-advance/energy/news/Pages/Contributes-to-the-EUs-work-to-electrify-the-transport-sector.aspx"><img class="ms-asset-icon ms-rtePosition-4" src="/_layouts/images/ichtm.gif" alt="" /><span style="background-color:initial">C</span><span style="background-color:initial">ontributes to the EU’s work to electrify the transport sector<br /></span></a><a href="" style="outline:0px"><img class="ms-asset-icon ms-rtePosition-4" src="/_layouts/images/icgen.gif" alt="" />Swedish Electromobility Centre</a><br /><span></span><a href="/en/areas-of-advance/Transport/profile-areas/Pages/Sustainable-Vehicle-Technologies.aspx"><img class="ms-asset-icon ms-rtePosition-4" src="/_layouts/images/ichtm.gif" alt="" />Sustainable Vehicle Technologies​</a><br /></div> <br /><br /><br /></span></div> Ramon Currius, Computer Science and Engineering<p>Online</p><p>​Realistic Real-Time Rendering of Global Illumination and Hair througc</p><div><br /></div> Over the last decade, machine learning has gained a lot of traction in many areas, and with the advent of new GPU models that include acceleration hardware for neural network inference, real-time applications have also started to take advantage of these algorithms.<br /><br /> In general, machine learning and neural network methods are not designed to run at the speeds that are required for rendering in high-performance real-time environments, except for very specific and typically limited uses. For example, several methods have been developed recently for denoising of low quality pathtraced images, or to upsample images rendered at lower resolution, that can run in real-time. <br /><br /> This thesis collects two methods that attempt to improve realistic scene rendering in such high-performance environments by using machine learning. <br /><br /> Paper I presents a neural network application for compressing surface lightfields into a set of unconstrained spherical gaussians to render surfaces with global illumination in a real-time environment. <br /><br /> Paper II describes a filter based on a small convolutional neural network that can be used to denoise hair rendered with stochastic transparency in real time Bastys, Computer Science and Engineering<p>Online</p><p>​Principled Flow Tracking in IoT and Low-Level Applications</p><br /><div>Significant fractions of our lives are spent digitally, connected to and dependent on Internet-based applications, be it through the Web, mobile, or IoT. All such applications have access to and are entrusted with private user data, such as location, photos, browsing habits, private feed from social networks, or bank details.</div> <br /> In this thesis, we focus on IoT and Web(Assembly) apps. We demonstrate IoT apps to be vulnerable to attacks by malicious app makers who are able to bypass the sandboxing mechanisms enforced by the platform to stealthy exfiltrate user data. We further give examples of carefully crafted WebAssembly code abusing the semantics to leak user data.<br /><br /> We are interested in applying language-based technologies to ensure application security due to the formal guarantees they provide. Such technologies analyze the underlying program and track how the information flows in an application, with the goal of either statically proving its security, or preventing insecurities from happening at runtime. As such, for protecting against the attacks on IoT apps, we develop both static and dynamic methods, while for securing WebAssembly apps we describe a hybrid approach, combining both.<br /><br /> While language-based technologies provide strong security guarantees, they are still to see a widespread adoption outside the academic community where they emerged. In this direction, we outline six design principles to assist the developer in choosing the right security characterization and enforcement mechanism for their system. We further investigate the relative expressiveness of two static enforcement mechanisms which pursue fine- and coarse-grained approaches for tracking the flow of sensitive information in a system. Finally, we provide the developer with an automatic method for reducing the manual burden associated with some of the language-based enforcements. risk and long-term AI safety<p>Online Zoom</p><p>​Short lecture series on AI risk and long-term AI safety. February 21-25, 2022. ​</p><strong>​<br /></strong><span><strong>Time: February 21-25, 2022</strong></span><p><span lang="EN-US"><strong>Place: Online, Zoom - register to get the link</strong><br /></span><span style="background-color:initial"><strong>S</strong></span><span style="background-color:initial"><strong>peaker: Olle Häggström<br /><a href="" target="_blank"><img class="ms-asset-icon ms-rtePosition-4" src="/_layouts/images/icgen.gif" alt="" />Register here</a></strong></span></p> <p><span lang="EN-US"><br /></span></p> <p><span lang="EN-US">This six-hour lecture series (in English) will treat basics and recent developments in <strong>AI risk and long-term AI safety</strong>. The lectures are meant to be of interest to Ph.D. students and researchers in AI-related fields, but no particular prerequisites will be assumed. Some of the basics will be taken from my 2021 book <a href="" target="_blank">Tänkande maskiner​</a>, but the field is developing rapidly and most of what I say in the lectures will go beyond what I did in the book.</span></p> <p><span style="background-color:initial">An</span><span style="background-color:initial"> ambition will be to post recordings of the lectures at CHAIR’s YouTube channel quickly after each lecture, so that participants missing a lecture will have the chance to catch up before the next one.</span></p> <p><span style="background-color:initial">The three two-hour lectures are scheduled as follows:</span></p> <p><span style="background-color:initial">1. Monday, February 21 at 15.15-17.00: </span><strong style="background-color:initial">How and why things might go wrong</strong><br /></p> <p><span lang="EN-US">2. Wednesday, February 23 at 15.15-17.00: <strong>Timelines, natural language processors and oracle AI<br /></strong></span></p> <p><span lang="EN-US">3. Friday, February 25 at 10.00-11.45: <strong>Research directions in AI alignment<br /></strong></span></p> <div><span lang="EN-US" style="margin:0px;padding:0px;border:0px;font:inherit;vertical-align:baseline;color:inherit"><b><br /></b></span></div>