News: Centre CHAIR related to Chalmers University of TechnologyFri, 25 Sep 2020 08:17:17 +0200 call: Affiliated WASP PhD Student Positions in AI<p><b>​The purpose of the call is to provide the opportunity for PhD students not funded by WASP to be part of the WASP Graduate School.</b></p>​<span style="background-color:initial">The Wallenberg AI, Autonomous Systems and Software Program announces a call for up to 15 affiliated WASP AI PhD student positions within AI at the five partner universities Chalmers, KTH, Linköping University, Lund University and Umeå University as well as the research groups at Örebro University and Uppsala University that are members of WASP AI. </span><div><br /></div> <div><span style="background-color:initial">To t</span><span style="background-color:initial">he </span><span style="background-color:initial">call: </span><a href="">Affiliated WASP PhD Student Positions in AI​</a><br /></div> <div></div> <div><br /></div> <div><br /></div> <div><span style="background-color:initial">C</span><span style="background-color:initial">halmers' representative in WASP: <a href="/en/Staff/Pages/crnkovic.aspx">Ivica Crnkovic​​</a></span><br /></div> <div><span style="background-color:initial"><br /></span></div> <div><em>Wallenberg AI, Autonomous Systems and Software Program (WASP) is Sweden’s largest individual research program, a major national initiative for strategically motivated basic research, education, and faculty recruitment. The program addresses research on artificial intelligence and autonomous systems acting in collaboration with humans, adapting to their environment through sensors, information, and knowledge, and forming intelligent systems-of-systems. </em><br /></div> <a href="" target="_blank"><div></div></a>Fri, 25 Sep 2020 08:00:00 +0200 resource for AI research<p><b>A new resource for Artificial Intelligence and Machine Learning AI/ML research has recently opened at C3SE, the national centre for scientific and technical computing at Chalmers.</b></p><div><p class="p1" style="font-stretch:normal;font-size:12px;line-height:normal;font-family:helvetica"><span style="background-color:initial;font-family:&quot;open sans&quot;, sans-serif;font-size:14px">T</span><span style="background-color:initial;font-family:&quot;open sans&quot;, sans-serif;font-size:14px">he system is built around Graphical Processing Units (GPU:s) accelerator cards, and consists of two types of compute nodes with multiple Nvidia GPU:s each. The new cluster is named Alvis and is dedicated for AI/ML research.   </span><br /></p></div> <div><div><br /></div> <div>The Alvis system at C3SE is a part of the Swedish National Infrastructure for Computing, SNIC and financed by WASP (The Wallenberg Artificial Intelligence, Autonomous Systems and Software Program) together with SNIC.    </div> <div><br /></div> <div>At the end of August phase 1 of the project was going into production after a test period. This means the system is now open for project assignments. </div> <div><br /></div> <div>Thomas Svedberg at Chalmers e-Commons/C3SE and project leader for the SNIC AI/ML project, has already seen several applications regarding using Alvis. <br /><span style="background-color:initial">– This is the only dedicated AI/ML system within SNIC and I think we will see a lot of applications.   </span></div> <div><span style="background-color:initial"><br /></span></div> <div><span style="background-color:initial">Researchers from all Swedish universities as well as research institutes can use Alvis and the other SNIC resources. To use SNIC resources you need to be a member of a corresponding SNIC project. To apply for a project visit the SUPR portal (see below).  </span></div> ​<div><br /></div> <a href="" target="_blank" title="C3SE about Alvis">More about Alvis from C3SE &gt; </a><div>​<a href="" target="_blank" title="SUPR portal for projects using SNIC resources">SNIC project portal, SUPR &gt;​</a></div></div>Wed, 09 Sep 2020 14:00:00 +0200 researchers address the question – how does it work?<p><b>​Researchers around the world are focusing on the task of finding a theoretical framework that can explain how deep learning works in practice. Professor Giuseppe Durisi at Chalmers has accepted the challenge.</b></p>​<span style="background-color:initial">We have become used to computers that can be trained to accomplish intelligent tasks such as image and speech recognition and natural language processing. To explain how this training is performed, we can compare it to how a child learns. For example, a child needs to see a certain number of cats in order to build the general knowledge 'cat'.</span><div><br /></div> <div>Deep neural networks are trained in a similar manner. We feed them with example, which are used to adjust the parameters of the network, until the network delivers correct answers. When the network provides correct answers even when faced with new examples, that is, examples that were not used in the training phase, we know that it has acquired some general knowledge.</div> <div><br /></div> <div>Deep neural networks have achieved sensational results, but there is one fundamental problem that concerns researchers and experts. We see that they work, but we do not fully understand why. A common criticism is that deep learning algorithms are used as &quot;a black box&quot; – which is unacceptable for all applications that require guaranteed performance, such as traffic safety applications.</div> <div><br /></div> <div>”Right now, we lack the tools to describe why deep neural networks perform so well”, says Giuseppe Durisi, professor of Information Theory.</div> <div><br /></div> <div>Here is one of the mysteries about deep neural networks. According to established results in learning theory, we would expect deep neural networks to perform poorly when trained with the amount of data that is typically used.  But practice shows that this is perfectly fine.</div> <div><br /></div> <div>”It is even the case that if you make the network more complex – which according to established knowledge would impair its ability to generalize, the performance will sometimes improve.”</div> <div><br /></div> <div>There is no theoretically based explanation for why this occurs, but Giuseppe Durisi speculates with another analogy with human learning.</div> <div><br /></div> <div>”In order to reach a deeper understanding and thus the ability to generalize based on a large number of examples, we are required to overlook, or forget, a certain amount of details that are not important. Somehow, deep neural networks learn which part of the data is worth memorizing and which part can be ignored.” </div> <div><br /></div> <div>Many research groups around the world are now working hard to come up with a theory explaining how and why deep neural networks work. In connection with a major international conference in July this year, a competition was announced to see which research team can come up with theoretical bounds able to predict the performance of deep neural networks.</div> <div><br /></div> <div> Tools from many different research fields can be used to establish such a theory. Giuseppe Durisi hopes that information theory can be the right one.</div> <div><br /></div> <div>“Yes, information theory is my area of expertise, but it remains to be seen if we will succeed. That is how research works – and it is really exciting to apply the theory I am familiar with to address the completely novel challenge of understanding deep neural networks. It will keep us busy for a while.”</div> <div><br /></div> <div>Giuseppe Durisi has several research projects under way and collaborates with colleagues in other fields. Within the Chalmers AI Research Centre, he collaborates with Fredrik Hellström, Fredrik Kahl and Christopher Zach, and in a WASP project, Giuseppe Durisi and Rebecka Jörnsten from Mathematical Sciences have recently recruited a doctoral student, Selma Tabakovic, who will work on this problem.</div> <div><br /></div> <div>When Giuseppe Durisi reflects on the future, he sees that a greater understanding of deep learning can contribute with additional benefits – besides providing guaranteed performance in safety critical systems.</div> <div><br /></div> <div>”With a theoretical understanding of how deep learning works, we could build smaller, more compact, and energy-efficient networks that may be suitable for applications such as Internet-of-Things. It would contribute to increase the sustainability of such a technology.” </div> <div><br /> </div> <div><br /> </div> <div> </div> <div><div>Research projects</div> <div><strong>INNER: information theory of deep neural networks</strong></div> <div>Fredrik Hellström, Giuseppe Durisi and Fredrik Kahl</div> <div>Chalmers AI Research Centre (CHAIR)</div> <div><br /> </div> <div><strong>Generalization bounds of Deep Neural Networks: Insight and Design</strong></div> <div>Selma Tabakovic, Rebecka Jörnsten and Giuseppe Durisi</div> <div>Wallenberg AI, Autonomous Systems and Software Program (WASP)​</div></div> <div><br /> </div> <div><br /> </div> <div><br /> </div> <div>A deep neural network is a computer program that learns on its own. It is called &quot;neural network&quot; because its structure is inspired by the neural network that forms the human brain. Deep learning is a machine learning method, and part of what we call artificial intelligence. </div> <div><br /> </div> <div><strong>Illustration above:</strong> A deep neural network is fed with training data (in this case images) and the learning algorithms interpret the images through a number of layers – for each layer the degree of abstraction increases. Once the network has learned to identify combinations of patterns in the image – the system is able to distinguish a dog from a cat even on completely new images that were not included in the training material. </div> <div><br /> </div> <div><br /> </div> <div><br /> </div> <div></div>Tue, 01 Sep 2020 07:00:00 +0200 granted CHAIR Consortium Seed Projects 2020<p><b>​Eight projects have been granted seed funding by the CHAIR Consortium.</b></p>​Approximately 2,4 MSEK have been granted to the 8 approved proposals. The criteria for the selection were those specified in the call.<br /><br /><strong>Miroslaw Staron, Helena Odenstedt Herges, Silvana Naredi, Linda Block and Mikael Elam</strong><br />eHRV - Using EEG to label HRV data for detection of Delayed Cerebral Ischema in Stroke Patients<br /><br /><strong>Jun Li, Jiajia Chen and Lei Chen</strong><br />Robust Federated Learning against Low-quality and Corrupted Data<br /><br /><strong>Kun Gao, Yongzhi Zhang and Xiaobo Qu</strong><br />Online Lithium-ion Battery State of Health Prognostics<br /> <br /><strong>Huu Le</strong><br />Learning to Solve Robust Visual Odometry<br /> <br /><strong>Lucy Ellen Lwakatare and Aiswarya Raj</strong><br />Automatic data validation in data-streams for machine learning<br /> <br /><strong>Fredrik Johansson and Alexander Schliep</strong><br />AI and Missingness in Diagnostics for Alzheimer’s Disease<br /> <br /><strong>Sina Rezaei Aghdam, Marija Furdek Prekratic and Alexandre Graell I Amat</strong><br />Enhanced Security and Privacy for Wireless Federated Learning<br /> <br /><strong>Christian Berger</strong><br />DegradeFX - Explicating and Measuring Data Degradation Effects on ML<br /><br />Here was the call (the deadline has passed)<br /><a href="/en/centres/chair/news/Pages/Call-CHAIR-Consortium-SEED-projects-2020.aspx">Call for CHAIR Consortium Seed projects 2020 &gt;</a>Fri, 26 Jun 2020 10:00:00 +0200 A conversation about AI risk and AI ethics in the age of covid-19<p><b>​Speakers: Jaan Tallinn and Olle Häggström</b></p>​<span style="background-color:initial">Jaan Tallinn was originally trained in theoretical physics in Estonia and was one of the founders of Skype. Today he is an investor in technology start-ups around the world, as well as a philanthropist focusing on existential risk and AI safety research. </span><div><br /><span style="background-color:initial"></span><div>Olle Häggström is a professor of mathematical statistics at Chalmers, where he also serves as chairman of CHAIR's AI ethics committee. </div> <div> </div> <div>They discuss a range of topics in AI risk and AI ethics, and whether in these fields there are lessons to be learned from the ongoing covid-19 crisis.</div> <div> </div> <div>The webinar was held on 19th May, 2020, and organised by the AI Ethics Committee , within Chalmers AI Research Centre.</div> <div><br /></div> </div>Wed, 20 May 2020 14:00:00 +0200 for CHAIR Consortium Seed Projects 2020<p><b>​Call for seed project proposals (budget not exceeding 300 kSEK) that address AI research projects with given themes.</b></p>​<span style="background-color:initial">Important dates:</span><div>Submission due : May  15, 2020  (the date is firm)</div> <div>Notification: June 25, 2020</div> <div>Project start :  Latest Oct 1st 2020</div> <div>Project end: Latest March 1st 2021</div> <div><br /></div> <div>The CHAIR Consortium consists of the following partners: Chalmers, Volvo Group, CEVT, Volvo Cars, Ericsson and Sahlgrenska University Hospital. While the application domains of the partners are diverse, they all face a set of common challenges in developing AI/ML solutions, which are addressed in this call. This call for seed projects asks for novel research ideas and solutions that address these common challenges.</div> <div> </div> <div>The project should be relevant for the following themes:</div> <div><br /></div> <h2 class="chalmersElement-H2">Theme 1: Advances in handling aspects of real data in machine learning</h2> <div>Most AI/ML methods depend on access to large volumes of high quality data. However, there is often a number of aspects that interfere with the requirements for which many machine learning methods have been developed which raises challenges such as:</div> <div><br /></div> <div><ul><li>Privacy and security of data that cannot be leaked or shared</li> <li>Low quality of data or annotations, reliability of data from untrusted sources</li> <li>Streaming data and corresponding algorithms </li> <li>Continuously updated data</li></ul></div> <div>The above items may be addressed by techniques also known as federated learning, robust learning, stream processing and streaming analytics, and adaptive machine learning. Hence, this theme calls for research into developing solutions addressing these aspects of real data.</div> <div><br /></div> <h2 class="chalmersElement-H2">Theme 2: Integration of knowledge- and data driven techniques</h2> <div>In many applications, there is a need to combine pure data driven methods with specific models based on specialized domain knowledge tailored for the task at hand. This raises challenges such as:</div> <div><ul><li>Incorporation of knowledge about physical laws into data driven methods</li> <li>Combination of a data driven model with a physics based model</li> <li>Generation of representative training data by combining real data, physics-based simulation, and re-sampling techniques</li></ul></div> <div>The above items may be addressed by techniques also known as gray-box system identification, hybrid learning, simulation-based learning, generative modeling, and knowledge- and data driven approaches. Hence, this theme calls for research into developing such hybrid solutions.</div> <div><br /></div> <div>This is a call for proposals that address one of these two themes. In the project proposal there should be clear and explicit relations to these themes. Applications that build teams with good gender balance are encouraged.</div> <div><br /></div> <div>The seed projects are expected to be relevant for some or for all CHAIR Consortium partners.  The funding is assumed for researchers employed by Chalmers. Other partners, including CHAIR consortium partners, could be included in the projects, but their funding is not covered by this call, and their efforts should be expressed in-kind (FTE).  </div> <div><br /></div> <h2 class="chalmersElement-H2">Requirements:</h2> <div><ul><li>Eligible applicants are Chalmers researchers including GU employees at shared departments (MV and CSE). PhD students or master students from Chalmers, or GU from the education programs at shared departments can be included in the project.</li> <li>The budget should not exceed 300 kSEK including indirect costs (OH). It can cover personnel costs, and other costs up to 10% of the project budget, excluding equipment.  Furter, 10% funding in kind is requested from the department.</li> <li>The  proposal should have a clear and explicit relation to one or both themes. It should also be relevant to the CHAIR consortium partners.</li> <li>The proposal should follow the formats and outline defined in two templates. Incomplete proposals, or proposals that do not follow the defined format and size of the templates will be desk-rejected.</li> <li>A researcher can only be part of one proposal. Researchers already funded by CHAIR in 2019 or 2020 cannot apply.</li></ul></div> <div><br /></div> <h2 class="chalmersElement-H2">Submission and notification</h2> <div>The proposals should be submitted as one PDF file consisting of the following parts</div> <div><ol><li>Chair Project proposal form (<a href="/en/centres/chair/Documents/CHAIR%20Project%20proposal%20template%202020.docx">CHAIR Project proposal template</a>)  - one page </li> <li>Project Plan (<a href="/en/centres/chair/Documents/CHAIR%20SEED%20Project%20Plan%20Template%202020.docx">CHAIR Project plan template​</a>​) - max 3 pages</li> <li>Ethical considerations, in the spirit of <a href="/en/centres/chair/research/Pages/Ethical-policy.aspx">the CHAIR Ethical Policy</a> as a part of the project plan.</li> <li>CVs of main applicants - one page per applicant</li></ol></div> <div>Submit the proposal to Easychair <a href=""></a></div> <div>Upon notification, the PIs should prepare and sign the project contract with the CHAIR Consortium. The PI’s manager will also sign the contract. </div> <div><br /></div> <h2 class="chalmersElement-H2">Evaluation Criteria:</h2> <div><ul><li>The research novelty of the proposal.</li> <li>The relevance for core partners (including the number that found it relevant).</li> <li>The possible impact of the research to the research field.</li> <li>The expected research results, and the potential of the research results for further research or utilisation.</li> <li>The feasibility of the project plan.</li> <li>The ability of the project members to deliver the results.</li> <li>The ethical reasoning. </li></ul></div> <div><br /></div> <div>The proposals will be evaluated by the <a href="/en/centres/chair/About-us/Pages/organisation.aspx#ot" target="_blank">CHAIR Operational Team​</a> and decision of funding will be taken by the CHAIR Core Partners Board. </div> <div><br /></div> <h2 class="chalmersElement-H2">Contact Information</h2> <div>The call general questions can be sent to Ivica Crnkovic (<a href=""></a>) or Kolbjörn Tunström (<a href="">​</a>).</div> <div>For questions related to the ethical aspects of the proposal you can contact Olle Häggström (<a href=""></a>).</div> <div><br /></div> <div>To contact Core partners for particular questions related to the project proposal you can contact</div> <div>CEVT: Shafiq Urréhman <a href=""></a></div> <div>Ericsson: Aneta Vulgarakis <a href=""></a> </div> <div>Sahlgrenska University Hospital: Anders Hyltander <a href=""></a> </div> <div>Volvo Cars: Erik Hjerpe <a href=""></a></div> <div>Volvo Group: Jenny Erneman  <a href="">​</a></div> Tue, 14 Apr 2020 09:00:00 +0200 annual report 2019<p><b>Each year we publish an Annual Report, which outlines the research, activities and projects in the previous year.</b></p><a href="/SiteCollectionDocuments/Centrum/CHAIR/Chalmers_AI_Research_Centre_Annual_report_2019_200330.pdf">​Click here to read the Chalmers AI Research Centre annual report 2019</a><br />Mon, 30 Mar 2020 00:00:00 +0200 with core partners kick-started<p><b>​Chalmers AI research centre has signed agreements with five strategic partners, which now form a core of cooperation with industry and the public sector. The centre is well positioned to be a key part of the regional as well as national AI ecosystem.</b></p>​<span style="background-color:initial">The newly formed consortium held a kick-off on 10th March. Chalmers President Stefan Bengtsson welcomed all partners and mentioned the high expectations that are on upcoming activities. The CHAIR consortium's new partners agreed and gave their perspectives during the meeting.</span><div><br /></div> <div>Director Ivica Crnkovic presented an overall plan and pointed to major challenges in society where AI has a potential to contribute.</div> <div>“CHAIR's partners represent different areas, but the AI challenges they face are similar. There are great opportunities for collaboration, and we can achieve good synergies”, says Ivica Crnkovic.</div> <div><br /></div> <div>Kristian Abel from Volvo Cars described Data Science and AI as crucial for developing and delivering future mobility, both in products and in their own operations.</div> <div>“CHAIR contributes to strengthening AI research and competence in the region and to enhance collaboration within the area of AI. We see CHAIR as an important addition to the Swedish AI scene”, says Kristian Abel from Volvo Cars.</div> <div><br /></div> <div>Professor Ann-Marie Wennberg from the Sahlgrenska University Hospital described the challenges facing the health care sector and pointed out that technical development and artificial intelligence have the potential to offer important solutions. She sees the collaboration between the partners within CHAIR as a natural step and a necessity for Sahlgrenska University Hospital.</div> <div><br /></div> <div>Ericsson is researching and developing AI technologies to automate telecom networks, improve efficiency and deliver optimal user experiences.</div> <div>“The collaboration with Chalmers AI Research Centre will increase its expertise and excellence in artificial intelligence and offer shared knowledge and experience”, says Aneta Vulgarakis from Ericsson Research.</div> <div><br /></div> <div>Johan Lundén from Volvo Group has great confidence in the opportunities AI will open for the Volvo Group and the ability to shape future sustainable transport.</div> <div>“In line with that, we have high expectations for the cooperation that we look forward to within CHAIR, says Johan Lundén”, Volvo Group.</div> <div><br /></div> <div>Shafiq Urréhman from CEVT agrees.</div> <div>“We see regional ecosystems as important for research and development and AI-related technologies are strategic areas for CEVT. CHAIR offers an important part of the transformation towards AI-based applications, where collaboration and building competence are crucial.”</div> <div><br /></div> <div>The five parties have entered into core partner agreements during the period 2020-2024, with a possibility of extension: CEVT, Ericsson, Sahlgrenska University Hospital, Volvo Cars and Volvo Group.</div> <div><br /></div>Thu, 12 Mar 2020 10:00:00 +0100're-setting-up-AI-infrastructure.aspx're-setting-up-AI-infrastructure.aspxWe&#39;re setting up AI infrastructure<p><b>​​Chalmers AI Research Centre (CHAIR) is currently setting up infrastructure to be able to offer Chalmers researchers GPU time to use for AI related research.</b></p><div> At this stage we have a working proof of concept as a part of the Vera cluster. The server currently consists of four GPU’s, while the ambition is to increase the scale of the concept substantially. </div> <div><br /></div> <div>Read more about <span><span><a href="">the topology of the GPU server here</a>, or click here to see an <a href="/SiteCollectionImages/Centrum/CHAIR/research/CHAIR_infrastructure_vera.pdf">overview image of the topology (PDF)</a>.</span></span><br /></div>   <div>If you want to get access to the CHAIR GPU for your research project please contact Yinan Yu, <a href=""></a>, Post doc at the department of Computer Science and Engineering.</div> <div><br /></div> <div> </div> <ul><li><a href="">Register here to be able to use the infrastructure</a><br /><br /> </li> <li><a href="">Step by step instructions on how to use the infrastructure<br /><br /></a> </li> <li><a href="">How to write a job script and start a training job</a> </li></ul> <div> </div> Fri, 31 Jan 2020 11:00:00 +0100 will soon be able to prewarn of disease<p><b>​Many serious diseases would be detected earlier if the health care had the technical means for examining X-ray images. Chalmers University of Technology and Sahlgrenska University Hospital now work together to develop a method based on artificial intelligence to assess computed tomographic images (3D X-ray) of the heart’s coronary arteries. The tool is developed not least thanks to image data from a large Swedish population study.</b></p>​<span style="background-color:initial">Health care has so far only just had a first taste of all the opportunities offered by artificial intelligence, AI. Sahlgrenska and Chalmers AI Research Center (Chair) recently launched a <a href="/en/centres/chair/news/Pages/Chalmers-and-Sahlgrenska-University-Hospital-in-research-cooperation.aspx">strategic research collaboration on AI in health care​</a>.</span><div><br /></div> <div>“AI is developing rapidly at the moment”, says Fredrik Kahl, professor of computer vision and image analysis at the department of Electrical Engineering at Chalmers. “There are many unexplored opportunities for AI in medical technology, for example to make early diagnoses and to support health care staff during surgery.”</div> <div><br /></div> <div><strong>The technology is making progress</strong></div> <div>Cardiovascular disease is still the most common cause of death in Sweden and the world. But conditions have never been better to identify individual risks for, for example, stroke, COPD, sudden cardiac arrest, myocardial infarction and other heart diseases. This is due to several advances.</div> <div><br /></div> <div>In addition to AI technology itself becoming more and more advanced, new technology in the health care system makes it possible to take pictures of the heart, lungs and blood vessels in a way not previously possible. It is also possible to image and measure the distribution of fat in the body. In addition, there is now a sufficiently large image bank to use thanks to the population study Scapis. The study comprises 30,000 Swedes and is a collaboration between six universities and six university hospitals. Images and information collected by Scapis are now used in several medical research projects where computers will learn to interpret computed tomographic images of human organs.</div> <div><br /></div> <div>“We are currently working with Sahlgrenska to develop an algorithm that can be used for segmentation and classification of three-dimensional computed tomographic images of the coronary arteries”, says Fredrik Kahl.</div> <div><br /></div> <div>Jennifer Alvén is also involved in the project. She is a doctoral student in medical image analysis and in the process of developing an algorithm that allows the computer system to read the coronary arteries all by itself.</div> <div><br /></div> <div>“It is great that the research is really taking off now”, says Jennifer Alvén. “I am training the computer system through deep learning so that it can recognize the coronary arteries of the heart and the areas where the vessels hold calcium and fat, which could lead to future heart problems.”</div> <div><br /></div> <div><strong>Learns to recognise signs of future disease</strong></div> <div>When the computer learns to locate the coronary arteries, it needs actual cases to compare with. In this case 600 X-ray images from the Scapis project, where radiologists have outlined the coronary arteries digitally. Each such image takes about half a working day for medical staff to assess. The computer will now be trained to do the same job as the medical doctors.</div> <div><br /></div> <div>“The goal is to have the 600 images ready at the turn of the year. It will be the world’s largest data collection of coronary arteries images in a research context”, says Jennifer Alvén.</div> <div><br /></div> <div>The AI assessment will be as accurate as the assessment made by humans but will go much faster once the computer is trained. Thus, analysing all coronary arteries images for the 30,000 people in the survey will no longer be an impossible task. In the next step, AI can help in discovering undetected connections and patterns, when a follow-up is done to find out which of the people in the study have really been affected by, for example, myocardial infarction and stroke.</div> <div><br /></div> <div><img src="/SiteCollectionImages/Institutioner/E2/Nyheter/Snart%20kan%20AI%20varna%20för%20sjukdom%20innan%20den%20uppstår/vesselwithandwithoutplaque_512px.jpg" class="chalmersPosition-FloatRight" alt="" style="margin:5px" /></div> <div>The pictures show two examples of cross sections of coronary arteries that the AI system is learning to assess. The outer dotted line shows the outer contour of the artery wall, and the solid inner line shows the contour of the artery itself, where the blood is flowing. In the left image the artery wall is thin and without plaque. In the right picture, however, coating is visible on the inside of the artery wall.<br /><br /><strong style="background-color:initial">One step closer to practical use</strong><br /></div> <div>Scapis data is also used in another project to study connections between the presence of fat within the pericardial sac and cardiovascular disease. The Chalmers researchers have developed a working algorithm for this, which has been passed on to health care software development specialists.</div> <div><br /></div> <div>“We hope that the algorithm for coronary arteries also can be passed on for health care use”, says Jennifer Alvén. “It would be interesting to include it in one of the larger platforms already available for coronary arteries surgery.”</div> <div><br /></div> <div><strong>Great potential to improve public health</strong></div> <div>There are many needs and possible uses for image analysis in health care. A clear example of this is cancerous tumours of the kidneys, which are often detected at a much later stage than they could in fact have been spotted on X-rays.</div> <div><br /></div> <div>“Early detection of cancerous tumours in the kidneys would benefit greatly by an automatic algorithm”, says Fredrik Kahl. “When studying computed tomography images of people later diagnosed with kidney cancer, it has been found that in about fifty percent of the cases, medical doctors would have been able to detect the tumour on the X-rays. The problem is that no one has been looking specifically for such tumours in these images. Here is a gap that AI could fill.&quot;</div> <div><br /></div> <div>Both researchers have experienced a positive attitude and great interest from medical staff at Sahlgrenska for new AI tools. The lead times, however, are always long before new methods can be introduced in health care.</div> <div><br /></div> <div>A possible future scenario is that all X-ray images taken, for whatever reason, undergo an automatic AI examination to detect signs of the most serious diseases as early as possible. This would mean a huge opportunity to reduce patients’ suffering and improve public health.</div> <div><br /></div> <div><em>Text: Yvonne Jonsson</em></div> <div><em><br /></em></div> <div><br /></div> <div><div><strong>Facts about the population study Scapis</strong></div> <div><ul><li>Scapis is a Swedish population study that examines the cardiovascular status of 30,000 randomly selected women and men aged 50–64 years. The recruitment phase has been completed and analysis of collected data is now underway.</li> <li>The purpose is to be able to identify individual risks such as stroke, COPD, sudden cardiac arrest, myocardial infarction and other heart diseases.</li> <li>The goal is to gain greater knowledge about the origin of the diseases in order to prevent them before they occur.</li> <li>Six universities and six university hospitals in collaboration lead and run Scapis.</li> <li>Scapis is funded by the Swedish Heart-Lung Foundation as the main financier and with significant contributions from the Knut and Alice Wallenberg Foundation, Vinnova, the Swedish Research Council and the university hospitals and the universities themselves.</li></ul></div> <div><a href="" target="_blank"><img class="ms-asset-icon ms-rtePosition-4" src="/_layouts/images/icgen.gif" alt="" />Read more about Scapis​</a><br /></div> <div><br /></div> <div><br /></div> <div><strong>For more information contact</strong></div> <div><a href="/sv/personal/Sidor/fredrik-kahl.aspx">Fredrik Kahl​</a>, professor of computer vision and image analysis at the department of Electrical Engineering at Chalmers University of Technology, <a href=""></a></div> <div><a href="/en/Staff/Pages/alven.aspx">Jennifer Alvén</a>, PhD student at the division of Signal processing and Biomedical engineering at the department of Electrical Engineering at Chalmers, <a href=""></a></div> <div><img src="/SiteCollectionImages/Institutioner/E2/Nyheter/Snart%20kan%20AI%20varna%20för%20sjukdom%20innan%20den%20uppstår/arterytree.gif" class="chalmersPosition-FloatLeft" alt="" style="margin:5px" /><br /><br /><br /><br />An animated example of an artery tree, where medically relevant arteries are outlined.<br /></div> <div><img src="/SiteCollectionImages/Institutioner/E2/Nyheter/Snart%20kan%20AI%20varna%20för%20sjukdom%20innan%20den%20uppstår/CTAwitharteries.gif" class="chalmersPosition-FloatLeft" alt="" style="margin:5px" /><br /><br /><br /></div> <div><br /></div></div> <div><br /></div> <div><br /></div> <div><br /></div> <div><br /></div> <div><br /></div> <div><br /></div> <div><br /></div> <div><br /></div> <div><br /></div> <div><p class="MsoNormal"><span lang="EN-US">A video showing computed tomography images of a human heart. R</span><span class="tlid-translation"><span lang="EN">ed contours outline where there are coronary arteries in each layer.​</span></span><span lang="EN-US"></span></p></div> <div><br /></div> <div><br /></div> <div><br /></div> <div><br /></div> <div><br /></div> <div><br /></div> Mon, 04 Nov 2019 00:00:00 +0100”I want to transform the way science is done”<p><b>​“I want to transform the way science is done, artificial intelligence (AI) holds the potential to do that” says Ross D. King, new professor of Machine Intelligence at Chalmers University of Technology.</b></p>​Professor Ross D. King, newly recruited to the Department of Biology and Biological Engineering at Chalmers, has started work building a third generation Robot Scientist called ‘Genesis’. <br /><br />A Robot Scientist is a robotic system that applies techniques from artificial intelligence to execute cycles of automated scientific experimentation. The Genesis AI will coordinate the continuous execution of around 10,000 parallel cycles of hypothesis generation and testing to improve its model of how cells work.<br /><br /><strong>A tool for human scientists</strong><br /><br />“Such automation will make scientific research cheaper and faster, which is needed if we are to meet such global challenges as climate change, food security, disease, etc.,” says Ross King. <br /><br />He says that the goal is not to replace human scientists, but to give them a tool to achieve their goals.<br /><br />“We are very pleased that we managed to recruit Dr Ross King to Chalmers. Having worked in the field for over 30 years, he is one of the most experienced machine learning researchers in Europe. This recruitment will truly strengthen our competence within this field,” says Stefan Bengtsson, President at Chalmers University of Technology.   <br /><br /><strong>Drew the short straw – ended up in computer science</strong><br /><br />Ross King’s main research interests lie at interface between computer science and science. This interest started during his undergraduate studies in microbiology at the University of Aberdeen when he literally drew the short straw and had to do an unwanted mathematic project rejected by his fellow students. <br /><br />To his surprise he enjoyed the computational project and it led him to study for a Master of Science degree in Computer Science. Following this he completed a PhD at The Turing Institute at the University of Strathclyde on developing machine learning methods for protein structure prediction. This was one of the first ever PhD’s on machine learning and bioinformatics.<br /><br /><strong>A thousand times more efficient than a human scientist</strong> <br /><br />In 2004 Ross King started his work on the first Robot Scientist, ‘Adam’, at the University of Wales at Aberystwyth. Adam was the first machine to autonomously discover scientific knowledge: the function of some orphan enzymes in the yeast <em>Saccharomyces cerevisiae</em>. The second Robot Scientist, ‘Eve’, found some lead compounds for neglected tropical diseases. Eve is moving to Chalmers.<br /><br />“These are modest but not trivial discoveries. However, working with the first Robot Scientists has been a proof of principle, they are prototypes. The next step is to scale up and to make the Robot Scientist a thousand times more efficient than a human scientist performing the same experiments,” says Ross King. <br /><br /><strong>Focus on yeast systems biology </strong><br /><br />Working at Chalmers will enable Ross King to implement the technology of the Robot Scientist in the field of systems biology. <br /><br />“I want to focus on systems biology with yeast as a model organism and Chalmers is probably the best place in Europe for that. Since yeast can be used to understand how human cells work, there are medical and pathological reasons for working with this organism. But in the future the idea of using AI, in the form of Robot Scientists, can be applied in different fields of science,” says Ross King.<br /><br />Stefan Hohmann, Head of Department, Biology and Biological Technology, says that through the recruitment of Ross King the department’s expertise in computational biology will expand greatly, in particular in machine learning. <br /><br />“His research, especially the robot scientist project, will tie together different activities at the department but also across Chalmers. Ross will link the department to computer science and large initiatives in artificial intelligence such as WASP (funded by the Wallenberg Foundation) and CHAIR (funded by the Chalmers Foundation). The department will also profit from Ross' extensive international network,” says Stefan Hohmann.  <br /><br /><strong>About Ross D. King </strong><br /><ul><li>He is 57 years old and was born in Edinburgh. </li> <li>Moved in October 2019 to Gothenburg from Manchester where he was Professor of Machine Intelligence at the University of Manchester.</li> <li>He led the team that designed and tested the first nondeterministic universal Turing machine. Such computers have the potential to outperform electronic and quantum computers.</li> <li>Some of his major interests are music, literature and nature. </li> <li>He has developed <a href="">an algorithm for converting protein coding DNA sequences into pieces of music</a> together with Colin Angus of The Shamen</li></ul> <br />Text: Susanne Nilsson Lindh<br />Photo: Johan Bodell<br />Wed, 23 Oct 2019 14:00:00 +0200 is the artist behind AI generated art<p><b>​Artificial intelligence is increasingly used in creative tasks. One example is the Google Dream project with its surreal imagery. But who is the actual artist behind the pictures, the computer or the humans who programme and use the computer? A new research article discusses the possibilities and problems with creative machines.</b></p><p>”If an algorithm is trained on paintings by Von Gogh it will generate a lot of pictures that will look like Van Gogh paintings. That is not what creativity is” explains Palle Dahlsted, Interaction Design division, Department of Computer Science and Engineering. In the <a href="">article Big Data and Creativity</a> he summarizes the experience and research on creativity and computer-generated art. </p> <p>AI often uses so called probabilistic machine learning algorithms in creative work. Palle Dahlstedt say these types of algorithms lack the ability to create something truly new. He makes an example using the music of Johann Sebastian Bach.</p> <p>“If we study Bach’s fiftheen inventions they are quite different, but they all have something in common. Each part has similarities with the previous ones but is also a new addition to the idea-content. You can train an AI system on the works of Bach and get it to generate new parts that will sound like Bach, but they will lack something new. If Bach had written a sixteenth invention, it would, like all previous inventions, gone beyond the previous idea-content and added something new.”</p> <p>Since AI-system are largely controlled by humans, Palle Dahlstedt mean it’s wrong to talk about creative computers or that they have the ability to independently create works of art.</p> <p>“These are fantastic technologies; I love them and work with them all the time. They allow us to do something we previously couldn’t, to work with visual or musical material on a pattern level instead of for example note level. But the agency, the ability to make creative decisions, are partly in the hands of the programmer of the algorithms and partly in the hands of the person deciding the data set and setting the parameters for the algorithm.”</p> <p> <img src="/SiteCollectionImages/Institutioner/DoIT/News/AI-art-PD.jpg" alt="A self-portrait by Palle Dahlstedt created using Google Dream." class="chalmersPosition-FloatRight" style="margin:5px" />Today’s AI-systems are not creative but imitative. “A computer can imitate the symptoms or the result but not the causes of a work of art. That is the essence of it being imitative. It is not able to predict the next style change, which is the truly creative act.</p> <p>In the article Palle Dahlstedt write about the possibilities for truly creative artificial intelligence. “There is nothing that in principle prevents it, but we’re not there yet. For it to be interesting, the AI must have a life, an influx of information and an interaction with reality. If the AI has a life then maybe it could be creative, but then an equally interesting question arises. Should an AI create art for itself or for us humans? The result may be something beyond our range of understanding.</p> <p><em>Photo on the right: A self-portrait by Palle Dahlstedt created using Google Dream.</em></p> <h2>Read more</h2> <p><a href="">Article <em>Big Data and Creativity</em> as preprint at Research Gate</a></p> <p><a href="">Article <em>Big Data and Creativity</em> in European Review</a></p> <p><a href="">The chapter <em>Between Material and Ideas: A Process-Based Spatial Model of Artistic Creativity</em> from the book <em>Computers and Creativity</em></a></p> <p><a href="">Watch Palle Dahlstedt's lecutre n Etudes: Artistic Intervention at the Big Data Symposium</a></p> <p><strong><br /> </strong><strong>The Department of Computer Science and Engineering is shared between Chalmers University of Technology and University of Gothenburg.</strong><br /> </p>Mon, 21 Oct 2019 11:00:00 +0200 Hackathon to teach us more about the Baltic Sea<p><b>​Guillemots return to the same ledge on Stora Karlsö, outside Gotland every spring, to lay eggs and raise their chicks. This year, they were filmed during their spring months and the extensive material holds plenty of interesting insights. But how do you structure 2000 hours of filmed material? To identify and analyze behaviors, events, and abnormalities is a time-consuming task for researchers. Enter AI and citizen science.</b></p><p>​In November, a two-day hackathon will be held in Gothenburg, with the aim of accelerating the research of guillemots and about the sensitive ecosystems of the Baltic Sea. Data scientists, programmers, and UX designers will team up with WWF, the Swedish University of Agriculture (SLU), Stockholm Resilience Center, the Baltic Seabird Project, and AI Innovation of Sweden, to work out ways to structure data on the seabirds.</p> <p><a class="cta-arrow" href=""><img class="ms-asset-icon ms-rtePosition-4" src="/_layouts/images/ichtm.gif" alt="" />Baltic Seabird AI/UX Hackathon- information and registration</a></p> <h2 class="chalmersElement-H2">Citizen Science and slow TV</h2> <p>Last year, WWF and SLU invited the public to watch and comment on events in a live broadcast from one of the ledges on Stora Karlsö. Comments paired with time codes have enabled researchers to make great leaps forward in analyzing the material, they simply know what to look closely at, and when it occurred. The broadcast on WWF's website was a success, resulting in 1300+ comments and many new insights. The interest in three months of live broadcasting from a barren rock shelf can be compared to Swedish Television (SVT) somewhat unexpected viewing success in “slow TV” this spring, Den stora älgvandringen – three months of the quiet, live broadcast of forests waiting for the yearly moose migration.<br /><br />&quot;The public's great interest and efforts in previous projects have spurred us to continue to drive the development of the citizen research initiative and to allocate additional resources for, for example, hackathon and platform development,&quot; says marine expert Metta Wiese, WWF.</p> <h2 class="chalmersElement-H2">The guillemots as ‘litmus test’ of a sensitive ecosystem</h2> <p>Ecosystems are important for marine balance, but many of them are unfortunately out of balance. By studying seabirds around the clock, scientists have gained new insights into their life patterns, an important key in working to recreate resilient seas. </p> <p>&quot;We know that the herring these birds catch around the colony to feed the chicks are vital to them. Finding out more about how long they need to be away to find a fish and if some individuals are better than others at finding them, is interesting from a purely biological point of view, but also important information when it comes to studying how the seabirds on an individual level respond to ecosystem changes. One typical change is the amount of fish in the Baltic Sea,&quot; says marine researcher Jonas Hentai-Sundberg at SLU.</p> <h2 class="chalmersElement-H2">What artificial intelligence can do</h2> <p>For Metta Wiese and Jonas Hentai-Sundberg, the hackathon will be a unique opportunity to collaborate cross-sectorally to save the seas. Using AI, researchers hope that the mapping of individuals and common events can be automated, emancipating resources of researchers and citizen science to observations and interpretations of the unusual and rare. </p> <p>&quot;Being able to identify individuals and events such as mating, brawl, feeding, arrivals and departures, gives the researchers the necessary tools to an in-depth understanding of the lives of the guillemots. Watching thousands of hours of film is impossible – but with automation, we hope to get a broader picture of everything happening on the ledge. If we then have knowledge of how the birds are moving, perhaps machine learning can be used to identify what they are doing and maybe who is doing it?&quot; says Jonas Hentai-Sundberg at SLU.</p> <p>AI Innovation of Sweden, WWF, and SLU are partnering with many more organizations for the hackathon. </p> <p><br /></p> <div class="call-out-box"><strong>Baltic Seabird AI/UX hackathon,</strong> November 21-22 <p><strong>Organization</strong><br /> WWF, SLU, Stockholm Resilience Centre, Baltic Seabird project, Chalmers AI Research Centre, AI Innovation of Sweden</p> <p><strong>With support from</strong><br /> Ocean Data Factory Sweden, SMHI, Space Data Lab, CGIT, Annotell, Zenuity</p></div> <p><br /></p> Mon, 21 Oct 2019 00:00:00 +0200 contributes to Big Data research in Bulgaria<p><b>On 7th October, a new European centre of excellence in Big Data and Artificial Intelligence was launched. The GATE Institute is located in Sofia and is a partnership between Sofia University, Chalmers University of Technology and Chalmers Industrial Technologies.</b></p><p>​<br />The initiative is made through funding for seven years from the EU's Horizon 2020 Widespread programme, which aims to promote competence and innovation across Europe and thereby strengthen European competitiveness and ability to meet societal challenges. The goal is to build research capacity and promote innovation power.<br /><br />Graham Kemp at the Department of Computer Science and Engineering at Chalmers has been working with preparations and the application since 2017 and is very pleased that the Institute has now been launched.<br /><br /><img src="/SiteCollectionImages/Areas%20of%20Advance/Information%20and%20Communication%20Technology/News%20events/GATE/Gate_GrahamKemp.jpg" alt="Graham Kemp" class="chalmersPosition-FloatRight" style="margin:5px;vertical-align:middle;width:200px;height:288px" />“We see an opportunity to share knowledge and experience in a field that is developing extremely rapidly. Our participation in GATE will lead to greater international impact, perspectives and interaction. In the longer term the GATE Institute will become a strong partner for collaboration in Eastern Europe”, says Graham Kemp.<br /><br />The research is focused on four strategic application themes: future cities, intelligent government, smart industry and digital health. GATE will employ over 100 researchers and install three new research labs at Sofia University, City Living Lab, Digital Twin Lab, in multidisciplinary collaboration with industry, as well as Virtual Reality and Big Data Visualisation (Open Visualisation Lab).<br /><br />The GATE Institute, as the only Big Data centre of excellence in Eastern Europe, will form a hub in a European network of more than 50 Big Data centres. GATE thus plays a strategically important role in expanding the network and contributing to knowledge transfer and innovation that will provide exchange not only at national or regional level, but for the whole of Europe. GATE will gather and unite everyone with interest in the research field – academia, government, industry and society.<br /><br />To the webpage of Gate: <a href="" target="_blank"> <br /></a></p> <p><em>This project has received funding from the European Union’s Horizon 2020 WIDESPEAD-2018-2020 TEAMING Phase 2 programme under Grant Agreement No. 857155.</em></p> <p><br />Captions</p> <p><strong>Group photo:</strong> Participants at the GATE kick-off meeting on 7th October 2019, in Sofia, Bulgaria. Back row, from left: Oana Radu (Research Executive Agency, European Commission), Ales Fiala (Research Executive Agency, European Commission), Boyan Stefanov (Sofia University), Eleonora Getsova (Sofia University), Nils Munk Wirell (Chalmers Industrial Technologies), Camilla Johansson (Chalmers Industrial Technologies), Yannis Patias (Sofia University), Iva Krasteva (Sofia University), Petya Stancheva (DG RTD, European Commission), Vassil Vassilev (London Metropolitan University)<br />Front row, from left: Irena Pavlova (Sofia University), Dag Wedelin (Chalmers University of Technology), Graham Kemp (Chalmers University of Technology), Ivica Crnkovic (Chalmers University of Technology), Sylvia Ilieva (Project Coordinator, Sofia University), Golaleh Ebrahimpur (Chalmers Industrial Technologies), Magda De Carli (DG RTD, European Commission), Dessislava Petrova-Antonova (Sofia University).<br /><br /><strong>The press conference:</strong> Eleonora Getsova moderates a press conference with (seated, left-to-right) Magda De Carli (DG RTD, European Commission), Prof Sylvia Ilieva (GATE project coordinator, Sofia University), Prof Anastas Gerdjikov (Rector, Sofia University), Dr Golaleh Ebrahimpur (CEO, Chalmers Industrial Technologies) and Prof Ivica Crnkovic (Chalmers University of Technology).<br /><br /><strong>Audience:</strong> The GATE opening event took place in the Ceremonial Hall of Sofia University St. Kliment Ohridski.<br /><br /><strong>Photos: </strong>Oleg Konstantinov<br /></p> <p><br /><a href="" target="_blank"></a></p>Mon, 14 Oct 2019 00:00:00 +0200 projects funded within deep neural networks and machine learning<p><b>​Chalmers AI Research Centre (Chair) will fund five PhD student projects at Chalmers. The projects focuses on different aspects of deep neural networks and machine learning.</b></p><p>​In the beginning of May, we <a href="/en/centres/chair/opportunities/Pages/phdprojects2019.aspx">opened a call for PhD student projects within AI</a>. </p> <div> </div> <p>Within the call we looked for projects that aimed to:</p> <div> </div> <ul><li>develop theoretical foundations and computational methods for AI, through research activities focused on algorithmic, mathematical, and statistical principles;</li> <li>use innovative AI tools to tackle foundational problems in other fields such as biology, physics, material science, etc.</li> <li>tackle core problems related to the development and the deployment of systems containing AI components.</li></ul> <div> </div> <p>We received 63 project proposals from researchers across almost all departments at Chalmers.Five projects have now been selected after an external review process. These projects will receive full support for a PhD student from Chair.</p> <div> </div> <p>“The large number of high-quality submissions to this call shows that our researchers have great interest in AI and have the capacity to perform excellent research within AI. We look forward to seeing the research output of the five selected projects and we regret we could not support more projects due to budget restrictions” says Giuseppe Durisi, Co-director of Chalmers AI Research Centre.</p> <div> </div> <h2 class="chalmersElement-H2">Selected projects within the call “PhD student projects within AI”</h2> <h3 class="chalmersElement-H3"> Deep Learning and likelihood-free Bayesian inference for intractable stochastic models</h3> <div> </div> <p><strong>Applicant:</strong> Umberto Picchini, Department of Mathematical Sciences</p> <div> </div> <p>We construct new deep neuronal networks (DNNs) to learn the parameters of complex stochastic dynamical models that do not have tractable likelihood functions. Specifically, we leverage the expressive approximation power of our DNNs to extract essential information from time-series data, both Markovian and not-Markovian, and then learn model parameters using likelihood-free methodology, such as approximate Bayesian computation. Special (though not exclusive) focus is directed to stochastic differential equation models and state space models (SSMs), where SSMs represent noisy observations of a latent Markovian process. The result will be a flexible <em>plug-and-play</em> machine learning methodology, allowing inference for complex stochastic models.</p> <div> </div> <p><br /></p> <div> </div> <h3 class="chalmersElement-H3">Energy-based models for supervised deep neural networks and their applications</h3> <div> </div> <p><strong>Applicants: </strong>Christopher Zach, Department of Electrical Engineering, and Morteza Haghir Chehreghani, Department of Computer Science and Engineering</p> <div> </div> <p>Despite deep learning-based methods being the state-of-the-art in many AI-related applications, there is a lack of consensus of how to understand and interpret deep neural networks in order to reason about their strengths and weaknesses. Energy-based models in machine learning have a long tradition as a framework to learn from unlabeled data, i.e. unsupervised learning. Recently, it has been shown that supervised learning of deep neural networks using the back propagation method is a limiting case of a suitably defined approach for learning energy-based models using a so-called contrastive loss. This connection is the basis for our interest in a tighter connection between deep learning and energy-based models.</p> <div> </div> <p><br /></p> <div> </div> <h3 class="chalmersElement-H3">Mechanisms for secure and private machine learning</h3> <div> </div> <p><strong>Applicants: </strong>Aikaterini Mitrokotsa, Department of Computer Science and Engineering, and Christos Dimitrakakis, Department of Computer Science and Engineering</p> <div> </div> <p>We envision secure and privacy-preserving machine learning algorithms for artificial intelligence applications in everyday life, that can provide confidentiality and integrity guarantees. In particular, we aim to: </p> <div> </div> <ol><li>Safeguard the privacy of individuals that participate by either (a) providing their data to build the system, or (b) being end-users of the system. </li> <li> Safeguard the integrity of the system by (a) ensuring its robustness to adversarial inputs (b) cryptographically limiting the possible points of adversarial manipulation.</li></ol> <div> </div> <div><br /></div> <div> </div> <p></p> <div> </div> <p></p> <h3 class="chalmersElement-H3">Stochastic continuous-depth neural networks</h3> <div><strong>Applicant: </strong>Moritz Schauer, Department of Mathematical Sciences</div> <div>We will advance the understanding of deep neural networks through the investigation of stochastic continuous-depth neural networks. These can be thought of as deep neural networks (DNN) composed of infinitely many stochastic layers, where each single layer only brings about a gradual change to the output of the preceding layers. We will analyse such stochastic continuous-depth neural networks using tools from stochastic calculus and Bayesian statistics. From that, we will derive practically relevant and novel training algorithms for stochastic DNNs with the aim to capture the uncertainty associated with the predictions of the network.</div> <p></p> <div> </div> <p><br /></p> <div> </div> <h3 class="chalmersElement-H3">VisLocLearn - Understanding and Overcoming the Limitations of Convolutional Neural Networks for Visual Localization</h3> <div> </div> <p><strong>Applicant: </strong>Torsten Sattler, Department of Electrical Engineering</p> <div> </div> <p>Visual localization is the problem of estimating the position and orientation from which an image was taken with respect to the scene. In other words, visual localization allows an AI system to determine its position in the world through a camera. Understanding why current approaches fail and proposing novel approaches that are able to accurately localize a camera are problems of high practical relevance. This is the purpose for the proposed project, VisLocLearn.</p>Tue, 08 Oct 2019 13:00:00 +0200