Events: Data- och informationsteknik events at Chalmers University of TechnologyFri, 25 Sep 2020 15:22:00 +0200 Machine Learning Seminar<p></p><p>The MinMax k-means clustering algorithm</p><div> Arman Rahbar, a Ph.D. student in Data Science and AI Division at CSE, Chalmers, will talk about a paper by Grigorios Tzortzis and Aristidis Likas published in Pattern Recognition called <span>The MinMax k-means clustering algorithm<span style="display:inline-block">.</span></span><br /></div> <div><br /></div> <div></div> <div>Abstract:<br /></div> <div>The MinMax k-Means clustering algorithm Applying k-Means to minimize the sum of the intra-cluster variances is the most popular clustering approach. However, after a bad initialization, poor local optima can be easily obtained. To tackle the initialization problem of k-Means, we propose the MinMax k-Means algorithm, a method that assigns weights to the clusters relative to their variance and optimizes a weighted version of the k-Means objective. Weights are learned together with the cluster assignments, through an iterative procedure. The proposed weighting scheme limits the emergence of large variance clusters and allows high-quality solutions to be systematically uncovered, irrespective of the initialization. Experiments verify the effectiveness of our approach and its robustness over bad initializations, as it compares favorably to both k-Means and other methods from the literature that consider the k-Means initialization problem.</div> <div><br /></div> opportunities with AI Sweden<p>Online, Zoom</p><p>​​​Welcome to a project ideas meeting! Here we want to identify opportunities for new projects and help AI Sweden to drive the AI agenda.</p><div><br /></div> AI Sweden serves as a platform for AI research and innovation activities in Sweden (read more about AI Sweden at <a href="" target="_blank"></a>). <div><br /></div> <div> Welcome to learn more about collaboration opportunities in AI projects and how to finance them! The goal is to find at least three concrete project ideas to start finding partners and funding for. You do not need to have a project idea ready for the workshop - we want you to build on your experience, have an open mind and be creative. </div> <div><br /></div> <div> We invite researchers at Chalmers and GU to an online workshop.</div> <h3 class="chalmersElement-H3"><a href="" target="_blank">Register and see the agenda at &gt;​</a></h3> AI Talks: Barbara Plank<p>Online, Zoom</p><p></p><h2 class="chalmersElement-H2">Learning across Adverse Conditions in Natural Language Processing</h2> <h3 class="chalmersElement-H3">October 1, 2020, 1:00-2:00 pm, Swedish time, Zoom<br />Zoom link: <a href="" target="_blank" title="" style="font-family:inherit"></a><span lang="EN-US" style="font-family:inherit;background-color:initial"> <span style="color:rgb(255, 0, 0)">(Password to be posted)</span></span></h3> <div><b>​<br /></b></div> <div> Welcome to an AI Talks under the theme Learning across Adverse Conditions in Natural Language Processing, with Barbara Plank, Associate Professor of Natural Language Processing (NLP) at the Computer Science Department at the IT University of Copenhagen. </div> <div><br /></div> <div> <strong>Abstract: </strong>Transferring knowledge to solve a related problem and learning from limited, unreliable inputs are examples of extraordinary human ability. State-of-the-art machine learning models based on deep learning often fail under such adverse conditions. So, how can we build Natural Language Processing technology which transfers better to new conditions, such as learning to process a new language, learning to answer new types of questions, or learning with uncertainty stemming from human labelling? In this talk, I will present some recent work to address these ubiquitous challenges using neural networks in NLP. In particular, I will include work on cross-lingual learning for NLP and work at the interface of language and vision. </div> <div><br /></div> <h3 class="chalmersElement-H3"><img src="/SiteCollectionImages/Centrum/CHAIR/events/AI_Talks_barbara_plank_180px.jpg" class="chalmersPosition-FloatLeft" alt="" style="margin:5px 10px" />About the speaker</h3> <div>Barbara Plank is Associate Professor of Natural Language Processing (NLP) in the Computer Science Department at ITU (IT University of Copenhagen) where she leads a research lab in natural language processing.​​​</div> <div><br /></div> Lopez Juan, Computer Science and Engineering<p>Online</p><p>​Practical Unification for Dependent Type Checking</p><div><h3 class="chalmersElement-H3">Abstract</h3></div> <div>When using popular dependently-typed languages such as Agda, Idris or Coq to write a proof or a program, some function arguments can be omitted, both to decrease code size and to improve readability.  Type checking such a program involves inferring a combination of these implicit arguments that makes the program type-correct.<br /><br />Finding such a combination of implicit arguments entails solving a higher-order unification problem.<br />Because higher-order unification is undecidable, our aim is to infer the omitted arguments for as many programs as possible with a reasonable use of computational resources. The extent to which<br />these goals are achieved affect how usable a dependently-typed proof assistant or programming language is in practice.<br /><br />Current approaches to higher-order unification are in some cases too inflexible, postponing unification of terms until their types have been unified (Coq, Idris). In other cases they are too optimistic, which sometimes leads to ill-typed terms that break internal invariants (Agda).<br /><br />In order to increase the flexibility of our unifier without sacrificing soundness, we use the twin types technique by Gundry and McBride. We simplify their approach so that it can be used within an existing type<br />theory without changes to the syntax of terms. We also extend it so that it can handle more classes of constraints. We show that the resulting solutions are correct and unique.<br /><br />Finally, we implement the resulting unification algorithm on an existing type checker prototype for a smaller variant of the Agda language, developed by Mazzoli and Abel. We make a suitable choice of internal term representation, and use few, if any, example-specific optimizations. We obtain a type-checker which avoids ill-typed solutions, and is also able to handle some challenging examples in time and memory comparable to the existing Agda implementation. <br /></div> Stylianopoulos, Computer Science and Engineering<p>Zoom, link above</p><p>Hardware-Aware Algorithm Designs for Efficient Parallel and Distributed Processing</p> <p><br /></p> <p>The introduction and widespread adoption of the Internet of Things, together with emerging new industrial applications, bring new requirements in data processing. Specifically, the need for timely processing of data that arrives at high rates creates a challenge for the traditional cloud computing paradigm, where data collected at various sources is sent to the cloud for processing. As an approach to this challenge, processing algorithms and infrastructure are distributed from the cloud to multiple tiers of computing, closer to the sources of data. This creates a wide range of devices for algorithms to be deployed on and software designs to adapt to.</p> <p><br /></p> <p>In this thesis, we investigate how hardware-aware algorithm designs on a variety of platforms lead to algorithm implementations that efficiently utilize the underlying resources. We design, implement and evaluate new techniques for representative applications that involve the whole spectrum of devices, from resource-constrained sensors in the field, to highly parallel servers. At each tier of processing capability, we identify key architectural features that are relevant for applications and propose designs that make use of these features to achieve high-rate, timely and energy-efficient processing.</p> <p><br /></p> <p>In the first part of the thesis, we focus on high-end servers and utilize two main approaches to achieve high throughput processing: vectorization and thread parallelism. We employ vectorization for the case of pattern matching algorithms used in security applications. We show that re-thinking the design of algorithms to better utilize the resources available in the platforms they are deployed on, such as vector processing units, can bring significant speedups in processing throughout. We then show how thread-aware data distribution and proper inter-thread synchronization allow scalability, especially for the problem of high-rate network traffic monitoring. We design a parallelization scheme for sketch-based algorithms that summarize traffic information, which allows them to handle incoming data at high rates and be able to answer queries on that data efficiently, without overheads.</p> <p><br /></p> <p>In the second part of the thesis, we target the intermediate tier of computing devices and focus on the typical examples of hardware that is found there. We show how single-board computers with embedded accelerators can be used to handle the computationally heavy part of applications and showcase it specifically for pattern matching for security-related processing. We further identify key hardware features that affect the performance of pattern matching algorithms on such devices, present a co-evaluation framework to compare algorithms, and design a new algorithm that efficiently utilizes the hardware features.</p> <p><br /></p> <p>In the last part of the thesis, we shift the focus to the low-power, resource-constrained tier of processing devices. We target wireless sensor networks and study distributed data processing algorithms where the processing happens on the same devices that generate the data. Specifically, we focus on a continuous monitoring algorithm (geometric monitoring) that aims to minimize communication between nodes. By deploying that algorithm in action, under realistic environments, we demonstrate that the interplay between the network protocol and the application plays an important role in this layer of devices. Based on that observation, we co-design a continuous monitoring application with a modern network stack and augment it further with an in-network aggregation technique. In this way, we show that awareness of the underlying network stack is important to realize the full potential of the continuous monitoring algorithm.</p> <p><br /></p> <p>The techniques and solutions presented in this thesis contribute to better utilization of hardware characteristics, across a wide spectrum of platforms. We employ these techniques on problems that are representative examples of current and upcoming applications and contribute with an outlook of emerging possibilities that can build on the results of the thesis. </p> on Research with Simon Olsson<p>Online, Zoom</p><p></p>​<span style="font-weight:700">Welcome to Chalmers AI Research Centre, CHAIR Spotlight on Research. In this series of AI short talks it is time to meet Simon Olsson, since quite recentely Assistant Professor for Applied Artificial Intelligence at the Data Science and AI section of Computer Science and Engineering </span><span style="background-color:initial;font-weight:700">at Chalmers. O</span><span style="font-weight:700">n Friday October 9th h</span><span style="background-color:initial;font-weight:700">e will talk about Machine learning for the molecular sciences.</span><span style="background-color:initial">​</span><div><br class="Apple-interchange-newline" />The first series of CHAIR Spotlight Research talks have the theme “New Chalmers researchers on the spot!”, meaning that researchers that came to Chalmers in the last three years are giving these talks. ​​​</div> <div><br /></div> <div><strong>Abstract:</strong> In this talk, I will briefly outline synergistic opportunities between the molecular sciences and machine learning. I will then outline two of our recent papers to illustrate how machine learning can play a critical role in the sampling and approximation problem facing molecular simulations. I will end on a few comments on how these insights can inspire future research in machine learning.</div> <div><br /></div> <h3 class="chalmersElement-H3">Registration will open soon.</h3> <div><br /></div> <div><strong><img src="/SiteCollectionImages/Centrum/CHAIR/events/simon_olsson_180px.jpg" class="chalmersPosition-FloatLeft" alt="" style="margin:5px 10px" />About the speaker:</strong> In 2013 the University of Copenhagen awarded Simon Olsson a Ph.D. in Bioinformatics for his work in probabilistic modeling of protein ensembles from averaged data under Thomas Hamelryck. Following his Ph.D., he was award several postdoctoral fellowships to spend time, first in Switzerland at the ETH Zürich and IRB Bellinzona, and later in Germany at the Freie Universität Berlin. Simons research focuses on the interface between machine learning and experimental, theoretical, and computational aspects of the natural sciences. As for October 2020, Simon is Assistant Professor for Applied Artificial Intelligence at the Data Science and AI section of Computer Science and Engineering, Chalmers</div> <div><br /></div> <div><h3 class="chalmersElement-H3">About CHAIR Spotlight on Research</h3> <div>Chalmers AI Research Centre, CHAIR Spotlight on Research is a series of AI short talks hosting researchers from Chalmers. The seminars are targeted towards experts from the Chair Consortium core partners as well as other Chalmers researchers.</div> <div><br /></div> <div>Our aim is to increase awareness of AI at Chalmers between Chalmers researchers and AI experts in industry. In the seminars, speakers present an overview of their current research and thoughts for new research, ideas, challenges – anything they believe to be of interest for other researchers. The seminar is taking place online and is scheduled to contain 30 minutes of presentation and 15 minutes of discussion. </div> <div> </div> <div>The seminars are open to all and are free of charge. CHAIR Spotlight Research talks are taking place on Fridays 13:00-13:45.​​</div></div> AI Talks: Arthur Gretton<p>Zoom</p><p>Learning to generate realistic-looking images</p><p></p> <div><h2 class="chalmersElement-H2"><span>Critics for generative adversarial networks: results and conjectures</span></h2></div> <h3 class="chalmersElement-H3"><font face="open sans, sans-serif">October 14, 2020, 1:00-2:00 pm, Swedish time, Zoom</font><br /><font face="open sans, sans-serif">Zoom link: </font><span lang="EN-US" style="background-color:initial"><font color="#5b97bf" face="open sans, sans-serif"><span></span> </font><font>(Password to be posted) </font></span></h3> <p><br /></p> <p><strong>Abstract: </strong>Generative adversarial networks (GANs) use neural networks as generative models, creating realistic images that mimic real-life reference samples (for instance, images of faces, bedrooms, and more). These networks require an adaptive critic function while training, to teach the generator network how to improve its performance. To achieve this, the critic needs to measure how close generated samples are to true samples, and to provide a useful gradient signal the generator network.</p> <p>I will explore the design of critic functions for GANs, including f-divergences as used in the original GAN design, and integral probability metrics (such as the Maximum Mean Discrepancy) as used in later GANs. I will provide some observations and conjectures on critic design: in particular, a problem-specific critic seems to be helpful, and the critic needs to be deliberately weakened to ensure good GAN performance.</p> <p></p> <h2 class="chalmersElement-H2">About the speaker</h2> <p><img class="chalmersPosition-FloatRight" src="/SiteCollectionImages/Centrum/CHAIR/events/AI_Talks_ArthurGretton_180px_rev.jpg" alt="" style="margin:5px" /> </p> <p>Arthur Gretton is a Professor with the Gatsby Computational Neuroscience Unit, and director of the Centre for Computational Statistics and Machine Learning (CSML) at UCL. He received degrees in Physics and Systems Engineering from the Australian National University, and a PhD with Microsoft Research and the Signal Processing and Communications Laboratory at the University of Cambridge. He previously worked at the MPI for Biological Cybernetics, and at the Machine Learning Department, Carnegie Mellon University.</p> <p>Arthur's recent research interests in machine learning include the design and training of generative models, both implicit (e.g. GANs) and explicit (high/infinite dimensional exponential family models), nonparametric hypothesis testing, and kernel methods.</p> <p>He has been an associate editor at IEEE Transactions on Pattern Analysis and Machine Intelligence from 2009 to 2013, an Action Editor for JMLR since April 2013, an Area Chair for NeurIPS in 2008 and 2009, a Senior Area Chair for NeurIPS in 2018, an Area Chair for ICML in 2011 and 2012, a member of the COLT Program Committee in 2013, and a member of Royal Statistical Society Research Section Committee since January 2020. Arthur was program co-chair for AISTATS in 2016, tutorials co-chair for ICML 2018, workshops co-chair for ICML 2019, program co-chair for the Dali workshop in 2019, and co-organsier of the Machine Learning Summer School 2019 in London.</p> Ethics online: A conversation about (mostly) near-term societal effects of AI<p>Online, Zoom</p><p>​Speakers: ​​Pontus Strimling and Olle Häggström 20 October, 2020</p><div><span style="background-color:initial"><br /></span></div> <div><span style="background-color:initial">Pontus Strimling leads the research theme &quot;New Technologies and the future of humanity&quot; at the Institute for Future Studies, and also serves as assistant manager at the Centre for Cultural Evolution at Stockholm University. He is a methodically broad researcher of social science who investigates how culture changes. Presently he is actively working on how AI will influence society in the coming 15 years and why values and norms change. </span><br /></div> <div><div><br /></div> <div> Olle Häggström is a professor of mathematical statistics at Chalmers, where he also serves as chairman of CHAIR's AI ethics committee. They will discuss a range of topics in the future of AI, with particular focus on Pontus' research on the technology's near-term societal effects. </div> <div><div><span class="text-normal"><span class="text-normal"><h2>About AI ETHICS at Chalmers</h2></span>A  series of seminars highlighting ethical perspectives of artificial intelligence. The series will feature invited speakers and Chalmers researchers with the aim of cultivating an informed discussion on ethical issues. The seminars are organised by the <a href="">AI Ethics Committee</a> , within Chalmers AI Research Centre (CHAIR). </span></div> ​<br /></div></div> Ethics online: What Fiction Can Teach Us About AI Ethics<p>Online, Zoom</p><p>​Speaker: Kathryn Strong Hansen​17 November 2020</p><strong>​​Abstract:</strong>  Kathy Strong Hansen will talk about the ways that fiction serves as a useful springboard to ethical reflection on AI issues by explaining her Tracks course on the same subject. In addition, she will point out some of fiction's other benefits for those in scientific and technical fields. ​<div><br /></div> <div><strong>About the speaker</strong></div> <div><img src="/SiteCollectionImages/Centrum/CHAIR/events/Kathryn_StrongHansen_180px.jpg" class="chalmersPosition-FloatRight" alt="" style="margin:5px" /><br />Kathryn Strong Hansen is a senior lecturer in language and communication at the Department for Communication and Learning in Science at Chalmers University of Technology. She has a Ph.D. in English literature from the University of Southern California. Her research ranges from pedagogical explorations of the ways that the study of fiction is beneficial in the teaching of science and technology to literary analysis of young adult literature.</div> <div><br /> <div><div> </div></div> <h3 class="chalmersElement-H3">Registration for this seminar will open in November</h3> <div><br /></div></div> ​​​​