Lunchtime Seminar
Archive Winter Semester 2017/2018
Predictive models for robotics
Lecturer:
Senka Krivic
Researcher at IIS group, University of Innsbruck
Date: Thursday, 1st of February 2018, 12:00 – 1:00
Venue: SR 1, ICT Building, Technikerstraße 21a, 6020 Innsbruck
Abstract:
Robots can handle complex tasks in highly controlled industrial settings. However, in real-world environments like the living room or an office, it is impossible to anticipate the conditions that robots will encounter. Real-world environments are very often dynamic, with a large number of objects and possible actions making the knowledge available to the robot incomplete. In this talk, I will present developed predictive models based on partial knowledge about the system and the environment that enable real-time operation of a robot in complex scenarios. Two examples will be addressed: online learning of predictive model for push-manipulation, and decreasing uncertainty in AI planning with state predictions.
Virtual payment channels over cryptographic currencies
Lecturer:
Lisa Eckey
Researcher at the Department of Computer Science, TU Darmstadt
Date: Thursday, 25th of January 2018, 12:00 – 1:00
Venue: SR 1, ICT Building, Technikerstraße 21a, 6020 Innsbruck
Abstract:
Blockchain technology and cryptographic currencies suffer a major setback: They do not scale! Payment channels promise to solve this problem and provide an efficient method for performing cheap transactions. In contrast to the traditional transactions, payment channels have the advantage that they allow nearly unlimited number of transactions between parties without involving the blockchain. In this talk Perun is presented, a new system for payment and state channels over cryptographic currencies. This solution for payment channels are more efficient than the existing solutions such as the Lightning Network and work over any statefull blockchain technology.
Fast and space-efficient indexing for main-memory database systems on modern hardware
Lecturer:
Robert Binna
Researcher at DBIS group, University of Innsbruck
Date: Thursday, 18th of January 2018, 12:00 – 1:00
Venue: SR 1, ICT Building, Technikerstraße 21a, 6020 Innsbruck
Abstract:
With the exponential growth of data available in the digital universe, fast and space efficient access to data is crucial. At the same time, the amount of main-memory has reached capacities that even large enterprise datasets can entirely be kept in RAM. To create state-of-the-art data management systems, it is important to optimize those systems for main-memory workloads on modern hardware, in particular for concurrency, cache and space efficiency. In this talk I will therefore present a novel trie based index structure for main-memory database systems, which provides space efficieny as well as fast access performance for arbitrary data sets.
Annotation based automatic action processing
Lecturer:
Elias Kärle
Researcher at STI, University of Innsbruck
Date: Thursday, 8th of January 2018, 12:00 – 1:00
Venue: SR 1, ICT Building, Technikerstraße 21a, 6020 Innsbruck
Abstract:
With a strong motivational background in search engine optimization theamount of structured data on the web is growing rapidly. The main searchengine providers are promising great increase in visibility throughannotation of the web page’s content with the vocabulary of schema.organd thus providing it as structured data. But besides the usage bysearch engines the data can be used in various other ways, for examplefor automatic processing of annotated web services or actions. In thistalk I am going to present an approach to consume and process schema.organnotated data on the web and give an idea how a best practice and usecases could look like.
Anonymous alone? On the anonymity of cryptographic currencies
Lecturer:
Malte Möser
PhD student in the Department of Computer Science at Princeton University and a Graduate Student Fellow at the Center for Information Technology Policy (CITP)
Date: Thursday, 14th of December 2017, 12:00 – 1:00
Venue: SR 1, ICT Building, Technikerstraße 21a, 6020 Innsbruck
Abstract:
Cryptographic currencies like Bitcoin balance two seemingly contradictory goals: radical transparency of the ledger and anonymity of users. A variety of privacy-enhancing techniques have been developed to strengthen anonymity, yet to this day their adoption remains limited. A closer look at existing deployments shows how in practice they often fall short of their promises due to inherent complexity and dependence on user behavior. In this talk I will give examples for such failures based on recent empirical work evaluating techniques for “mixing” coins in cryptographic currencies. I’ll discuss the importance of empirical measurements and conclude with a set of privacy guidelines for cryptocurrency designers.
Fluid simulation in computer graphics
Lecturer:
Fernando Zorrilla Medrano
Research assistant at IGS group, University of Innsbruck
Date: Thursday, 7th of December 2017, 12:00 – 1:00
Venue: SR 1, ICT Building, Technikerstraße 21a, 6020 Innsbruck
Abstract:
Fluid dynamics are one of the most challenging problems in science. The Navier-Stokes equations describe them but they remain unsolved. Many fluid solvers have achieved results very accurated to reality making use of iterative numerical methods with a very high computing time. In computer graphics a high accuracy is not as important as a faster computing time. Many different approaches to reduce computing time are in the state of the art of this field. He will present his Master thesis “Shallow water simulation” and introduce the topic of his PhD.
Multimodal Social Signals Analysis
Lecturer:
Nicu Sebe
Head of Dept. of Information Engineering and Computer Science
Leader of Multimedia and Human Understanding Group (MHUG), University of Trento
Date: Thursday, 30th of November 2017, 12:00 – 1:00
Venue: SR 1, ICT Building, Technikerstraße 21a, 6020 Innsbruck
Abstract:
Social perception is the main channel through which human beings access the social world, very much like vision and hearing are channels through which people access the physical world. For a long time, researchers have addressed the computational implementation of vision and hearing in domains like computer vision and speech recognition, but only early attempts have been made to do the same with social perception. We believe social perception to be one of the missing links in the communication between humans and computers, and so in this presentation I will present our recent research in social signals analysis (e.g., head pose, camera-based heart rate estimation). We will concentrate on behavior modeling and recognition, with an emphasis on sensing and understanding users’ actions and intentions for achieving multimodal human-computer interaction in natural settings, in particular pertaining to human dynamic face and body behavior in context dependent situations (task, mood/affect). Perspectives on multisensory observation will be addressed.
Tiling with decreasing diagrams and explicit sharing
Lecturer:
Vincent van Oostrom
Researcher at CL group, University of Innsbruck
Date: Thursday, 16th of November 2017, 12:00 – 1:00
Venue: SR 1, ICT Building, Technikerstraße 21a, 6020 Innsbruck
Abstract:
Term rewriting is a model of computation in which (functional) programs can be implemented at a natural level of abstraction. For example, the factorial function can be implemented using two simple term rewrite rules, one for the zero-case and one for the successor-case. Repeatedly applying these rules until none applies anymore, yields the result.
We discuss two aspects of this modelling:
How to establish that a given term rewriting system implements a function (irrespective of what the function actually is)? Will a result be obtained (existence, termination), and is the result independent of the order of applying rules (uniqueness, confluence)?
We present one of the core-methods of the CSI confluence tool, based on tiling the plane with so-called decreasing diagrams.How to guarantee to do only needed work, i.e. how to avoid doing non-necessary or duplicate steps? This is usually resolved by (sharing) graph rewriting. We present three basic operations involved, sharing, deletion, and boxing, and discuss the pitfalls of implementing these lazily.
Ontology-based Business Process Execution Log Extraction in Process Mining
Lecturer:
Ario Santoso
Research assistant at QE group, University of Innsbruck
Date: Thursday, 19th of October 2017, 12:00 – 1:00
Venue: SR 1, ICT Building, Technikerstraße 21a, 6020 Innsbruck
Abstract:
Process mining aims at discovering, monitoring, and improving business processes by extracting useful insights from business process execution logs. Through process mining, decision makers can discover process models from data, compare expected and actual behaviors, and enrich models with key information about their actual execution. To be applicable, process mining techniques require the input data to be explicitly structured in the form of an event log, which records various events related to the execution of various process instances. Unfortunately, in many real world settings, event logs are not always explicitly present. Often, event logs have to be extracted by integrating information from various data sources. To apply process mining in this widespread setting, there is a need for techniques that are able to support data and process analysts in the data preparation phase, and in particular in the extraction of event logs from legacy information systems. In this talk, a technique for event logs extraction that makes use of the ontology-based data access paradigm will be presented. This technique allows us to have a high-level view over the data repositories, capture different process-related views on the same data, and also facilitate multi-perspective process mining.
In Code We Trust? Measuring the Control Flow Immutability of All Smart Contracts Deployed on Ethereum
Lecturer:
Michael Fröwis
Research assistant at SEC group, University of Innsbruck
Date: Thursday, 12th of October 2017, 12:00 – 1:00
Venue: SR 1, ICT Building, Technikerstraße 21a, 6020 Innsbruck
Abstract:
Program code stored on the Ethereum blockchain is considered immutable, but this does not imply that its control flow cannot be modified. This bears the risk of loopholes whenever parties encode binding agreements in smart contracts. In order to quantify the issue, we define a heuristic indicator of control flow immutability, evaluate it based on a call graph of all smart contracts deployed on Ethereum, and find that two out of five smart contracts require trust in at least one third party. Besides, the analysis reveals that significant parts of the Ethereum blockchain are interspersed with debris from past attacks against the platform. We leverage the call graph to develop a method for data cleanup, which allows for less biased statistics of Ethereum use in practice.
Scientific Workflows on Manycore Computers
Lecturer:
Matthias Janetschek
Research Assistant at DPS group, University of Innsbruck
Date: Thursday, 5th of October 2017, 12:00 – 1:00
Venue: SR 1, ICT Building, Technikerstraße 21a, 6020 Innsbruck
Abstract:
Modern shared-memory heterogeneous manycore systems have becomeincreasingly complex to program, and traditional programming methods arestruggling to fully exploit the performance of today’s heterogeneoussystems. One of the most successful paradigms for implementinghigh-performance computing applications in Grid- and Cloud-computing arescientific workflows, originally designed to easily create large andcomplex applications by reusing existing (often legacy and monolithic) software components and assembling them through well-defined controlflow and data-flow dependencies. Because of the similarity in terms ofscale and heterogeneity, workflow systems represent today a promisingalternative for development and execution of scientific applications onshared-memory heterogeneous manycore architectures. To test this thesiswe implemented a new Manycore Workflow Runtime Environment (MWRE) whichwe specifically designed for the unique requirements of shared-memoryheterogeneus manycore systems. MWRE is compiler-based and translatesworkflows from a XML-based workflow specification language into anequivalent C++-based program. Experimental results show that MWRE canachieve performance similar to OpenMP while better supportingheterogeneity and concepts like full-ahead scheduling.