Seminars

The information on this page is not completed and schedule is tentative. Please check back later for more information.

  • Title: The State of Hobbyist Microcontroller Based Graphics and Interactivity
    Speaker: Trevor Tomesh

    Date: Monday October 19
    Time: 12:30pm - 1:30pm
    Place: CL 417

    Abstract: While hobbyists have experimented with producing graphics from microcontroller units since the late nineties, there is very little in the way of academically valid documentation on such projects. Since embedded computing is becoming more ubiquitous and MCU driven displays are now appearing everywhere -- including on our wrists -- this application is becoming of particular interest, especially for hobbyists / makers. The purpose of this seminar is to provide an overview of hobbyist MCU graphics projects. It is presented in a chronological sequence and includes primary research interviews from various players along with technical overviews of noteworthy projects.

  • Title: Machine Learning Techniques for Fraud Detection
    Speaker: Khulood Almonami

    Date: Wednesday October 21
    Time: 12:30pm - 1:30pm
    Place: CL 417

    Abstract: Fraud detection is a significant subject and refers to processes that usually rely on machine learning techniques to protect the online systems of organizations and businesses. An important question, therefore, is how to build a fraud detection system that achieves a high quality of performance. We studied numerous fraud detection systems in detail in order to identify advantages and disadvantages of underlying machine learning techniques. This study is a starting point for establishing the foundations of a robust fraud detection system for online auctions. Our goal is to illustrate an approach that will reduce the number of successful fraud attempts.

  • Title: Knowledge Extraction from the Web -- Overview and Challenges
    Speaker: Dr. Denilson Barbosa

    Date: Friday October 23
    Time: 12:30pm - 1:30pm
    Place: CL 417

    Abstract: This talk offers a broad overview of some of the work on natural language understanding done at the University of Alberta, with emphasis on information extraction for the purpose of constructing or updating knowledge bases. This is a very hot and challenging research topic, given the numerous practical applications it enables, including natural language understanding, question answering and semantic search. We will present some of these applications, the current architecture of the state-of-the-art in knowledge base construction, and discuss in more detail our own work on extraction of knowledge from natural language text. We will also briefly discuss the problem of Knowledge Fusion, which aims at finding a consistent knowledge base given a large number of disparate and possible conflicting extractions. We will conclude with a brief list of open problems in the area.

    Bio: http://www.cs.ualberta.ca/~denilson
    Denilson Barbosa (PhD, Toronto, 2005) is an Associate Professor of Computing Science at the University of Alberta. His interests are on databases, the Web, information retrieval and natural language processing, with recent emphasis on information extraction from semi-structured and unstructured sources. He serves as Associate Editor of IEEE Transactions on Knowledge and Data Engineering. He was a PC co-Chair of the 2015 Canadian AI Conference and he has served on the PC of several database and Web conferences; he was also Associate Editor of ACM's SIGMOD Record. He is the recipient an Alberta Ingenuity New Faculty Award, an IBM Faculty Award, the Best Paper Award at the 2010 IEEE Conference on Data Engineering, and he supervised the recipients of the Best Undergraduate Poster Award at the 2012 ACM SIGMOD Conference. He was a Visiting Scientist at the Max-Planck Institute for Informatics in Germany from July 2014 to April 2015, and a Visiting Professor (BIT) at the Free University of Bozen-Bolzano in Italy, during the Summer of 2008. He was a principal investigator and the Theme Leader for Data Quality of Canada's NSERC Business Intelligence Network.

  • Title: Introduction to Artificial Neural Networks
    Speaker: Kevin Yu

    Date: Wed, Oct 28, 2015
    Time: 12:30pm - 1:30pm
    Place: CL 417

    Abstract: Modelled after the neural structures of the cerebral cortex, neural networks are computing systems composed of a large number of simple, interconnected elements. Neural networks process information collectively through its nodes in such a way that it can adapt to, and learn from, its inputs. Applications of neural networks include artificial intelligence, data processing, pattern recognition, data mining, and robotics. In this presentation, I will provide an introduction to artificial neural networks, including a history, technical analysis, real-life applications, and an example of a neural network in action.

  • Title: In-auction Fraud Detection Using Machine Learning
    Speaker: Swati Ganguly

    Date: Mon, Nov 2, 2015
    Time: 12:30pm - 1:30pm
    Place: CL 417

    Abstract: With the growing popularity of online shopping and internet auctions, fraud and scams are becoming very common. Because of the lack of identity transparency and visibility of product details, it is easier to carry out illegitimate activities in online auctions. As per the Consumer Sentinel Network 2015 report published by Federal Trade Commission, Internet auction fraud accounts to about 1% of the total crime reported. Moreover, in 2014 Internet Crime Complaint Centre reports a loss of $11 million due to auction fraud. Regarding pre and post auction fraud, measures like policing, security enhancement and law enforcement have been enforced. Whereas for in-auction fraud, machine learning techniques like SVM, Neural Network, Bayesian Networks, Decision Trees, Hidden Markov Model, Regression Analysis are applied.

  • Title: Determining Three-way Decision Regions by Using Game-Theoretic Rough Sets to Solve Gini Objective Functions
    Speaker: Yan Zhang

    Date: Wed, Nov 4, 2015
    Time: 12:30pm - 1:30pm
    Place: CL 417

    Abstract: Three-way decisions divide all objects into three disjoint regions, i.e., acceptance, rejection and non-commitment decision regions. The determination of decision regions remains a challenge in three-way decisions and rough sets. In this research, we use Gini coefficient to measure impurity of decision regions. Gini objective functions are formulated to optimize impurities of multiple decision regions. Moreover, we use Game-theoretic rough set model (GTRS) to find the solutions to Gini objective functions. Compromise solutions can be obtained with GTRS by formulating competition between decision regions. The Gini objective functions, game formulation, Nash equilibrium, and iteration learning mechanism are investigated. An example to demonstrate how compromise decision regions can be obtained by using GTRS to solve Gini objective functions is presented.

  • Title: FABRIK: An iterative approach to the Inverse Kinematics problem
    Speaker: Victoria Verlysdonk

    Date: Fri, Nov 6, 2015
    Time: 12:30pm - 1:30pm
    Place: CL 417

    Abstract: The Inverse Kinematics (IK) problem is defined as finding a suitable set of joint configurations that produce the desired positions for the end effectors in a kinematic chain. In the context of computer animation, a solution to the IK problem preferably provides realistic, agile, smooth and accurate motion for the end effectors. In this seminar, we discuss the heuristic method Forward And Backward Reaching Inverse Kinematics (FABRIK) (Aristidou and Lasenby, 2011), which proposes an iterative approach to the IK problem that avoids computation of rotational angles or matrices, and we compare this method to other popular methods in terms of reliability and computational cost.

  • Title: Multi-Objective Optimization in Multi-Attribute Multi-Unit Combinatorial Reverse Auctions
    Speaker: Shubhashis Shil

    Date: Monday, November 9, 2015
    Time: 12:30pm - 1:30pm
    Place: CL 417

    Abstract: Almost all real-world optimization applications are basically non-linear programming problems that involve multiple conflicting objectives. Evolutionary Approaches (EAs) have been shown to be the best to address Multi-Objective Optimization Problems (MOOPs) because they guarantee the solution optimality. We represent the winner determination problem in Combinatorial Reverse Auctions (CRAs) as MOOPs. Thanks to EAs, we can generate a set of Pareto optimal solutions. By nature dealing with only multi-items makes CRA a NP-complete problem. Considering multi-units and multi-attributes with implicit and explicit constraints adds more complexity. We would like to apply Genetic Algorithm, a well-known evolutionary optimization method, along with prominent mechanisms, such as fitness assignment, fitness sharing, convergence, diversity, and elite solutions storage with external population.

  • Title: Learning erasing pattern languages from queries when given the shortest positive example
    Speaker: Fahimeh Bayeh

    Date: Friday, November 13, 2015
    Time: 12:30pm - 1:30pm
    Place: CL 417

    Abstract: Machine learning of pattern languages has applications in different areas like biology and text mining. Various methods have been proposed with the goal of learning pattern languages from minimal given information. In the query model, information is given to the learning algorithm in the form of answers to queries, where the latter are chosen by the learning algorithm itself. One type of query models is query learning with additional information. In this model, it is assumed that a word in the target language is given before the querying begins. In this presentation, I will discuss the effects of providing the shortest string in the target language as additional information.



To Top of Page