Seminars

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

  • Title: Principles and Implementation of Dashboard Design
    Speaker: Regan Meloche

    Date: Oct 31 (F)
    Time: 3:30pm - 4:30pm
    Room: CL 410

    Abstract: The rate at which modern software systems can gather and process data is unprecedented. This data, however, is useless if it cannot be interpreted and communicated effectively. Information visualization plays an important role in any domain where data analysis is necessary, including scientific research, business management, and social media analytics. This presentation will discuss a number of key design principles that can be applied to dashboard design, which will be highlighted using relevant examples. These principles include human perception and visual encoding. A platform for developing web-based dashboards (D3) will be emphasized as a key implementation technique for the design principles discussed earlier. The benefits of using D3 will be illustrated using a series of live demos.

  • Title: Introduction to mining data streams and an anomaly detection approach for tweet sentiment data streams
    Speaker: Khantil Patel

    Date: Nov 05 (W)
    Time: 3:30pm - 4:30pm
    Room: CL 410

    Abstract: Today, social media has become a popular platform where people like to share their thoughts, feelings, opinions and reactions to world events. Such activities on social media create huge amounts of data every minute. The Twitter social media platform like Twitter shares the user’s activities in terms of tweets through a data streaming application programming interface (API). Tweets arrive in real-time at high frequency. In this presentation we model tweets as a time series of tweet sentiments. We review the existing data stream mining approaches and propose our approach for mining data stream using the well-known Storm framework. We propose an anomaly detection scheme that will allow us to detect whether a stream of tweet sentiments contains abnormal peaks.

  • Title: Ontology-based GIS for Land Use Suitability Analysis and Simulation
    Speaker: Munira Al-Ageili

    Date: Nov 07 (F)
    Time: 3:30pm - 4:30pm
    Room: CL 410

    Abstract: This work presents the development of Ontology-Based Information Extraction (OBIE) System for automating the extraction of criteria and values applied in land-use suitability analysis (LUSA) from Saskatchewan Regulations documents. The system we developed integrates a variety of tools and resources in order automate the information extraction process and populate the ontology. The work on LUSA OBIE system includes the selection of application domain related documents, domain ontology construction, linguistic pre-processing, ontology-based semantic annotation and ontology population. The ontology is used to represent domain knowledge and guide the information extraction process. The ontology is also populated with the extracted information, and can be used for semantic queries and reasoning, or can be can be exported to a knowledge base, a database or an xml file for further analysis.

  • Title: Sentiment Analysis of Tweets
    Speaker: Credell Simeon

    Date: Nov 12 (W)
    Time: 3:30pm - 4:30pm
    Room: CL 410

    Abstract: The increasing popularity of the microblogging application Twitter in recent years has resulted in several research studies on the sentiment analysis of tweets. Tweets which are user-generated statements of opinion are classified as either positive, negative or neutral based on their expressions of favorability, unfavorability or objectivity towards a particular topic respectively. Furthermore, tweets are restricted to 140 characters and contain many Twitter-specific features which increase the difficulty of the sentiment analysis task. In order to overcome this problem, studies have applied supervised machine learning techniques which determine sentiment based on learning algorithms, and lexicon-based approaches which use the prior polarities of opinion words. This seminar will compare and contrast both approaches.

  • Title: An Addition-deletion Strategy for Reduct Construction
    Speaker: Kuifei Liu

    Date: Nov 14 (F)
    Time: 3:30pm - 4:30pm
    Room: CL 410

    Abstract: Rough sets theory has been applied in many fields, such as artificial intelligence and data mining. In rough set theory, an important concept is attribute reduction. In this talk, we first discuss a general framework for reduct construction algorithms from a two-level view: a control strategy and an attribute selection heuristics. We then examine an addition-deletion strategy for constructing an attribute reduct. The attribute selection heuristic is given by an entropy-based fitness function. Finally, we briefly summarize some advantages and disadvantages of the addition-deletion strategy compared to other types of strategy.

  • Title: Constructing objects with sharp features from volumetric data
    Speaker: Fatemeh Bayeh

    Date: Nov 17 (M)
    Time: 3:30pm - 4:30pm
    Room: CL 410

    Abstract: Using volumetric data for representing objects in a 3D world is fast and it facilitates some operations on constructed surfaces. To generate objects from volumetric data, we need to extract information about the surface that is implicitly present in the volumetric data, and use this information to construct the surface. The well-known method for this purpose is Marching Cubes. However, this method is not able to generate sharp features. If the size of each volume element is decreased and an octree is used, the created surface will be more accurate, but still sharp features are not constructed properly. We propose a method to construct objects with finer sharp features using Hermit Data (i.e. intersection points and normal vectors).

  • Title: Preference-based Constraint Optimization for Online Shopping
    Speaker: Bandar Mohammed

    Date: Nov 19 (W)
    Time: 3:30pm - 4:30pm
    Room: CL 410

    Abstract: Constraints and preferences coexist in many real world applications. In this seminar I will talk about the benefits of managing both preferences and constraints for interactive applications through the online shopping system that I recently developed. Given online shoppers' requirements and preferences, the proposed system provides a set of suggested products meeting the users' needs and desires. This is an improvement to the current shopping websites where the clients are restricted to choose among a set of alternatives and not necessarily the selections that meet users' satisfaction and needs. In addition to eliciting constraints and preferences from the users, the system has also the ability to recommend products based on its learning component.

  • Title: Periodicity-based Swimming Performance Feature Extraction and Parameter Estimation
    Speaker: Yang Zhao

    Date: Nov 21 (F)
    Time: 3:30pm - 4:30pm
    Room: CL 410

    Abstract: This paper presents a novel approach for extracting swimming performance parameters from acceleration data using techniques traditionally applied to audio analysis. The recorded acceleration data is treated as sampled audio data, with the stroke rate (one of the main parameters to extract) treated as the fundamental frequency. A pitch detection algorithm is then adapted to this domain and applied to the data. Lap counts are derived from an analysis of the output frequency series as percussive events, and stroke counts are obtained from a pitch-aligned adaptive spectral extraction algorithm. The comparable accuracy from the experiments shows the superiority of this approach over conventional methods, especially considering that this method is not dependent on filtering and is robust to sensor displacement and axial alignment.

  • Title: AI Adaptations for Fighting Games
    Speaker: Daniel Lavin

    Date: Nov 24 (M)
    Time: 3:30pm - 4:30pm
    Room: CL 410

    Abstract: In a fighting (or action) video game, a person plays against a sequence of one or more enemies, using combinations of kicks, punches, or special attacks, until either the player or the enemies' health points are reduced to zero. Players need quick reflexes, knowledge of their own abilities, and sometimes the ability to predict their opponent's next move. Reinforcement learning is an AI technique that can follow and learn from an opponent's patterns, and find optimal solutions through trial and error. Since each person has a different playstyle, using reinforcement learning allows the game AI to adapt to a specific player's abilities. In this presentation, we examine several reinforcement learning techniques and explain why they are appropriate for fighting games.

  • Title: A Volumetric Approach to the Real-Time Simulation of Granular Soils
    Speaker: Andrew Geiger

    Date: Nov 26 (W)
    Time: 3:30pm - 4:30pm
    Room: CL 410

    Abstract: There has been an increased demand recently for physically realistic simulations of granular soils, such as sand, dirt, and loam, within dynamic 3D environments. In these applications, it is critical that the simulated soils behave naturally when excavation activities are performed by the user at arbitrary locations on the surface of the soil. More specifically, a realistic simulation of granular soils requires the grains of the soil to be separable and joinable, through interaction with external objects, and to experience soil slippage, a phenomenon that causes masses of soil to slide downwards along steep, unstable slopes. In this presentation, we will discuss our volumetric approach to the simulation of granular soils which is designed for parallel execution on modern GPUs.

  • Title: Hierarchical Parallel Genetic Algorithms and the University Timetabling Problem
    Speaker: Barret Rennie

    Date: Nov 28 (F)
    Time: 3:30pm - 4:30pm
    Room: CL 410

    Abstract: The university course timetabling problem is an NP-complete combinatorial problem that universities face every term. The International Timetabling Competition 2007 specified a specific variant of the post-enrolment problem (that is, the problem where classes are scheduled after student enrolment happens). Constraints involved in this problem are introduced, as well as the subject of hierarchical parallel genetic algorithms (HPGA). An implementation of such an HPGA (called Spaghetti) was developed over a research term. Attempts were made to use Spaghetti to solve several instances of this problem. The algorithms involved in improving the quality of solutions in Spaghetti are also be discussed.

  • Title: Your Brain Inside of a Computer: Neural Nets
    Speaker: Ian Hauser

    Date: Dec 01 (M)
    Time: 3:30pm - 4:30pm
    Room: CL 410

    Abstract: Neural Nets are a tool used in the field of Artificial Intelligence to help computer systems solve problems and learn new concepts. They employ a kind of bio-mimicry; emulate the neurons and connections of a human brain in order to learn how to solve problems the way people do. This seminar will discuss the uses of Neural Nets as they relate to common Computer Science problems, show what kind of issues with traditional problem solving approaches they surpass, give some real-world examples of Neural Nets in use today, and discuss why they fall short of the ultimate goal of Artificial Intelligence – re-creating human intellect.

  • Title: Decorator Design Pattern: An Exciting Way to Add Flexibility to Our System
    Speaker: Abbas Shahid Khwaja

    Date: Dec 03 (W)
    Time: 3:30pm - 4:30pm
    Room: CL 410

    Abstract: At times there comes a situation when creating separate classes for every distinct type becomes a non-efficient and non-effective procedure. It becomes costly to compute a complex system when there is large number of classes involved, such that they are linked in inheritance. To discuss it, I will be briefing on how these problems can be solved using one of the most useful design pattern, i.e., Decorator Pattern. With the help of examples, I will be explaining in detail about what Decorator Pattern is all about and how we can use it. I will also be explaining in details about the design and implementation of Decorator Pattern and how we can add functionality using classes. Instead of rewriting the code, I will be using Open Closed Principle to show how we can simply use the old code by extending the new features.

  • Title: CANCELLED! Speech Recognition: How it Works and How Far it Has Come
    Speaker: Brandon Prevost

    Date: Dec 05 (F)
    Time: 3:30pm - 4:30pm
    Room: CL 410

    Abstract: Speech recognition is the process of taking spoken word as input to a computer program and converting it into coding patterns to which meaning is assigned. Speech recognition software has recently become increasingly imbedded into our lives. It is built into our operating systems, telephone menu systems, smart phones, and cars. People use speech recognition daily but how does it actually work? This presentation will explore the various approaches to speech recognition such as template matching and feature analysis. It will also explore the changes in algorithms and models used in the development of speech recognition software that have allowed computers to recognize strings of spoken dialog instead of only single words.



To Top of Page