Online Learning for Pinpoint Selling

This internship applies data mining and machine learning techniques to increase conversion rates in the interactive online marketing industry by providing personalized recommendation to specific customer segments. It will consist of two parts. The first part is to classify customers according to what message would resonate. The intern will apply some techniques to improve the accuracy of classification. The second part is to order resources from most to least interesting for website.

Real-time Animation of Deformable Objects for Computer Games

Radical Entertainment is a video game developer which creates and develops games for all current and next generation platforms. Each year, computer games become graphically more realistic due to advances in computer technology and animation. The object of this project is simulation of deformable objects (such as a mattress or jelly-like substances) in real-time for computer gaming purposes. The equations of motion for such objects is quite different and more difficult to solve than those used for rigid bodies.

Large-scale Physical Factors Associated with the 1999 – 2005 Canadian Prairie Drought

Droughts are among Canada’s costliest natural disasters, having major impacts on a wide range of sectors including agriculture, forestry, industry, municipalities, recreation and aquatic ecosystems. The 1999 – 2005 Canadian Prairie drought was the worst drought in at least 100 years in parts of the region and caused considerable environmental and economic hardship. Although much is known about droughts, there is still no complete understanding of why they occur.

Convex Decomposition of a Triangle Soup

Radical Entertainment is a video game developer which creates and develops games for all current and next generation platforms. Collision detection forms an indispensable part in today’s 3D game engines. Due to increasing size of the 3D objects used in game development, high-performance collision detection running directly on these objects requires a significant amount of work. To allow for real-time collision detection between a 3D object and other objects in the game environment, the typical approach is to first find for each object a set of best-fitting convex hulls.

Active Learning of Hierarchically Parameterized Policies

Next Level Games is a full-service videogame developer based in Vancouver, BC. This intern research project will investigate mathematical solutions to the incredibly difficult problem of sequential decision making in an uncertain, partially observed, multi-agent environment with realistic motor dynamics. The problem is formalized under the reinforcement learning framework, where the agent observes the world state, takes action, receives a reward and observes the new state.

User Behavior Modeling and Scalabitliy Analyzing for VoIP Network

Eyeball Networks is a leading developer of software for the VoIP, video telephony and instant messaging industry. In this project, the intern will investigate and model user behavior of VoIP networks developed by the company. Based on a statistical user behavior model, he will then develop a test tool to collect specific Quality-of-Service performance metrics, so as to analyze the scalability of system, and to formulate its relation with the user behavior model.

Transliteration from Arabic to English

Machine translation has been an active field in Natural Language Processing. Although the quality of translation cannot reach that of a human, it is getting closer. One of the major sub-tasks in this field is ‘transliteration’, which is mapping the letters from the source language to the letters of target language. It would be useful for translating the proper names, location names and any out-of-vocabulary word found in the source text.

Segment Pool Allocation Strategy

Video games must be able to present things in ‘real time’, with no delays thus performance is critical. A modern processor can deal much quicker with information that is in one continuous section of memory as opposed to divided into smaller chunks. When information is split, it causes a delay in processing. The intern’s research will focus on optimizing a new system for memory management that will attempt to achieve greater performance through a system that organizes memory into collections of whole blocks in order to prevent the need for data to be split.

Interaction of Log-Probabilities and Taxonomic Hierarchies

The intern’s project aims to provide a formal mathematical foundation to the weightings applied to terms used to describe complex models (eg mineral deposit models or landslide hazards) when measuring the similarity between models and existing natural phenomena such as mineral occurrences or seemingly dangerous slopes. These weightings are important to correctly matching models with existing natural phenomena, an activity which experts carry out mentally and which GeoReference Online Ltd has implemented on a computer.

Rate-Distortion Optimized Streaming of Fine-Grained Scalable Video Sequences

Video streaming over the Internet is increasing in popularity. Many video objects are made available online and many users are streaming video everyday. In addition, the increasing processing power of personal computers, and the availability of high-speed Internet services, encourages users to demand more, higher-quality video content. To provide high-quality and to accommodate clients with various resources, fine-grained scalable (FGS) video coding has been proposed as part of the recent MPEG-4 and H.264 standards.

Pages