Semantic question-answering systems for customer service automation

Recent advances in applications of deep learning in natural language processing has provided potential opportunities in building robust information retrieval and conversational models that require far less hand-crafted features for understanding the intent of queries and ultimately building question-answering systems. In particular, there has been several advances in factoid question-answering systems and some recent attempts to moving beyond factoid questions.

Learning Depth from Transmissive Diffraction Mask based Sensor

Most cameras today, take photographs by measuring the amount of light in a scene. This means that they discard any information about the directionality of light, which results in a two-dimensional representation of the three-dimensional scene. We have developed a new type of camera that can detect the direction of light as well as image intensity. The 3D scene can be estimated by leveraging this additional information. Due to our manufacturing constraints, the signal from our camera is not ideal, hence the reconstruction process is not straightforward.

Building Scalable Business Transaction and Data Mining Systems For Insurance Workloads

Farmers of North America (FNA) and FNA Strategic Agriculture Institute (FNA STAG) are two Canadian organizations dedicated to maximizing farm profitability. They collect and analyze demographic, legal, marketing and relevant data about its producers and partnering commodity organizations to understand the farmer market need and create strategies for business operation functionality. With this project, the organizations will get two database systems, the market/consumer research and distributed database.

A New Framework for Method-Level Dynamic Software Updating

Dynamic Software Updating (DSU) is a necessity in the operation of a large computing infrastructure that must deliver high availability. An issue very relevant to the industry is that legacy applications running in data centers were often designed and constructed without due consideration for the need of partial upgrades in the future. The goal of this research is the development of a framework that allows for an automatic retrofitting of legacy applications to enable a selected set of individual methods to be dynamically updated.

Investigation on object recognition on very large volumes of image data

Istuary Innovation Labs, Inc partners with innovators, entrepreneurs, technical talent and experts in North America to build technology companies for fast-growing markets, such as China, and to capitalize on quickly expanding and unmet demand. Founded in 2013, Istuary’s core business consists of Innovation Labs and Idea Labs. We partner with Istuary Venture Capital to fund development of technology internally through Innovation Labs and externally through Idea Labs.
Investigation on object recognition on very large volumes of image data will be conducted.

Advancing Unstructured Data Extraction and its Use in Geoscience

Unstructured data refers to data that is present in reports, web pages, newspapers and other media. Such data is the most common data that we see around us and yet no modern tools exist to extract information from it. In this project we will develop techniques to extract the data and apply it to geoscientific reports in order to aid in the discovery of new mines and other geoscience applications.

Photonic sensors for rapid and selective detection of bacteria in water - Year two

In recent years monitoring and protection of food and water resources became a priority of governments worldwide. Bio-hazards are potential threat for these resources thus need to be addressed both in industry and in academia. Therefore, developing an accurate, fast and cost effective technique for detection of pathogenic strains called for increased demand on the areas targeted by the fiber-optic systems.

Investigating the Implementation of Machine Learning Algorithms on Adiabatic Quantum Solvers Year Two

Machine learning is an active field of research and development to provide tools and technologies for finding significant patterns in data. Behind every face detection and face recognition software in digital cameras or social network websites a constantly under-development machine learning algorithm is working. Nowadays in any practical applications of machine learning we have to analyze huge amounts of data. Using classical approaches to train machine learning algorithms for some classes of algorithms is either very slow, requiring a lot of computing resources, or inefficient.

Improving the Performance of Java Virtual Machine (JVM) Garbage Collection using Transactional Memory

As the multi-processing power of computers continues to grow, traditional methods of memory management become more and more problematic. The purpose of this project is to quantitatively analyze existing memory management tactics to determine whether they can be improved using techniques that take advantage of modern hardware, or instead, whether brand new methods for managing memory need to be developed. As the act of memory management is one which is crucial to almost every application that runs on IBM’s J9 JVM, clearly identifying a path forward will be extremely beneficial.

Spatial Sentiment Visualization for Text Data

Large numbers of online comments about a public consultation from citizens makes it difficult for decision-makers on public projects. To clarify various cognitions of participants is important. But reading through all comments, especially long ones, is unrealistic due to the high cost. An example dataset is 974 comments about whether residents support a new 0.5% Metro Vancouver Congestion Improvement Tax from PlaceSpeak. This brings out the need for text analysis.