Projets novateurs réalisés

AI-enabled Performance Enhancement for the Reconfigurable Multi-Player RAN

In 5G and beyond networks, softwarization of network functions, as well as disaggregation of software and hardware, are the recent moves pushing Radio Access Networks (RAN) to be ultra-agile, reconfigurable and flexible. This flexibility comes along with complexity that goes beyond traditional algorithms’ capabilities to optimize the RAN. In addition, in future RANs, multiple-players interacting within the same RAN environment will increase the burden on proper decision making. Many researchers in academia and the telecom industry have turned to AI/ML to handle the rising complexity of wireless networks. Dr. Erol-Kantarci, one of the leading researchers in the area, will join forces with Ericsson to address this bleeding-edge challenge and develop advanced machine learning tools for future RANs. With this project, her team will develop hierarchical and planned distributed learning techniques under partial observability to optimize the multi-player RAN based on policies. These techniques will provide Ericsson an edge over the rapidly changing technology scene.

Developing Indicators of Ecosystem Stress Using Long-term Data From Pristine and Physically-altered Experimental Lakes

Climate change and expansion of industrial development into the boreal zone are two, prominent stressors that Canadian boreal lakes face. This puts the quantity and quality of freshwater fish habitat at risk, and the subsequent productivity of fish populations in a given system. The objective if this work is to define the components of fish habitat (e.g. temperature, oxygen, nutrients, prey availability, structural features) that are most strongly linked to fish productivity. We will use a combination of (1) long-term fisheries and limnological monitoring data (~40 – 50 years) from pristine lakes at the IISD-Experimental Lakes Area (IISD-ELA) in northwestern Ontario, (2) fisheries and limnological data collected from lakes before and after they were manipulated to mimic impacts from hydroelectric developments (i.e. large reductions in water balance), and (3) multi-year, high-resolution data tracking the movement and habitat use of lake trout in several systems, in order to define the best ecosystem indicators of the productivity of fish populations. This work will help to identify climate and industrial development impacts on fish populations, and less-costly and more efficient ways to monitor fish productivity in stressed ecosystems.

Detecting, Extracting and Merging Receipts from Uploaded Smartphone Images

Sensibill provides financial tools like digital receipt data that help banks and credit unions better know and serve their customers. Users can upload digital images through tools and the company would do image processing first and then use processed images to analyze. However, the previous image processing algorithm is time-consuming for users and doesn’t satisfy the use case of long receipts that don’t fit on a single image. The project is aimed to build a method that can efficiently generate one clear final image from multiple uploaded images.

An Improved Approach to Watershed Management and Adaptive Decision Making in the Great Lakes

With collaboration between the Council of the Great Lakes Region, Pollution Probe and Lambton College, the proposed project is focused on continuing the development of an artificial intelligence visualization tool to enable users to select growth constraints and visualize resulting changes to watershed health, predict how watersheds will evolve over time and prescribe actions to protect them. This AI tool’s purpose is to analyze historical watershed data and to predict changes to a watershed over time. Looking into water quality and the minimum and maximum of water quality thresholds.

Robust Taxonomy out of Receipt Item Labels

The primary objective of this project is to implement a product taxonomy model that can reliably categorize receipt item labels to generate more personalized financial insights. Sensibill leverages unstructured receipt data to support personal and business finance management. Reliable categorization of receipt line items into specific merchant categories not only reduces the time for manual entry, but also helps customers better understand their spending patterns, as well as support banks and credit unions in providing individualized recommendations that are aligned with customers’ financial goals. Considering the continuously expanding and often closely related sets of product categories, a hierarchical product taxonomy that leverages the relationship – from general to specific – amongst receipt item categories would power downstream features, including more fine-grained categorization of historical purchases and delivery of more personalized financial insights. To achieve such results, we will implement and evaluate natural language processing and machine learning approaches for hierarchical product categorization.

Convolutional Neural Network for Demand Forecasting

Many retailers are interested in forecasting demand for the products they sell. Deloitte has used machine learning methods to tackle this problem in the past. However, this requires the creation of hand-crafted features based on product sales data, which is a costly and time-intensive process. Using alternative models to perform this task would remove the need for laborious data manipulation. It will also allow model enhancements to scale across many clients rather than requiring from-scratch data manipulation for each new client. Hence, this project will involve the development of a new machine learning model to predict product demand. The model will be trained using historical sales data. Iterative stages of model architecture and fine-tuning will give rise to the final model. Various enhancements to the model architecture will be explored. The main objective is for the final model to be integrated into a modular demand prediction solution at Deloitte.

Enzyme Immobilization on Surface-Enhanced 3D-Printed Structure

Large field trials as platforms to develop digital tools for intercropping

Evaluation of the OISEAU Application

The project is a program evaluation of the mobile application OISEAU: Agents of Nature, designed by the non-profit organization Morning Star Enterprises. Morning Star has developed individual OISEAU applications for six Calgary Parks locations and the launch is the summer 2013.The application is designed to increase children’s exposure and connection with nature, as well as improve their content knowledge and physical activity. Research consistently shows that children’s physical, cognitive, emotional, and moral development are enhanced by experiences with nature and that interactions with nature aid the development of self-concept, personal identity, and environmental protection. The research will be conducted by Ph.D. student Maxine Crawford, and supervised by Dr. Mark Holder, both of UBC Kelowna.

Augmented Reality as an Assistive Technology for Individuals with Visual Impairments

Safety tests and calorimetric aging investigation of lithium-ion batteries

The influence of working memory processes on the selective retrieval of episodic memory contents