The ability of the health system to manage a massive influx of patients is based on the combination of four factors: the personnel, the equipment, the physical spaces and the system in place. A combination better known in jargon as the 4 "S" (staff, stuff, structure / space, system). A fifth factor that is often misunderstood is synchronicity.
Hydro-Québec is a public utility that generates and distributes electricity. Despite selling most of its electricity in Québec, its most lucrative sales are in the neighboring markets. To ensure the best possible quality of service, the transmission system must remain stable, but to maximize profits, the company also wants to increase its transmission capacity to maximize energy exports. The transfer limit is now conservatively estimated based on a certain combination of simulated network configurations.
Robotics vehicles deployed at Hydro-Québec up to now are still mainly manually operated and human intervention is continuously required. The project aims to equip Hydro-Québec's current and future fleet of inspection robots with autonomous inspection capabilities. The intern will leverage breakthroughs in artificial intelligence to enable robotic vehicles to realize real-time automated visual inspection of the company's infrastructure and use a simply and securely deployable robotic vehicle to perform the company’s first fully autonomous power line components inspection mission.
The goal of the project is to facilitate the research and development of new drugs using machine learning. More specifically, exploring new techniques to model molecules and how they can be represented. Doing so will involve training deep learning models on multiple small datasets with the objective of improving the generalization performance on new tasks. The trained models will have to be accurate even in the context of new types of molecules. With the small amount of data available, out-of-distribution techniques will be used.
In a trend towards increasingly complex aircraft, tomorrow’s pilots must possess excellent situational awareness, problem solving, leadership and communication skills. Pilots trained for these competencies will be best equipped to handle unforeseen situations safely. In this project at Paladin AI, intern will focus on working with real flight simulator data that has been labeled by human experts. Intern will work with this data to develop new analytics techniques for inferring pilot competency. The intern will identify one or multiple markers of either good or poor pilot performance.
This project aims to predict demand and inventory shortages for better inventory management in health care institutions. This will initially allow for a reduction in the expenses generated by emergency orders placed during a stock shortage. Indeed, it will be possible for the establishment to order a larger quantity of the product from the supplier or to place an order with another supplier.
The software Antidote (www.antidote.info) is capable of correcting English and French corpuses. It detects thousands of types of errors. We want to add new types of correction using modules based on deep learning. The detection of missing words is a type of correction that we want to address in this internship. Here is an example of a sentence where Antidote displays a break (analysis problem), but which a deep learning model could correct:
If the project performs well, the intern will be able to try an integration of his model in our Antidote software.
The first objective is to use ML to reduce the modeling error in predicting the end-of-growth of a batch, reducing the emission of CO2 and water consumption of synthesized products. The second objective is to formulate the algorithms to facilitate its integration into our analytics solution. The third objective is to validate shared learning when applied for 1) forecasting other events and 2) forecasting the same events using similar but different datasets from different users.
Deep learning technology is a great tool to learn complex patterns and make prediction based on this learning. In order to get the most accurate predictions, one needs to train those neural networks on vast amount of labelled data. Labelling data is a time consuming and costly task. Using semi supervised learning, it should be possible to label a fraction of the dataset and let the neural network learn by itself on the rest of the, unlabelled, data, thus greatly reducing the overhead of using deep learning technology.
Brain MRI scans are a critical component in the diagnosis of neurodegenerative disorders and their use will only increase in the following years. However, there is a wide diversity in terms of the image quality and resolution obtained across different sites and there is a need for robust methods that can handle such diversity. The goal of this project is to develop and validate the performance of state-of-the-art lesion detection methods for 3D brain MRIs.