A framework for measuring, tracking and reduction of carbon emissions from large deep learning models - BC-934

Project type: Innovation
Desired discipline(s): Engineering - computer / electrical, Engineering, Computer science, Mathematical Sciences, Mathematics
Company: ALSGeoAnalytics
Project Length: 4 to 6 months
Preferred start date: As soon as possible.
Language requirement: English
Location(s): Vancouver, BC, Canada
No. of positions: 1
Desired education level: Undergraduate/BachelorMaster's
Open to applicants registered at an institution outside of Canada: No

About the company: 

ALS’ Geoanalytics services integrate data science, cutting-edge AI technology and geoscience expertise to help you eliminate manual processes and transform your data into valuable insights. Our comprehensive services empower mining and exploration managers and technical experts with real-time decision-making for improved output and faster time-to-market.

Describe the project.: 

This project aims to develop an innovative framework to measure, track, and reduce the carbon footprint of large deep learning models. As the adoption of artificial intelligence (AI) grows, the energy consumption and associated emissions from training and deploying large-scale models have become a significant environmental concern. This research and innovation project addresses the pressing need for sustainable AI practices by delivering tools and methodologies to quantify and minimize the environmental impact of machine learning (ML).

The project’s ultimate goal is to deliver a comprehensive methodology and software tool that allows organizations to quantify and minimize the carbon emissions of their AI workflows.

The project’s deliverables include:
• Carbon Emission Tracking Framework: A robust system to monitor and measure the energy consumption and carbon emissions associated with ML workflows.
• Optimization Strategies: A suite of energy-efficient algorithmic enhancements and workflows for reducing emissions during training and deployment.
• Actionable Insights and Software Tools: An intuitive software platform that not only tracks emissions but provides tailored recommendations for carbon reduction.
• Validation and Benchmarks: A comprehensive evaluation of the framework’s effectiveness against real-world AI models and industry-standard benchmarks.

The innovation lies in the development of a multi-faceted approach combining data-driven models with real-world energy metrics to provide actionable and scalable solutions. Tasks for the research candidate include developing carbon footprint tracking systems, optimizing training pipelines for energy efficiency, and validating the framework against state-of-the-art benchmarks

This project bridges the gap between technological innovation and environmental responsibility and offers practical, actionable solutions for industry-wide adoption, making it a valuable resource for achieving a more energy efficient future.

Required expertise/skills: 

• Knowledge on Machine Learning (ML), Deep Learning (DL).
• Proficieny in programming in Python language, and expertise in working with Python ML frameworks such as scikit-learn and PyTorch.
• Knowledge of Restful APIs.
• Knowledge of Databases is an asset.
• Knowledge of React is an asset.