Optimize the size and performance of AI models for low capacity devices
a. Project Description
Machine Learning and Artificial Intelligence in general slowly become an integral part in the majority of software systems and applications. In combination with the advancements in pervasive and ubiquitous computing, it is now implied that AI models need to be deployable on any device. However, it is not clear yet what the implications are of low capacity devices, such as mobile phones, on the deployment, performance and accuracy of AI models. In this project, we aim to firstly explore these implications and quantify them, in order to guide the optimization of AI models in terms of resource consumption and constrained by requirements on their performance and accuracy. Next, we will integrate these considerations in MLOps pipelines to automate the optimization and make it an integral part of the model’s lifecycle. The expected outcome of the project is the development of models to simultaneously capture the aspect of the AI models as software artifacts (performance and resource consumption) and as AI artifacts (accuracy). Based on these models, we will also develop tools for the optimal and adaptive deployment of AI models on low capacity devices.
b. Tasks and responsibilities
The hired student will work towards the review of MLOps technologies and tools with a focus on the deployment of AI models on IoT and other low capacity devices, as well as a review of the state-of-practice on adaptive AI model deployment and operations. The student will develop the theoretical foundation as well as practical experience on the optimization of MLOps pipelines. The student will aim to publish in top-tier journals, including IEEE Transactions on Software Engineering, Elsevier Journal of Systems and Software, and conferences, such as ICPE, SEAMS, ACSOS, ICSE, and others. The student will also be responsible for supervising and mentoring MSc and BSc students working on the project. The position is open for Winter, Summer or Fall 2024.
c. Required Skills
The student will be asked to demonstrate adequate understanding or expertise in the following topics through relevant courses (on undergraduate or graduate level) or through relevant publications in international conferences or journals. The student should consider applying if they have the expert-level skills and at least 50% of the good-level skills.
- Expert programming skills, preferably in Java or Python.
- Good knowledge on software performance engineering, self-adaptive systems and DevOps.
- Adequate knowledge on AI, machine learning, deep learning and relevant technologies.
d. Application process
Upon contacting the professor to inquire for the position, the student is also asked to submit the following documents:
- A copy of the most recent version of their CV or Resume.
- A copy of the transcripts of their undergraduate and/or master studies, if available.
- The aforementioned documents are also required by the EECS application process for the PhD or MSc programs (along with a statement of purpose). The candidate student is highly encouraged to complete the EECS application in parallel to contacting the professor. More information about the EECS application can be found here: https://lassonde.yorku.ca/eecs/academics/graduate/future-students/#phd.
- The names and contact information of 3 referees.
-
A review for one of the three following articles. The review (maximum one page) should contain a summary of the paper, its strengths and weaknesses and comments about the improvement or extension of the work presented in the paper.
- Raj, E., Buffoni, D., Westerlund, M. and Ahola, K., 2021, October. Edge mlops: An automation framework for aiot applications. In 2021 IEEE International Conference on Cloud Engineering (IC2E) (pp. 191-200). IEEE.
- Peltonen, E. and Dias, S., 2023, October. LinkEdge: Open-sourced MLOps Integration with IoT Edge. In Proceedings of the 3rd Eclipse Security, AI, Architecture and Modelling Conference on Cloud to Edge Continuum (pp. 67-76).
- Antonini, M., Pincheira, M., Vecchio, M. and Antonelli, F., 2022, May. Tiny-MLOps: A framework for orchestrating ML applications at the far edge of IoT systems. In 2022 IEEE international conference on evolving and adaptive intelligent systems (EAIS) (pp. 1-8). IEEE.
- Fujii, T.Y., Hayashi, V.T., Arakaki, R., Ruggiero, W.V., Bulla Jr, R., Hayashi, F.H. and Khalil, K.A., 2021. A digital twin architecture model applied with MLOps techniques to improve short-term energy consumption prediction. Machines, 10(1), p.23.
- (Only for PhD candidates) An example of a proposal (as evidence of writing) written by the student for a research project relevant to the position or of a topic selected by the student. The proposal should include background, motivation, methodology and a plan for evaluation. The proposal should be maximum 2 pages.
- The candidate student should submit these documents by email to the professor with the subject “MLOps PhD 2024” for PhD candidates or “MLOps MSc 2024” for MSc candidates). No email will be considered unless it has this subject and the required attachments (CV, transcripts, review, proposal). In the email, the student should express their interest to the position and provide the corresponding evidence to the required skills as this appears in the attached documents.