| Grant number: | 21/06981-0 |
| Support Opportunities: | Regular Research Grants |
| Start date: | July 01, 2022 |
| End date: | June 30, 2024 |
| Field of knowledge: | Physical Sciences and Mathematics - Computer Science - Computer Systems |
| Agreement: | MCTI/MC |
| Principal Investigator: | Marcelo Caggiani Luizelli |
| Grantee: | Marcelo Caggiani Luizelli |
| Host Institution: | Universidade Federal do Pampa (UNIPAMPA). Campus Alegrete. Alegrete , SP, Brazil |
| City of the host institution: | Alegrete |
| Associated researchers: | Arthur Francisco Lorenzon ; Fábio Diniz Rossi ; Oscar Mauricio Caicedo Rendon ; Roberto Irajá Tavares da Costa Filho ; Rodrigo Neves Calheiros ; Weverton Luis da Costa Cordeiro |
| Associated scholarship(s): | 23/00400-0 - An Evaluation Environment for Unsupervised ML Approaches,
BP.TT 22/15340-0 - Design of scalable algorithms for optimization problems related to the orchestration of unsupervised ML approaches, BP.TT |
Abstract
Data plane programmability is redesigning the way we manage and operate forwarding devices. However, most of the algorithmic decisions performed by data planes are still deterministic and control-plane dependent. We believe that it is possible to break this dependency and make the data plane intelligent so that they learn infrastructure states autonomously. In this project, we propose Spinner, the first effort towards to operationalize unsupervised Machine Learning approaches (Machine Learning - ML) in programmable devices. Despite existing efforts to make data planes intelligent, little has been done to design unsupervised ML algorithms that fit the architectural constraints of programmable devices. Unsupervised learning algorithms (e.g., data clustering) are useful when the data profile is not known in advance and the learning process takes place continuously. Executing such approaches in the data plane has the potential to reduce the volume of data collected/transmitted to ML Control Plane applications, as well as reducing the overall decision-making time. Despite the potential for executing ML techniques on the data plane, designing and operating them on programmable devices is particularly challenging for three reasons: (i) programmable architectures are constrained concerning arithmetic operations; (ii) domain-specific languages (e.g. P4) and current architectures do not provide iteration-based structures and, therefore, the implementation of classical ML algorithms is infeasible; and (iii) the available memory resources are limited, which reduces the amount of information stored and processed by a device. To fill in these gaps, Spinner aims to simplify the process of implementing and operating unsupervised ML algorithms in programmable devices. The idea is to design models and algorithms that can fit architectural constraints to ensure the accuracy of ML models. Later, we intend to operate them in a programmable infrastructure. The results that can be accomplished in this project have the potential to make a strong impact in this research area in the coming years. (AU)
| Articles published in Agência FAPESP Newsletter about the research grant: |
| More itemsLess items |
| TITULO |
| Articles published in other media outlets ( ): |
| More itemsLess items |
| VEICULO: TITULO (DATA) |
| VEICULO: TITULO (DATA) |