2026-05-05T16:11:13.748Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:1:8555 2021-11-09T20:55:55Z 1

Experiment-driven system for machine learning based research management 2021 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/1/8555 https://www.eventiotic.com/eventiotic/library/paper/642 I. Radosavljević З. Коњовић Ђ. Обрадовић А. Митровић С. Гаврић М. Видовић Н. Нешић Due to the recent rapid expansion of machine learning and its more frequent use as a valuable tool in scientific research the lack of tools that incorporate research project management and tools for machine learning is becoming apparent. Although there are many tools that can be used for project management or machine learning experiments, most of them are intended for commercial use and usually do not support features necessary for scientific research such as experiment versioning, reviewing and publishing of the experiment results. In this paper the authors propose a software solution that provides a unified environment for running machine learning experiments and research project management. The solution is a microservice oriented web platform that can be used to manage research projects, catalogue datasets, define, validate, and execute machine learning experiments. Some of the advantages of the presented solution are its ability to integrate experiments directly into the research project management process, and to automatically version and validate those experiments with respect to the constraints imposed by the authors or reviewers of the experiments. An additional advantage of the solution is the use of the graphical DSL to define experiments, thus allowing researchers who are not skilled in programming to use the platform. conferenceObject 72 76 ICIST 2021 Proceedings