MODEL-ORIENTED, DATA-ORIENTED OR HYBRID APPROACH IN MACHINE LEARNING

Authors

          Maksudov Kh.T. – Candidate of Physical and Mathematical Sciences, Associate Professor of the Department of Digital Economics, Polytechnic Institute of Tajik Technical University, Khujand, Republic of Tajikistan, kh.maqsudov@gmail.com.
             Inomov B.B. – senior lecturer, department of digital economy, Polytechnic Institute of Tajik Technical University, Khujand, Republic of Tajikistan, behruzinomov@gmail.com.

Annotation

        The article provides a detailed analysis of the three main approaches in machine learning – model-oriented, data-oriented and hybrid. The advantages and disadvantages of each are examined. It is shown that most modern artificial intelligence systems rely on a model-oriented approach that focuses on improving the architecture and hyperparameters of machine learning models. However, a data-oriented approach that focuses on the quality and life cycle of data can significantly increase the accuracy and reliability of models. The specifics of implementing a data-oriented infrastructure are considered in detail, including understanding the subject area, versioning data, and other aspects. It is noted that the transition to a data-oriented approach has many advantages. It is concluded that in the future it is advisable to increasingly rely on a data-oriented approach in machine learning, since data quality is critically important at all stages of the life cycle of artificial intelligence systems

Key words

artificial intelligence, data-oriented method, model-oriented method, code,
model quality