Anastasiia Grishina

Anastasiia Grishina

Position

Enrichment student

Cohort year

2023

Bio

Anastasiia has been a Ph.D. candidate in computer science at Simula Research Laboratory and the University of Oslo since 2021. In her research, Anastasiia focuses on automated analysis and repair of errors in source code with neural machine translation methods and transformer architectures. The vision of her Ph.D. project is to make coding and debugging less tedious, as well as to explore to which extent the existing generative models can perform the same tasks.

Prior to her Ph.D. project at Simula, Anastasiia worked on natural language processing for industrial applications at Philips Research and the Eindhoven University of Technology. She completed Erasmus Mundus Joint Master's Degree in Pervasive Computing and Communications for Sustainable Development, which spanned five countries and three universities. During her Master's studies, Anastasiia analyzed sustainability aspects of data center operations at ENEA Casaccia in Rome, Italy, and created actionable recommendations for improving the efficiency of energy usage. Anastasiia's initial background is in Applied Mathematics, with Bachelor's and Master's Degrees from Novosibirsk State University. Outside of the academic realm, she enjoys winter sports, hiking, and traveling.

Research interests

During her enrichment period at the Alan Turing Institute from October 2023-March 2024, Anastasiia is planning to broaden her research in source code analysis and repair. She will focus on hybrid methods of automated program repair that combine textual and structural properties of code to repair errors, also referred to as bugs. She will explore how graph representations of code, such as Abstract Syntax Trees, and graph processing methods impact code repair performance in comparison to using neural machine translation techniques without graphs. She will also experiment with the amount of input context added to the buggy code and its effect on code repair performance of generative models.