My research project focuses on teaching AI to think logically. Current methods based on neural networks just produce an output based on an input. Knowledge emerges from patterns of neural activations, but there is no concrete representation of concepts. I am working on building neural networks that output symbols and relations, with which the AI framework can do logical reasoning. For example, if the input is a picture of an equation, the neural network produces the numbers and symbols in the equation. The reasoning layer then conducts the required mathematical calculation. To facilitate the learning of such symbols, I am investigating human-in-the-loop methods. In cases where the framework is unsure, it can ask humans to verify the symbols.
I chose to pursue a PhD with the STAI CDT to get a broader experience than with a standard PhD. I wanted to learn about multiple fields in AI and engage in projects and research with other people. It is much more fun to be in a cohort and enables you to see different perspectives.
Undergraduate Qualification: BSc (Hons) Artificial Intelligence and Computer Science, University of Edinburgh
Masters Qualification: MSc Computing (Artificial Intelligence and Machine Learning), Imperial College London
Conferences: Rader, A. P., Mocanu, I. G., Belle, V., & Juba, B. (2021). Learning Implicitly with Noisy Data in Linear Arithmetic. In Proceedings of 30th International Joint Conference on Artificial Intelligence (IJCAI-21) (pp. 1410-1417). IJCAI Inc. https://doi.org/10.24963/ijcai.2021/195