Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Research Topic for the ParisTech/CSC PhD Program
*Field (cf. List of fields below): Information and Communication Sciences and Technologies Subfield (Applied Physics, Chemistry, Mathematics, Mech. Eng. etc): Computer Engineering,Robotics, Artificial Intelligence.Title: Explainable neural-symbolic learningParisTech School: ENSTA ParisTechAdvisor(s) Name: Ass. Prof. Natalia Diaz RodriguezAdvisor(s) Email: [email protected](Lab, website): asr.ensta-paristech.fr https://flowers.inria.fr/
Short description of possible research topics for a PhD: Current Deep Learning methods are powerful but datahungry, sample inefficient and hard to interpret.On the other hand, classic knowledge based approaches such as logics, or rule based systems areendowed with less generalization capabilities. However, the latter are fully explainable. The objectiveof this project is to combine the best of both paradigms by using recent advances on relational deeplearning and symbolic reinforcement learning to make black box neural models more explainable. Forthis purpose, specific datasets will be created in order to test abilities for a neural network to learnspatial and relational ontological properties that are inherent to description logics. Data, tasks andmethods will be proposed to ameliorate the problem of standard memory architectures struggling attasks that heavily involve understanding the way in which entities are connected, estimating causeeffect relations, or conditional independences. The learning of finer grained semantic levels ofdescription logics will be tested for explainability in computer vision reallife applications in order toproduce interpretable rules or hypotheses to allow 1) more trusted AI models, and 2) models closer tohuman commonsense reasoning.
Required background of the student (Which should be the main field of study of the applicant before applying): Computer Science/Engineering, Programming, Deep learning and Reinforcement LearningA list of 5(max.) representative publications of the group: (Related to the research topic)[1] T Lesort, N DíazRodríguez, D Goudou, JeanFrançois, Filliat. SRL Toolbox: Environments, Datasets and Evaluation Metrics for State Representation Learning. 2018.[2] Raffin, A Hill, R Traoré, T Lesort, N DíazRodríguez, D Filliat State Representation Learning for Control: An Overview. 2018. arXiv:1809.09369[3 S Doncieux, D Filliat, N DiazRodriguez, T Hospedales, R Duro, A Coninx, … Openended Learning: a Conceptual Framework based on Representational Redescription. 2018. Frontiers in Neurorobotics 12, 59[4] M Garnelo, K Arulkumaran, M Shanahan. Towards deep symbolic reinforcement learning Deep Reinforcement Learning . NIPS Workshop 2016.[5] Evans, R. and Grefenstette, E. Learning explanatory rules from noisy data. 2018. Journal of Artificial Intelligence Research , 61:1–64.