| KEYWORD |
Area Engineering
AutoML techniques for neuro-symbolic AI
Reference persons LIA MORRA
Research Groups DAUIN - GR-09 - GRAphics and INtelligent Systems - GRAINS
Thesis type ARTIFICIAL INTELLIGENCE, AUTO-ML, DEEP LEARNING, MACHINE LEARNING
Description Artificial Intelligence agents are required to learn from their surroundings and to reason about the knowledge that has been learned in order to make decisions. While state-of-the-art deep learning typically uses distributed representations, reasoning is normally useful at a higher level of abstraction. As a result, attempts at combining symbolic AI and neural computation into neural-symbolic systems have been on the increase. Neural-symbolic learning and reasoning is an area of research that aims at integrating relational/symbolic reasoning and knowledge representation (e.g., first order logic language for knowledge representation) by embedding their constructs in deep neural networks. The goal of this project is to automate the process of designing, implementing and setting the hyper-parameters of neuro-symbolic systems. In fact, NeSy techniques like LTNs introduce many additional design choices and hyper-parameters, such as: i) how to define constraints, which to include, in which order and how to weight them; ii) how to redefine the neuro-symbolic loss, which functions to use and how to parameterize them; iii) how to deal with common dataset issues such as class imbalance and dataset bias; iv) how to prevent NeSy losses to promote unwanted behaviors, such as reasoning shortcuts.
Several thesis are available on tackling these issues. Depending on the candidate background and predisposition, the problem can be tackled either from a more theoretical or experimental standpoint. Starting from an analysis of the current literature, techniques to be investigated include: i) Analysis of the mathematical and numerical properties of different frameworks; ii) Comparison with other frameworks and losses; iii) application of techniques from the fields of hyper-parameter optimization (Random search, Bayesian Optimization), Neural Architectural Search (NAS), genetic algorithms and curriculum learning. Experiments will be conducted on standard benchmarks as well as applications on semantic image interpretation explored by the group.
Prerequisites: programming skills (Python, Pytorch or other deep learning framework); good analytical skills. Prior knowledge of neuro-symbolic techniques is not required – essential material to study on the topic will be provided. The candidate we are looking for is highly motivated and interested towards research-oriented activities.
Deadline 28/02/2025
PROPONI LA TUA CANDIDATURA