Thursday, 23rd November - 12 noon Efthymia Tsamoura : Seminar

Title: On Learning Latent Models with Multi-Instance Weak Supervision.

Abstract:

We consider a weakly supervised learning scenario where the supervision signal is generated by a transition function σ of labels associated with multiple input instances. We formulate this problem as multi-instance Partial Label Learning (multi-instance PLL), which is an extension to the standard PLL problem. Our problem is met in different fields, including latent structural learning and neuro-symbolic integration. Despite the existence of many learning techniques, limited theoretical analysis has been dedicated to this problem. We provide the first theoretical study of multi-instance PLL with possibly an unknown transition σ. We make minimal assumptions on the data distributions. In fact, we prove learnability even under the “toughest" distributions that concentrate their mass on a single instance. In addition, we provide learning guarantees under widely used surrogate losses for training classifiers subject to logical theories. We are the first to provide this theoretical analysis, closing a gap in the neuro-symbolic and latent structural learning literature. This work will be presented in NeurIPS 2023: https://arxiv.org/pdf/2306.13796.pdf.

Bio:

Efi Tsamoura is a Senior Researcher at Samsung AI, Cambridge, UK. In 2016, she was awarded an early career fellowship from the Alan Turing Institute, UK, and before that, she was a Postdoctoral Researcher in the Department of Computer Science of the University of Oxford. Her main research interests lie in the areas of logic, knowledge representation and reasoning, and neuro-symbolic integration. Her research has been published in top-tier AI and database venues (SIGMOD, VLDB, PODS, AAAI, ICML, NeurIPS, etc.). Efi started the Samsung AI neuro-symbolic workshop series “When deep learning meets logic” and has been a keynote speaker in the 2023 Extended Semantic Web Conference.