*********************************
There is now a CONTENT FREEZE for Mercury while we switch to a new platform. It began on Friday, March 10 at 6pm and will end on Wednesday, March 15 at noon. No new content can be created during this time, but all material in the system as of the beginning of the freeze will be migrated to the new platform, including users and groups. Functionally the new site is identical to the old one. webteam@gatech.edu
*********************************
Title: Differentiable Neural Logic Networks and their Application onto Inductive Logic Programming
Committee:
Dr. Fekri, Advisor
Dr. Kiyavash, Chair
Dr. Davenport
Abstract:
The objective of this research is to present a novel paradigm for learning algorithmic and discrete tasks via Deep Neural Networks (DNN) by using Boolean logic algebra. Despite the impressive performance of DNNs in a variety of tasks, there has been limited success in using DNNs for learning discrete algorithmic problems. To achieve this objective, I first present the basic differentiable operators of a Boolean system such as conjunction, disjunction and exclusive-OR and I show how these elementary operators can be combined in a simple and meaningful way to form differentiable Neural Logic (dNL) Networks. I examine the effectiveness of the proposed dNL framework in learning Boolean functions and discrete-algorithmic tasks. Further, I demonstrate that, in contrast to the implicit learning in the MLP approach, the proposed neural logic networks can learn the logical functions explicitly allowing the verification and interpretation by human. In this thesis proposal I will also show how this explicit representational feature of dNL can be exploited to reformulate the Inductive Logic Programing into a differentiable optimization problem which can be solved via typical gradient methods. This novel neural based ILP solver is capable of learning auxiliary and invented predicates as well as learning complex recursive predicates and can outperform the current state of the art ILP solvers in a variety of benchmark algorithmic tasks as well as larger scale relational classification tasks.
