Fusing model-based and function-based approaches to AI

Abstract

This project is based on a distinction between model-based and function-based approaches to artificial intelligence (AI), with a further distinction between symbolic and probabilistic model-based approaches. The main insight is that the field of AI has focused on different approaches during different phases of its history, and by largely disjoint communities, even though each approach has much to offer yet comes with limitations. The PI???s Automated ReasoningGroup at UCLA has been somewhat unique in having made significant contributions to both symbolic and probabilistic model-based approaches, particularly as far as integrating such approaches. Moreover, it has pioneered work on compiling models into tractable circuits, which can be viewed as a class of functions; in line with the current function based approaches realized by neural networks and deep learning. This project will capitalize on theseaccomplishments and associated expertise for the goal of fusing model-based and function-based approaches, using new insights and innovative research directions. In particular, we plan to capture domain knowledge using partialmodels, that integrate logical constraints, uncertainty, independence and causality. This will allow us to utilize both symbolic and probabilistic model-based approaches for the purpose of capturing the (partial) knowledge we may have about a domain. We then plan to synthesize a function structure from a partial model, while directing the function towards a specific query, as in neural networks. We will finally train the synthesized function structure using labeled and unlabelled data, in a style similar to, but that extends deep learning approaches to combine supervised and unsupervised learning.

Document Details

Document Type
DoD Grant Award
Publication Date
Jul 26, 2018
Source ID
N000141812561

Entities

People

  • Adnan Darwiche

Organizations

  • Office of Naval Research
  • United States Navy
  • University of California, Los Angeles

Tags

Fields of Study

  • Computer science

Readers

  • Adaptive Control and Estimation with Uncertainty in Dynamic Systems.
  • Artificial Intelligence
  • Neural Network Machine Learning.

Technology Areas

  • AI & ML
  • AI & ML - DoD AI Strategy
  • AI & ML - Neural Networks