ABSTRACT
Due to recent success in solving previously intractable problems such as image recognition and generation, deep learning has exploded in popularity in academia and industry. Because of these successes, deep neural networks appear to be a viable computational model for analyzing and generating the nondeterministic relationships that are found in architectural design. However, the foundations of deep learning are far from new and are based on well-known principles from statistics, information theory, and tensor analysis. In this text, we lay out a broad framework for identifying characteristics of architectural artificial intelligence (AAI) based on domain-specific abilities. Focusing on one of these abilities, we explain foundational principles of deep learning relative to architectural design. We explain the inner workings of neural networks as applied to an industry-relevant task. Finally, we speculate on the necessity, or lack thereof, of domain-specific knowledge encoded into models that would meet the criteria for AAI.
