Deep Learning-based Technology Fitness Landscape: A Biological Analogy

This research note presents a deep learning-based technology fitness landscape premised on a technology embedding space and the estimated improvement rates of all domains in it. The technology embedding space is trained via neural embedding techniques on both intrinsic (semantic) features and connective (citation) information to derive high-dimensional embedding vectors for the 1,757 technology domains curated by Singh et al. (2021), covering 97.2% of the patent database. The estimated improvement rates of these 1,757 domains were also drawn from Singh et al. (2021). The technology fitness landscape exhibits a high hill related to information, electronics, and electrical technologies and a vast low plain of the remaining domains. The construction of the technology fitness landscape based on neural embedding training presents a global picture and bird's eye view of the co-evolution of heterogeneous technology domains in the unified technology space.