Expressing Liminality in Data within the Context of Deep Learning

In this paper we describe the concept of Liminality and how it manifests in Feature Spaces of Data during processes of Deep Learning, with particular focus on data during transition, transformation and other states where Liminality manifests in Data Feature Spaces and between various dimensions, Hyper-planes and Decision Boundaries. To go about this, we approached a decision boundary and gave it a threshold, which would behave as an event horizon, that way data can be measured to be in that threshold or out of it. In our work, we discovered that deploying solutions, by consulting a metric of liminality leads to an ability to understand whether a model being used is appropriate, or not, by seeing how its rearrangement, affects the data's proximity to the threshold. In summary, this paper addresses the problem of lack of depth in the study of the evaluation of a Deep Learning algorithm, through the use of a liminality metric which measures which data is subject to certain techniques like feature selection, and pruning, or models that select progressively narrower sections of data from a given feature space.