Environmental Adaptation and Differential Replication in Machine Learning

Abstract

When deployed in the wild, machine learning models are usually confronted with an environment that imposes severe constraints. As this environment evolves, so do these constraints. As a result, the feasible set of solutions for the considered need is prone to change in time. We refer to this problem as that of environmental adaptation. In this paper, we formalize environmental adaptation and discuss how it differs from other problems in the literature. We propose solutions based on differential replication, a technique where the knowledge acquired by the deployed models is reused in specific ways to train more suitable future generations. We discuss different mechanisms to implement differential replications in practice, depending on the considered level of knowledge. Finally, we present seven examples where the problem of environmental adaptation can be solved through differential replication in real-life applications.

Keywords: copying; differential replication; editing; knowledge distillation; machine learning; natural selection.

Figure 1

The problems of (a) transfer learning and environmental adaptation for (b) a case where the new new feasible set overlaps with part of the existing hypothesis space and (c) a case where there is no such overlap. The gray and red lines and dots correspond to the set of possible solutions and the obtained optimum for the source and target domains, respectively. The shaded area shows the defined hypothesis space.

Figure 2

Inheritance mechanisms in terms of their knowledge of the data and the model internals.