Lottery Tickets in Evolutionary Optimization: On Sparse Backpropagation-Free Trainability
This work addresses the generalization of sparse trainability to evolutionary methods, offering insights into optimization dynamics for researchers in machine learning and evolutionary algorithms.
The paper investigated whether the lottery ticket phenomenon extends to evolutionary optimization, finding that highly sparse trainable initializations exist for evolution strategies and differ from gradient descent-based sparse training, with a novel pruning method enabling sparser initializations and showing transferability across tasks and optimization methods.
Is the lottery ticket phenomenon an idiosyncrasy of gradient-based training or does it generalize to evolutionary optimization? In this paper we establish the existence of highly sparse trainable initializations for evolution strategies (ES) and characterize qualitative differences compared to gradient descent (GD)-based sparse training. We introduce a novel signal-to-noise iterative pruning procedure, which incorporates loss curvature information into the network pruning step. This can enable the discovery of even sparser trainable network initializations when using black-box evolution as compared to GD-based optimization. Furthermore, we find that these initializations encode an inductive bias, which transfers across different ES, related tasks and even to GD-based training. Finally, we compare the local optima resulting from the different optimization paradigms and sparsity levels. In contrast to GD, ES explore diverse and flat local optima and do not preserve linear mode connectivity across sparsity levels and independent runs. The results highlight qualitative differences between evolution and gradient-based learning dynamics, which can be uncovered by the study of iterative pruning procedures.