Current browse context:
cs.LG
Change to browse by:
References & Citations
Computer Science > Machine Learning
Title: Structured Ensembles: an Approach to Reduce the Memory Footprint of Ensemble Methods
(Submitted on 6 May 2021 (v1), last revised 17 Sep 2021 (this version, v2))
Abstract: In this paper, we propose a novel ensembling technique for deep neural networks, which is able to drastically reduce the required memory compared to alternative approaches. In particular, we propose to extract multiple sub-networks from a single, untrained neural network by solving an end-to-end optimization task combining differentiable scaling over the original architecture, with multiple regularization terms favouring the diversity of the ensemble. Since our proposal aims to detect and extract sub-structures, we call it Structured Ensemble. On a large experimental evaluation, we show that our method can achieve higher or comparable accuracy to competing methods while requiring significantly less storage. In addition, we evaluate our ensembles in terms of predictive calibration and uncertainty, showing they compare favourably with the state-of-the-art. Finally, we draw a link with the continual learning literature, and we propose a modification of our framework to handle continuous streams of tasks with a sub-linear memory cost. We compare with a number of alternative strategies to mitigate catastrophic forgetting, highlighting advantages in terms of average accuracy and memory.
Submission history
From: Jary Pomponi [view email][v1] Thu, 6 May 2021 09:56:01 GMT (531kb,D)
[v2] Fri, 17 Sep 2021 08:51:12 GMT (501kb,D)
Link back to: arXiv, form interface, contact.