Towards efficient feature sharing in MIMO architectures
This work addresses a specific bottleneck in MIMO architectures for improving efficiency on mobile and AR/VR devices, but it is incremental as it builds on existing frameworks.
The paper tackled the problem of inefficient parameter use in Multi-Input Multi-Output (MIMO) architectures, where subnetworks fail to share generic features, limiting applicability on small devices; the proposed unmixing step improved feature sharing and model performance in preliminary experiments on CIFAR-100.
Multi-input multi-output architectures propose to train multiple subnetworks within one base network and then average the subnetwork predictions to benefit from ensembling for free. Despite some relative success, these architectures are wasteful in their use of parameters. Indeed, we highlight in this paper that the learned subnetwork fail to share even generic features which limits their applicability on smaller mobile and AR/VR devices. We posit this behavior stems from an ill-posed part of the multi-input multi-output framework. To solve this issue, we propose a novel unmixing step in MIMO architectures that allows subnetworks to properly share features. Preliminary experiments on CIFAR-100 show our adjustments allow feature sharing and improve model performance for small architectures.