Latent Weights Do Not Exist: Rethinking Binarized Neural Network Optimization
This work addresses optimization challenges for BNNs, which are important for efficient deep learning in resource-constrained environments, representing a novel approach rather than an incremental improvement.
The paper tackles the problem of optimizing Binarized Neural Networks (BNNs) by redefining latent weights as inertia rather than real-valued analogs, and introduces the Binary Optimizer (Bop) as the first optimizer specifically designed for BNNs, demonstrating its performance on CIFAR-10 and ImageNet.
Optimization of Binarized Neural Networks (BNNs) currently relies on real-valued latent weights to accumulate small update steps. In this paper, we argue that these latent weights cannot be treated analogously to weights in real-valued networks. Instead their main role is to provide inertia during training. We interpret current methods in terms of inertia and provide novel insights into the optimization of BNNs. We subsequently introduce the first optimizer specifically designed for BNNs, Binary Optimizer (Bop), and demonstrate its performance on CIFAR-10 and ImageNet. Together, the redefinition of latent weights as inertia and the introduction of Bop enable a better understanding of BNN optimization and open up the way for further improvements in training methodologies for BNNs. Code is available at: https://github.com/plumerai/rethinking-bnn-optimization