Know Where To Drop Your Weights: Towards Faster Uncertainty Estimation

TL;DR

This paper introduces Select-DC, a method that reduces computational costs of uncertainty estimation in neural networks by applying DropConnect to selected layers, maintaining accuracy while enabling faster, low-latency uncertainty estimation.

Contribution

The paper proposes Select-DC, a novel approach that models epistemic uncertainty efficiently by using a subset of layers with DropConnect, reducing GFLOPS needed for uncertainty estimation.

Findings

Significant reduction in GFLOPS compared to Monte Carlo DropConnect

Marginal performance trade-off with faster uncertainty estimation

Insights into how layer-wise DropConnect affects predictive entropy

Abstract

Estimating epistemic uncertainty of models used in low-latency applications and Out-Of-Distribution samples detection is a challenge due to the computationally demanding nature of uncertainty estimation techniques. Estimating model uncertainty using approximation techniques like Monte Carlo Dropout (MCD), DropConnect (MCDC) requires a large number of forward passes through the network, rendering them inapt for low-latency applications. We propose Select-DC which uses a subset of layers in a neural network to model epistemic uncertainty with MCDC. Through our experiments, we show a significant reduction in the GFLOPS required to model uncertainty, compared to Monte Carlo DropConnect, with marginal trade-off in performance. We perform a suite of experiments on CIFAR 10, CIFAR 100, and SVHN datasets with ResNet and VGG models. We further show how applying DropConnect to various layers in the network with different drop probabilities affects the networks performance and the entropy of the predictive distribution.