|
|
Assessment of Uncertainty of Neural Net Predictions in the Tasks of Classification, Detection and Segmentation
Vlasák, Jiří ; Kohút, Jan (referee) ; Herout, Adam (advisor)
This work focuses on comparing three widely used methods for improving uncertainty estimations: Deep Ensembles, Monte Carlo Dropout, and Temperature Scaling. These methods are applied to six computer vision models that are pretrained as well as trained from scratch. The models are then evaluated on computer vision datasets for classification, semantic segmentation, and object detection using a wide range of metrics. The models are also evaluated on distorted versions of these datasets to measure their performance on out-of-distribution data. These modified models achieve promising results. Ensembles outperform the other models by as high as 70 % in accuracy and 0.2 in IOU on the distorted MedSeg COVID-19 segmentation dataset while also outperforming the other models on the CIFAR-100 and FMNIST datasets.
|
|
|
Assessment of Uncertainty of Neural Net Predictions in the Tasks of Classification, Detection and Segmentation
Vlasák, Jiří ; Kohút, Jan (referee) ; Herout, Adam (advisor)
This work focuses on comparing three widely used methods for improving uncertainty estimations: Deep Ensembles, Monte Carlo Dropout, and Temperature Scaling. These methods are applied to six computer vision models that are pretrained as well as trained from scratch. The models are then evaluated on computer vision datasets for classification, semantic segmentation, and object detection using a wide range of metrics. The models are also evaluated on distorted versions of these datasets to measure their performance on out-of-distribution data. These modified models achieve promising results. Ensembles outperform the other models by as high as 70 % in accuracy and 0.2 in IOU on the distorted MedSeg COVID-19 segmentation dataset while also outperforming the other models on the CIFAR-100 and FMNIST datasets.
|
|
|
DEnFi: Deep Ensemble Filter for Active Learning
Ulrych, Lukáš ; Šmídl, Václav
Deep Ensembles proved to be a one of the most accurate representation of uncertainty for deep neural networks. Their accuracy is beneficial in the task of active learning where unknown samples are selected for labeling based on the uncertainty of their prediction. Underestimation of the predictive uncertainty leads to poor exploration of the method. The main issue of deep ensembles is their computational cost since multiple complex networks have to be computed in parallel. In this paper, we propose to address this issue by taking advantage of the recursive nature of active learning. Specifically, we propose several methods how to generate initial values of an ensemble based of the previous ensemble. We provide comparison of the proposed strategies with existing methods on benchmark problems from Bayesian optimization and active classification. Practical benefits of the approach is demonstrated on example of learning ID of an IoT device from structured data using deep-set based networks.
|
guest ::
