Uncertainty estimation for time series classification - Exploring predictive uncertainty in transformer-based models for variable stars

Open Access

Table A.1

Comparison of UEs in the misclassification task on the MACHO, ATLAS, and OGLE-III test sets for stochastic attention and the DEs method.

Method	UEs	MACHO	ATLAS	OGLE-III
Baseline	SMP	75.6±1.6	85.9±2.1	82.2±1.1
	PV	71.4±2.3	82.0±2.2	73.4±1.7
	BALD	70.0±2.6	80.8±2.4	74.1±1.8

Gumbel-softmax SA	SMP	-2.4±1.9	-2.1±2.3	-2.0±1.1
	PV	-5.5±2.2	-2.6±2.6	-0.4±1.9
	BALD	-6.9±2.3	-4.7±2.9	-4.0±2.5

Notes. The baseline values represent the mean and standard deviation of the absolute ROC AUC scores (%). For the Stochastic Attention method, values indicate the difference in ROC AUC relative to the baseline.

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.