OpenCBD: A Network-Encrypted Unknown Traffic Identification Scheme Based on Open-Set Recognition

Table 1

Different models for open-set recognition.

Model	Reference	Open-set recognition category	Methodology	Extreme value theory	Advantages
OpenMax	[15]	Discriminative model	EVT-based calibration classification		First deep open-set classifier without using background samples
G-OpenMax	[16]	Generative model	Unknown generation classification		Combining generative adversarial networks and OpenMax
CROSR	[17]	Discriminative model	Distance		First neural network architecture which involved hierarchical reconstruction blocks
C2AE	[18]	Discriminative model	Reconstruction error		Algorithms using class conditional autoencoders
Neural-network-based representation	[19]	Discriminative model	EVT-free calibration classification		A loss function was proposed such that instances from the same class are close to each other, while instances from different classes are farther apart
PEELER	[20]	Discriminative model	Distance		Combining few-shot classification and open-set recognition
ORE	[21]	Discriminative model	Distance		An incremental object detector is proposed
CD-OSR	[22]	Discriminative model	EVT-free calibration classification		Automatically reserve space for unknown classes under test, naturally bringing new class discovery capabilities