Skip to main content

Table 8 Average ranks for autoencoder architectures when used as deep embeddings wit one-class SVM (on must-link constraints) and for Isolation Forest (on cannot-link constraints)

From: Overlapping community finding with noisy pairwise constraints

Architecture

Small networks

Large networks

Encoder + SVM

Encoder + IF

Encoder + SVM

Encoder + IF

AE1

3

2

2

3

AE1_L1

4

4

4

4

AE2

2

3

1

2

AE2_L1

5.5

5.5

5.5

5.5

AE3

1

1

3

1

AE3_L1

5.5

5.5

5.5

5.5