An investigation of privacy preservation in deep learning-based eye-tracking

Table 3 Performance for people with multiple recordings

	Train	Valid	Test
Total^a	19	14	21
\({{\text{Instance}}^{\text{model}}}^{\text {b}}\)	11	4	11
\({{\text{Person}}^{\text{model}}}^{\text {c}}\)	16	8	12
1 − P-value^Person	0.996	0.21	0.44

^aThe total number of recordings in each set that belongs to a person who has another recording present in the training set of the target model (eye-tracking model)
^bOnly provided for the sake of completeness and is the number of picked recordings as inside (predicted label 1), despite being trained on them with labeling as outside (label 0)
^cThe number of these recordings that had been picked as inside (predicted label 1) in model trained on person labels

ISSN: 1475-925X