In well-trained DNN+HMM face-recognition models, an eyes-focused eye movement pattern yields higher recognition accuracy than a nose-focused pattern¶

Claim · All claims
Evidence · 1 study · design r2 · q2 · i2 medium
Currency · newest study 2022 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i2 Well-trained DNN+HMM models with the eyes-focused pattern showed higher face recognition accuracy than nose-focused models (effect size d = 0.75). → Hsiao 2022

Evidence¶

Hsiao 2022¶

Hsiao, J. H., An, J., Hui, V. K. S., Zheng, Y. & Chan, A. B. (2022). Understanding the role of eye movement consistency in face recognition and autism through integrating deep neural networks and hidden Markov models. npj Science of Learning 7:28. https://doi.org/10.1038/s41539-022-00139-6

q2 · i2 · design · r2

Simulation of 80 trained DNN+HMM models on the LFW-a face dataset; representative eye movement patterns were derived by EMHMM clustering. Well-trained (epoch 500) eyes-focused models outperformed nose-focused ones, with printed effect size d = 0.75 and "MA = 0.534, MB = 0.500".

"In the well-trained models, those with pattern A (eye-focused) exhibited higher accuracy than pattern B (nose-focused), t(78) = 3.33, p = 0.001, d = 0.75, MA = 0.534, MB = 0.500"

Discussion¶

Related Claims¶

• In partially trained DNN+HMM models simulating children, eye movement consistency (lower entropy) predicts recognition accuracy but eye movement pattern does not — related
• In typically developing children aged 6–11, face recognition performance is predicted by eye movement consistency but not by eye movement pattern — related
• A more eyes-focused scanning pattern and higher eye movement consistency are each correlated with better face recognition performance — related
• In adults, face recognition performance is predicted by eye movement pattern, with eye movement consistency adding no additional variance — related