Face-voice association towards multimodal-based authentication using modulated spike-time dependent learning

• Main Authors: Yusoff, Nooraini, Ibrahim, Mohammed Fadhil
• Format: Conference or Workshop Item
• Language: English
• Published: 2015
• Subjects: QA75 Electronic computers. Computer science
• Online Access: http://repo.uum.edu.my/15518/1/PID161.pdf; http://repo.uum.edu.my/15518/; http://www.icoci.cms.net.my/proceedings/2015/TOC.html

We propose a reward based learning to associate face and voice stimuli. In particular, we implement learning in a spiking neural network paradigm using modulated spike-time dependent plasticity (STDP).The face and voice stimuli are paired with a temporal delay, and the network is trained to associate the paired face-voice with a target response.The learning rule is dependent on a reward policy in which the network is given a positive reward for a correct response to a face-voice stimulus pair, or the network receives a negative reward for an incorrect response. Despite a stochastic environment, the learning result of real images and sound indicates a good performance with 77.33% accuracy.The result demonstrates that a machine can be trained to associate a pair of biometric inputs to a target response.