[Unsolved problem in auditory localization]
Auditory localization is perceived by comparing the signal input between two ears through integrating the information on the interaural time difference (ITD) and the interaural level difference for each frequency range. The dominant models for processing ITDs are based on the Jeffress model (1948) and predict neurons that fire maximally at a common ITD across their responsive frequency range. In Jeffress-type models, sensitivity to ITDs would require neural delay lines with differences in path lengths between the two ears. However, there is no evidence to prove or disprove such models physiologically.

[Statistical modeling approach for auditory localization]
Kunio Tanabe (ISM), Toshio Irino (Wakayama Univ.) and I are facing the above problem in the Jeffress-type models. Our objective is to anew propose an alternative meta-model by statistical modeling approach. We attempt to find possible constraints on neural networks through statistically modeling of ITDs and establish a verifiable paradigm for auditory localization. In practice, signal inputs to both ears from various directions are analyzed using inner ear models and the neural firing patterns are extracted. From the patterns, the ﾒoptimalﾓ statistical model for auditory localization is estimated using the Penalized Logistic Regression Machine (PLRM) which was developed by Tanabe (2001). The inductive PLRM, which is different from the conventional machine with parametric models designed by human, can automatically select a proper model only based on data.
Future work includes post-analyses of the model and identification of neural networks required for auditory localization.

Members
Tomoko Matsui (ISM)