Specificities of learning in combined-feature search: implications for underlying coding mechanisms

Eleven subjects were trained with 1200 trials of search for a 180°-rotated letter L amongst 90°-counterclockwise-rotated Ls, then tested with three transfer stimuli in which target, distractors, or both were changed. According to signal detection theory, learning both increases signal enhancement and external noise exclusion by enlarging the distance between target and distractor response distributions (Dosher and Lu, 1998 Proceedings of the National Academy of the USA 95 13988 - 13993). Similarly, the SERR (search via recursive rejection ) model (Humphreys and Muller, 1993 Cognitive Psychology 25 43 - 110) would assume a learning process based upon boosting of target template and distractor grouping and rejection. Finally, the FRS (Fogel, Rubenstein, and Sagi) model of texture segmentation [Sagi, 1991, in Channels in the Visual Nervous System: Neurophysiology, Psychophysics and Models Ed. B Blum (London and Tel Aviv: Freund) pp 397 - 424] would assume a learning process based on the tuning of the appropriate filters to the orientation of target and/or distractor elements at the spatial frequency corresponding to the size of whole elements. Our results show that learning transfers to a stimulus where low-pass-filtered orientation of the target and distractor is the same (changing either target or distractors); learning does not transfer to a stimulus where both target and distractors are changed; learning abolishes the differences between present and absent responses. Learning may be based upon either additive internal noise reduction and external noise exclusion (Dosher and Lu, 1998), or target template enhancement and distractor grouping and rejection (Humphreys and Muller, 1993).