The tilt aftereffect (TAE, Gibson and Radner, 1937) is a simple but intriguing visual phenomenon. After staring at a pattern of tilted lines or gratings, subsequent lines appear to have a slight tilt in the opposite direction (figure 1). The effect resembles an afterimage from staring at a bright light, but it represents changes in orientation perception rather than in color or brightness.
 |
Most modern explanations of the TAE are loosely based on the feature-detector model of the primary visual cortex (V1), which characterizes this area as a set of orientation-detecting neurons. Experiments showed that these neurons became more difficult to excite during repeated presentation of oriented visual stimuli, and the desensitization persisted for some time (Hubel and Wiesel, 1968). This observation led to the fatigue theory of the TAE: perhaps active neurons become fatigued due to repeated firing, causing the response to a test figure to change during adaptation. Assuming the perceived orientation is some sort of average over the orientation preferences of the activated neurons, the final perceived orientation would thus show the direct TAE (Coltheart, 1971). The fatigue theory has been discredited for a number of reasons, chief among which is that individual V1 neurons actually do not appear to fatigue (Finlayson and Cynader, 1995; McLean and Palmer, 1996). In fact, their response to direct stimulation is essentially unchanged by adaptation to a visual stimulus (Vidyasagar, 1990). The now-popular inhibition theory postulates that tilt aftereffects instead result from changing inhibition between orientation-detecting neurons (Tolhurst and Thompson, 1975).
The inhibition hypothesis has recently been incorporated into theories of the larger purpose and function of the cortex. Barlow (1990) and Földiák (1990) have proposed that the early cortical regions are acting to reduce the amount of redundant information present in the visual input. They suggest that aftereffects are not flaws in an otherwise well-designed system, but an unavoidable result of a self-organizing process that aims at producing an efficient, sparse encoding of the input through decorrelation (see also Field, 1994; Miikkulainen et al., 1997; Sirosh et al., 1996). Based on these theories, Dong (1995) has shown analytically that perfect decorrelation can result in direct tilt aftereffects which are similar to those found in humans. Only very recently, however, has it become computationally feasible to test the inhibition/decorrelation theory of the TAE in a detailed model of cortical function, with limitations similar to those known to be present in the cortex.
A Hebbian self-organizing process (the Receptive-Field Laterally Interconnected Synergetically Self-Organizing Map, or RF-LISSOM; Miikkulainen, Bednar, Choe, and Sirosh, 1997; Sirosh and Miikkulainen, 1994a, 1997; Sirosh, 1995) has been shown to develop feature detectors and specific lateral connections that could produce such aftereffects. The RF-LISSOM model gives rise to anatomical and functional characteristics of the cortex such as topographic maps, ocular dominance, orientation, and size preference columns, and the patterned lateral connections between them. Although other models exist that explain how the feature-detectors and afferent connections could develop by input-driven self-organization, RF-LISSOM is the first model that also shows how the long-range lateral connections self-organize as an integral part of the process. The laterally connected model has also been shown to account for many of the dynamic aspects of the adult visual cortex, such as reorganization following retinal and cortical lesions (Miikkulainen et al. 1997; Sirosh and Miikkulainen 1994b; Sirosh, Miikkulainen, and Bednar 1996). These findings suggest that the same self-organization processes that drive development may be acting in the adult (see Fregnac 1996; Gilbert 1998 for review). This hypothesis is further explored in this paper with respect to the TAE.
The current work is a first study of the functional behavior of the RF-LISSOM model, specifically the response to stimuli similar to those known to cause the TAE in humans. The model permits simultaneous observation of activation and connection patterns between large numbers of neurons. This makes it possible to relate higher-level phenomena to low-level events, which is difficult to do experimentally. The results provide detailed computational support for the idea that tilt aftereffects result from a general self-organizing process that aims at producing an efficient, sparse encoding of the input through decorrelation.