Perceptual Learning

Dr. rer. nat. M. Herzog
Prof. Dr. med. M. Fahle

‘Learning’ has been defined as ‘a modification of observable behaviour as a result of preceding experience’, and we define perceptual learning as any change of responses to sensory stimulation following after (extended) training of a perceptual task. More specifically, we concentrate on those parts of the learning process that seem to be relatively independent from conscious, or declarative forms of learning (such as learning a poem) but that rely on rather ‘low-level’ modifications of the central nervous system and that have some resemblance with procedural forms of learning (such as learning to ride a bike). Our results indicate that the notion, generally accepted until a few years ago, that the primary visual cortex in adults lacks plasticity, must be wrong. Perceptual learning differs from other forms of learning in that it involves structural and/or functional changes at least partly in primary sensory cortices, and may account for complex phenomena, including some that are often thought to be cognitive.

Project aims: We investigate the extend to which training can alter performance in perceptual tasks, especially so-called hyperacuity tasks where observers reach thresholds far below the diameter (and spacing) of foveal photoreceptors. This investigation is expected to lead to a better understanding not only of improved perceptual performance through training in normal observers and in patients suffering from disorders of the neuronal parts (and maybe even of the optical parts) of the visual system. It might also lead to insights into the neuronal mechanisms underlying (perceptual) learning, including its dependence on top-down influences from ‘higher’ cortical areas — influences commonly called ‘attention’ and ‘motivation’.

Methods: The methods employed include behavioural and psychophysical experiments on visual performance in normal observers and its changes through training, tests of patients suffering from circumscribed lesions of the central nervous system (e.g. after a stoke, leading to e.g. visual field defects or amnesia), registration of the electrical activity of the central nervous system and its changes through training by means of multi-channel EEG recordings (and, more recently, MEG recordings and functional Magnetic Resonance Imaging (fMRI)).

Results: Through training, performance in some hyperacuity tasks improves by a factor of two or more within an hour. This improvement is specific for a number of rather low-level characteristics of the stimulus such as its exact orientation: improvement of thresholds does not transfer across a change in stimulus orientation of 10 degrees or more (Fig. 1). The improvement is also at least partly specific for the eye used during monocular training (Poggio, Fahle, Edelman, 1992), for the exact task trained (Fahle, Edelman, Poggio, 1995), with no transfer between a stimulus consisting of three (almost) aligned dots where the observer has to discriminate between a lateral offset of the middle point relative to an imaginary line through the endpoints (= vernier discrimination, inset of Fig. 2) and (almost) the same stimulus but where the observer has to discriminate between an offset towards one or the other of the endpoints, along the imaginary line between them (= bisection task; Fahle & Morgan 1996). There is also no transfer between different positions in the visual field, that is, improvement through learning is specific for visual field position. This specificity would argue for a relatively early location, along the visual pathways, of the neuronal changes underlying perceptual learning, and there are electrophysiological results both in primates (Gilbert, 1995) and psychophysical results in humans (Fahle & Skrandies, 1994) indicating changes in neuronal activity in the primary visual cortex. In line with this hypothesis, we find significant improvement through training in simple visual tasks even in patients suffering from amnesic syndromes - even if the patients have no memory whatsoever that they ever trained the task, their performance improves and stays at the higher level for at least several weeks. On the other hand, we find a strong influence of error feedback and attention on perceptual learning, indicating that the adult primary visual cortex is much more plastic than previously thought. Therefore, training might improve performance not only in normal observers, but also in patients suffering from circumscribed brain damage from different causes.

Figure 1

Figure 1: Improvement of performance (percentage of correct responses) as a function of training in a hyperacuity task. Orientation of the vernier stimulus was rotated by 10° after one hour of training; performance subsequently dropped to pretraining.

Figure 2

Figure 2: Means and standard errors of performance in a vernier and a bisection task. Six observers started with the vernier, the other six with the bisection task. The tasks were exchanged after one hour of training. Base-line differences between the 2 tasks will therefore cancel each other out. The 2nd day of testing started with a last measurement of the 1st stimulus type (data point immediately to the left of the hatched line).


  1. Poggio T, Fahle M, Edelman S. Fast perceptual learning in visual hyperacuity. Science 1992; 256: 1018-1021
  2. Fahle M, Edelman S, Poggio T. Fast perceptual learning in hyperacuity. Vision Research 1995; 35: 3003-3013
  3. Fahle M, Morgan M. Perceptual learning: No transfer of learning between vernier and bisection tasks. Current Biology 1996; 6: 292-297
  4. Fahle M, Skrandies W. An electrophysiological correlate of perceptual learning in humans. German J. Ophthalmology 1994; 3: 427-432