Perceptual learning and representational learning in humans and animals by ProQuest


More Info
									Learning & Behavior
2009, 37 (2), 141-153

              perceptual learning and representational learning
                           in humans and animals
                                                             József fiser
                                              Brandeis University, Waltham, Massachusetts

                Traditionally, perceptual learning in humans and classical conditioning in animals have been considered as
             two very different research areas, with separate problems, paradigms, and explanations. However, a number of
             themes common to these fields of research emerge when they are approached from the more general concept
             of representational learning. To demonstrate this, I present results of several learning experiments with human
             adults and infants, exploring how internal representations of complex unknown visual patterns might emerge
             in the brain. I provide evidence that this learning cannot be captured fully by any simple pairwise associative
             learning scheme, but rather by a probabilistic inference process called Bayesian model averaging, in which the
             brain is assumed to formulate the most likely chunking/grouping of its previous experience into independent
             representational units. Such a generative model attempts to represent the entire world of stimuli with optimal
             ability to generalize to likely scenes in the future. I review the evidence showing that a similar philosophy and
             generative scheme of representation has successfully described a wide range of experimental data in the domain
             of classical conditioning in animals. These convergent findings suggest that statistical theories of representa­
             tional learning might help to link human perceptual learning and animal classical conditioning results into a
             coherent framework.

   The phenomenon of learning is central to humans and                 tational paradigm of classical conditioning in a Bayesian
animals, and as such it is the focus of an intense scien­              framework and thus afforded better insight into the under­
tific investigation. However, due to the complex nature of             lying computation than did classical models. In the con­
learning, the areas of research on learning are multiple,              cluding part of the article, I highlight parallels between
with little crosslinking between these areas, even in the              the two approaches and propose how representational
more restricted domain of behavioral and system neuro­                 learning might shed new light on the computational nature
science. The goal of the present review is to highlight                of a large class of learning phenomena.
some recent empirical and theoretical developments in
both human and animal learning that suggest a possibly                      ClassiCal perCeptual learning
more general framework within which learning could be                                 in humans
interpreted and investigated successfully. The more gen­
eral framework is grounded in a statistical reformulation                 Perceptual learning has traditionally been defined
of different learning theories; more specifically, express­            as a practice­induced improvement in humans’ ability
ing learning as an implicit Bayesian inference carried out             to perform specific perceptual tasks (Fahle & Poggio,
by humans and animals.                                                 2002). In perceptual learning paradigms, the subject is
   The structure of the article is as follows. First, I describe       presented with a well­defined task explained verbally
the classical definition of human perceptual learning and              by the experimenter—orientation discrimination, for ex­
the way this classical paradigm could be modified a rep­               ample (Fiorentini & Berardi, 1980; Furmanski & Engel,
resentational learning paradigm, to examine how humans                 2000; Petrov, Dosher, & Lu, 2006), texture discrimination
acquire new complex representations of their environ­                  (Ahissar & Hochstein, 1997; Karni & Sagi, 1991), mo­
ment. Next, I explain a number of results in human learn­              tion direction discrimination (Matthews, Liu, Geesaman,
ing within this representational learning paradigm, and I              & Qian, 1999), or a hyperacuity test (Poggio, Fahle, &
show how these results are naturally explained within a                Edelman, 1992). After repetitive training (typically in­
Bayesian framework. In the last part of the article, I shift           cluding feedback), the subject’s performance improves as
gears and argue that classical conditioning in animals is              quantified by threshold or reaction time measures. This
a sort of perceptual learning and can be reformulated in               experimental paradigm has been explored extensively in
a representational learning paradigm, in much the same                 both the psychophysical (Dosher & Lu, 1998; Furmanski
way as classical perceptual learning in humans can be. I               & Engel, 2000; Gold, Bennett, & Sekuler, 1999) and the
present the results of studies that reformulated represen­             neurophysiological domains (Gilbert, Sigman, & Crist,

                                                       J. Fiser,

                                                                   141                      © 2009 The Psychonomic Society, Inc.
142      Fiser

2001; Schoups, Vogels, Qian, & Orbán, 2001), and it is          from a large set of potential descriptions, during classi­
considered the dominant experimental approach to human          cal perceptual learning subjects improve their threshold
sensory learning that does not strictly deal with abstract      of discrimination between well­specified simple patterns,
cognitive tasks such as concept learning (Fahle & Poggio,       which can be done by increasing the sensitivity of existing
2002; Fine & Jacobs, 2002).                                     detectors. Second, whereas representational learning is
   Despite this widespread view, the classical paradigm         largely detached from conscious access, the attribute to be
captures only one type of change occurring in the brain         learned during c
To top