Time/Room: Friday, May 15, 2015, 5:00 – 7:00 pm, Pavilion
Organizer(s): Chris Baker and Hans Op de Beeck; NIMH, USA; University of Leuven, Belgium
Presenters: Krishna Srihasam, Rufin Vogels, David J. Freedman, Andrew E Welchman, Aaron Seitz
Symposium Description
The capacity for learning is a critical feature of vision. It is well established that learning is associated with changes in visual representations and the underlying neural substrate (e.g. sharper behavioral discrimination and sharper neural tuning for trained visual features such as orientation or shape). However, the brain regions involved vary from experiment to experiment, ranging from primary visual cortex to all higher levels in the visual system. One working hypothesis suggests that the hierarchical level at which neural plasticity is most prominent is related to the complexity of the stimuli and the task context, but results do not necessarily support this prediction. Further, the nature of the changes is often inconsistent between studies. In this symposium we emphasize the viewpoint that in order to understand how learning changes the brain, it is critical to consider the underlying complexity and distributed nature of the visual system. The group of speakers we have assembled will present work using a variety of different approaches from behavior to TMS to fMRI in both monkeys and humans. The consistent theme across talks will be that a fuller and better understanding of neural plasticity might be achieved by considering how learning impacts processing from neurons to circuits to regions in the context of the distributed neural architecture of vision. Individually, the speakers will highlight specific properties of the visual system that have an important role in visual learning but are often not considered in theories of learning. First, brain regions differ in their average response and selectivity even before learning, and might each have a different role in learning, making a search for THE visual learning area unrealistic. Further, simple classification schemes such as low-level areas subserve low-level learning and high-level areas high-level learning might vastly underestimate how effects of learning are distributed across hierarchical levels. In addition, the regions critical for a task might change as a function of learning. Even more in detail, different cell types have different roles in visual processing and are possibly changed in different ways through learning. Finally, computational and behavioral approaches also emphasize that learning involves multiple learning processes, and understanding their interaction is crucial. Through these examples we will showcase the complexity of the processes involved in visual learning at the behavioral, neural, and computational level. This symposium should be of broad interest to the VSS community from students to faculty, providing a multidisciplinary overview of current approaches to visual learning. Often visual learning is studied in specific limited domains and the goal of this symposium is to try to integrate findings across different levels and different scales of visual processing, taking into account the complexity of the neural system.
Presentations
Novel module formation reveals underlying shape bias in primate infero-temporal cortex
Speaker: Krishna Srihasam; Department of Neurobiology, Harvard Medical School, Boston, MA
Authors: Margaret S. Livingstone; Department of Neurobiology, Harvard Medical School, Boston, MA
Primate inferotemporal cortex is divided up into domains specialized for processing specific object categories, such as faces, text, places, and body parts. These domains are in stereotyped locations in most humans and monkeys. What are the contributions of visual experience and innate programs in generating this organization? The reproducible location of different category-selective domains in humans and macaques suggests that some aspects of IT category organization must be innate. However, the existence of a visual word form area, the effects of expertise and our recent finding that novel specializations appear in IT as a consequence of intensive early training indicate that experience must also be important in the formation or refinement of category-selective domains in IT. To ask what determines the locations of such domains, we intensively trained juvenile monkeys to recognize three distinct sets of shapes: alphanumeric symbols, rectilinear shapes and cartoon faces. After training, the monkeys developed regions that were selectively responsive to each trained set. The location of each specialization was similar across monkeys, despite differences in training order. The fact that these domains consistently mapped to characteristic locations suggests that a pre-existing shape organization determines where experience will exert its effects.
Learning to discriminate simple stimuli modifies the response properties of early and late visual cortical areas
Speaker: Rufin Vogels; Laboratorium voor Neuro- en Psychofysiologie, Dpt. Neurowetenschappen, KU Leuven Campus Gasthuisberg, Belgium
Authors: Hamed Zivari Adab; Laboratorium voor Neuro- en Psychofysiologie, Dpt. Neurowetenschappen, KU Leuven Campus Gasthuisberg, Belgium
Practicing simple visual detection and discrimination tasks improves performance, a signature of adult brain plasticity. Current models of learning with simple stimuli such as gratings postulate either changes in early visual cortex or reweighting of stable early sensory responses at the decision stage. We showed that practice in orientation discrimination of noisy gratings (coarse orientation discrimination) increased the ability of single neurons of macaque visual area V4 to discriminate the trained stimuli. Then we asked whether practice in the same task also changes the response properties of later visual cortical areas. To identify candidate areas, we used fMRI to map activations to noisy gratings in the trained monkeys, revealing a region in the posterior inferior temporal (PIT) cortex. Subsequent single unit recordings showed that the PIT neurons discriminated better the trained compared with the untrained orientations, even when the animals were performing an orthogonal task. Unlike in previous single unit studies of learning in early visual cortex, more PIT neurons preferred trained compared with untrained orientations. Thus, practicing a simple discrimination of grating stimuli cannot only affect early visual cortex but also changes the response properties of late visual cortical areas. Perturbation of the activity in PIT reduced the coarse orientation discrimination performance in the trained animals, suggesting that this region is indeed part of the network underlying the performance in the task. We suggest that visual learning modifies the responses of most if not all areas that are part of the cortical network which supports the task execution.
Learning-dependent plasticity of visual encoding in inferior temporal cortex
Speaker: David J. Freedman; Department of Neurobiology, The University of Chicago
Authors: Jillian L. McKee; Department of Neurobiology, The University of Chicago
Our ability to recognize complex visual stimuli depends critically on our past experience. For example, we easily and seemingly automatically recognize visual stimuli such as familiar faces, our bicycle, or the characters on a written page. Visual form recognition depends on neuronal processing along a hierarchy of visual cortical areas which culminates in inferior temporal cortex (ITC), which contains neurons which show exquisite selectivity for complex visual stimuli. Although both passive experience and explicit training can modify or enhance visual selectivity in ITC, the mechanisms underlying this plasticity are not understood. This talk will describe studies aimed at understanding the impact of experience on visual selectivity in ITC. Monkeys were trained to perform a categorization task in which they classified images as novel or familiar. Familiar images had been repeatedly viewed over months of prior training sessions, while novel images had not been viewed prior to that session. Neurophysiological recordings from ITC and prefrontal cortex (PFC) revealed a marked impact of familiarity on neuronal responses in both areas. ITC showed greater stimulus selectivity than PFC, while PFC showed a more abstract encoding of the novel and familiar categories. We also examined familiarity-related changes in ITC encoding within individual sessions, while monkeys viewed initially novel stimuli ~50 times each. This revealed enhanced stimulus selectivity with increasing repetitions, and distinct patterns of effects among putative inhibitory and excitatory neurons. This may provide a mechanism for familiarity-related changes in ITC activity, and could help understand how ITC stimulus selectivity is shaped by learning.
Training transfer: from functional mechanisms to cortical circuits
Speaker: Andrew E Welchman; University of Cambridge, UK
Authors: Dorita F Chang; University of Cambridge, UK
While perception improves with practice, the brain is faced with a Goldilocks challenge in balancing the specificity vs. generality of learning. Learning specificity is classically established (e.g. Karni & Sagi, 1991, PNAS 88, 4966-4970), however, recent work also reveals generalisation that promotes the transfer of training effects (e.g., Xiao et al, 2008, Cur Biol, 18, 1922-26). Here I will discuss how we can understand the neural mechanisms that support these opposing drives for optimising visual processing. I will discuss work that uses perceptual judgments in visual displays where performance is limited by noise added to the stimuli (signal-in-noise tasks) or clearer displays that push observers to make fine differentiation between elements (feature difference tasks). I will review work that suggests different foci of fMRI activity during performance of these types of task (Zhang et al, 2010, J Neurosci, 14127-33), and then describe how we have used psychophysical tests of learning transfer to understand the mechanisms that support learning (Chang et al, 2013, J Neurosci, 10962-71). Finally, I will discuss recent TMS work that implicates a wide high-level network involved in generalisation of training between tasks.
Moving beyond a binary view of specificity in perceptual learning
Speaker: Aaron Seitz; Department of Psychology University of California, Riverside
A hallmark of modern perceptual learning is the nature to which learning effects are specific to the trained stimuli. Such specificity to orientation, spatial location and even eye of training (Karni and Sagi, 1991), has been used as psychophysical evidence of neural basis of learning. However, recent research shows that learning effects once thought to be specific depend on subtleties of the training procedure (Hung and Seitz, 2014) and that within even a simple training task that there are multiple aspects of the task and stimuli that are learned simultaneously (LeDantec, Melton and Seitz, 2012). Here, I present recent results my from my lab and others detailing some of the complexities of specificity and transfer and suggest that learning on any task involves a broad network of brain regions undergoing changes in representations, readout weights, decision rules, feedback processes, etc. However, importantly, that the distribution of learning across the neural system depends upon the fine details of the training procedure. I conclude with the suggestion that to advance our understanding of perceptual learning, the field must move towards understanding individual, and procedurally induced, differences in learning and how multiple neural mechanisms may together underlie behavioral learning effects.