Matteo Carandini
Speaker of Workshop 1
Will talk about: Looking for canonical neural computations in the visual system
Matteo Carandini’s research aims to understand how populations of neurons in cortex process sensory information, and how they integrate it with information from within the brain to guide perception and action. He works at University College London, where together with Kenneth Harris he runs the Cortical Processing Laboratory (www.ucl.ac.uk/cortexlab). The laboratory uses a combination of experiment and computational analysis. Data are acquired in the mouse brain, with techniques such as multielectrode recordings, optogenetics, widefield and multiphoton imaging, operant conditioning, and virtual reality simulation. Carandini holds the GlaxoSmithKline / Fight for Sight Chair in Visual Neuroscience. He received a Senior Investigator award from the Wellcome Trust, and an Advanced Investigator award from the European Research Council.
The primary visual cortex (V1) codes fundamental attributes of visual stimuli, representing them in the coordinated and dynamically-changing activity of populations of neurons. Does this representation follow simple mathematical rules? Are these rules stable or do they change according to stimulus properties such as a recent history, strength, or configuration? Working with Andrea Benucci and Neel Dhruv, we addressed these questions in a series of experiments performed in cats and mice, where we recorded from populations of V1 neurons using multielectrode arrays. In response to sequences of stimuli of different orientations, the cortex adopts a very simple linear coding scheme, with a stereotyped response that is simply applied additively to different groups of neurons depending on the stimulus orientation. This basic linear representation, however, is only the scaffolding for more complex, nonlinear operations that make the cortex extremely adaptive. First, in response to sums of stimuli of different contrast, the cortex engages in a range of behaviors, from simple summation to winner-take-all competition. A simple model based on divisive normalization summarizes all these effects. Second, the cortex shows a marked ability to adapt to the statistics of the stimuli: it changes the selectivity and responsiveness of neurons just as needed to counteract any biases in the recent history of stimulation. A simple model based on equalization summarizes these effects. These nonlinear operations cascade from area to area: adaptation to the statistics of the stimuli in one stage of the visual system has profound effects on the inputs to the subsequent stage, and these effects can be predicted by a very simple model based on summation with fixed weights. We have great hope and reasonable expectation that the rules that we have uncovered are not specific to area V1 but are rather canonical rules of operation of cortical populations. These rules may act as guide to research in the underlying mechanisms and circuits and in the neural computations that lead to perception and behavior.
