The main aim of my research is to understand how the brain (and specifically its individual neurons) makes decisions based on information from the senses (typically, vision) and from past experience (learning a task, learning the value of a choice). I work in close collaboration with Kenneth Harris, with whom I share the Cortical Processing Laboratory. We are also part of the International Brain Laboratory.

Below are my main research contributions. Of possible interest, see also my failed covers for Neuron.


Some of our main discoveries concern a neural computation called divisive normalization, whereby the activity of a neuron is divided by the summed activity of other neurons. I started working on normalization when I was a graduate student at New York University, collaborating with David Heeger at Stanford University. I refined the model, proposed a biophysical substrate for it based on synaptic inhibition, and provided direct support for it in neurons of the visual cortex of the primate brain (Science 1994; J Neurosci 1997).

After opening my own laboratory, I extended the model to other brain regions. Working with Mante and Bonin I showed that normalization is also at work in earlier stages of the visual system (J Neurosci 2005; Nature Neurosci 2005; Neuron 2008). Working with Busse I extended the model from single neurons to entire populations, showing that a single equation can capture the activity of vast numbers of neurons in the cerebral cortex (Neuron 2009).

These results and those obtained by others in other species and systems led Heeger and I to claim that normalization is a canonical neural computation, one that is performed in multiple neural systems to serve multiple functions (Nature Rev Neurosci, 2012).

In related work, my laboratory investigated the neural circuits that provide normalization in the cerebral cortex. We addressed this question with progressively more advanced techniques. First we did simple neurophysiology (Neuron 2002), then working with Katzner we turned to pharmacology (J Neurosci 2011), and finally working with Sato, optogenetics (Nature Neurosci 2014) paired with intracellular recordings (Nature Neurosci 2016). These experiments allowed me to revise my earlier proposal that normalization in cortex is due to increases in inhibition. Rather, the results reveal that the mechanism controlling responsiveness in the cortex relies on changing the mutual excitation between neurons.


A related thread of investigations concerns the roles of neural inhibition in cortical computations. Working first as a postdoc with Ferster, I showed that synaptic inhibition in visual cortex has the same sensory selectivity as excitation (J Neurophysiol, 2000). Later, working with Katzner I showed that inhibition in cortex controls neuronal responsiveness but not selectivity (J Neurosci 2011). Working with Atallah and Scanziani, I showed that this effect was due to the most widespread type of cortical inhibitory neurons, those that express Parvalbumin (Neuron 2012).

Working with Haider and Hausser, we were then the first to measure the strength of synaptic inhibition in the cerebral cortex during wakefulness (it had always been measured under anesthesia). This led to the discovery that during wakefulness inhibition dominates responses (Nature, 2013).


A third area of investigation concerns the phenomenon of visual adaptation, the reduction in responsiveness seen in the visual system after prolonged visual stimulation. Working with Ferster, we discovered a cellular explanation for adaptation in the cerebral cortex (Science 1997). Working with Barlow, I then tested an influential hypothesis: that adaptation causes decorrelation (Philos Trans R Soc B, 1997). Later, working with Dhruv we revealed how adaptation cascades from one brain region to the subsequent one (Neuron, 2014) and working with Benucci we showed how adaptation acts as a homeostatic mechanism that equalizes response levels across large populations of neurons (Nature Neurosci, 2013).


Currently, the main area of our work centers on the laws governing the responses of populations of neurons in visual cortex (Nature Neurosci 2009; Neuron 2009). Working with Benucci, I characterized stimulus-driven travelling waves in cortex (Neuron 2007), and working with Nauhaus and Ringach, we discovered that such waves depend on brain state (Nat Neurosci 2009; Neuron 2012).

A key set of findings concern the relationship between the activity of single neurons and that of populations. Working with Nauhaus and Ringach, we discovered a law relating the selectivity of individual neurons and their location in visual cortex (Neuron 2008). Working with Okun and Harris, we discovered that synaptic connectivity makes some cortical neurons respond in tune with the nearby population an others independently of it (“choristers” vs. “soloists”, Nature 2015). These measures allowed us to reveal the nature of cortical “noise”, the variability in sensory responses observed in visual cortex.

I had established that the relationship between intracellular currents and firing rate plays a key role in amplifying response variability (PLoS Biol 2004). Later work with Lin, Scholvinck, and Harris showed that the main sources of variability in visual cortex are shared across neurons (Neuron 2015; J Neurosci 2015).

Current work is investigating the sources of these non-sensory signals in sensory cortex. Work with Ayaz and Saleem investigated how these are related to locomotion (Curr Biol 2013) and to navigation (Nature Neurosci 2013).


To enable this research, our laboratory has developed multiple techniques to measure the activity of neuronal populations. We characterized a widespread electrical measure, the local field potential, relating it to the activity of the underlying populations and synaptic currents (Neuron 2009; Neuron 2016). We also perfected the use of multielectrode arrays in visual cortex (e.g. Nat Neurosci 2009; Neuron 2009; Nat Neurosci 2009), and helped develop new methods for imaging with voltage-sensitive dyes (Neuron 2007) and voltage-sensitive fluorescent probes (J Neurosci 2015; Neuron 2015).

The latest of these technologies are Neuropixels probes, which have 1,000 recording sites. These sites record from hundreds of neurons distributed across brain regions. For instance, working with Steinmetz and Harris, we recently demonstrated that using two Neuropixels probes we could record from over 500 neurons in 5 regions of the mouse brain simultaneously. Our laboratory has played a key role in the development of the first generation of these probes and, thanks to Marius Pachitariu, in the software tools needed to parse their output (bioRxiv, 2016), and I lead the consortium funded to develop the second generation of the probes in the next five years. Meanwhile, advances in image processing in our laboratory have made us the first laboratory to image simultaneously the activity of 10,000 neurons in the awake brain (bioRxiv, 2016).


Our current research expands on these findings and techniques, and focuses on studying the visual system in awake mice that are engaged in a variety of behaviors, from simple perceptual decisions to more complex forms of decision and navigation, both in physical reality and in virtual reality (Nature Neurosci 2013; bioRxiv, 2016). In devising this research in mice, we can build on our track record of innovative psychophysical measurements in humans. In early studies, we provided one of the first tests of an influential Bayesian model of speed perception (Vision Res 2002), and provided psychophysical evidence against a role for inhibition in normalization (J Vis 2002). More recently, working with Busse and with Dakin and Gardner we revealed an unexpected role of superstition in biasing the perceptual decisions of mice (J Neurosci 2011) and in humans (PNAS, 2016).