Tatiana Engel
Modeling: Attention and Learning

Personal Background

I joined Brains in Silicon and Prof. Tirin Moore's neurophysiology lab as a postdoctoral researcher in 2012, excited about the opportunity to combine large-scale cortical simulations on Neurogrid and analysis of multi-electrode recordings in behaving monkeys in order to derive more accurate biophysical models of cognitive functions. Neurogrid challenges many models proposed in computational neuroscience that typically comprise only few thousand neurons with little (if any) heterogeneity among them.

In Neurogrid, much like in the brain, almost any neural parameter varies significantly across the population, and scaling up to million neurons produces new unexpected behaviors. Therefore computational models have to stand the test of heterogeneity and scale to function on Neurogrid. At the same time, neurophysiological multi-electrode data collected in the Moore lab provide just the necessary constraints and inspiration for large-scale modeling of cognitive processes.

Before joining Brains in Silicon, I was a postdoctoral researcher in the lab of Prof. Xiao-Jing Wang at Yale University. There I worked on building computational and neural circuit models of category learning, working memory and decision making. Prior to that, I obtained my PhD in Theoretical Physics from Humboldt University of Berlin in Germany, and my MSc in Physics from Lomonosov Moscow State University in Russia.


Research Goals

Neuronal population dynamics in the mammalian cerebral cortex is dominated by endogenously generated activity, which results in high variability of spontaneous and stimulus-evoked responses. This endogenous activity, and resulting variability, can be internally controlled within the brain with high spatial and temporal precision. For example, selective visual attention reduces and decorrelates activity fluctuations in neurons responding to stimuli in the attended spatial location. This reduction in correlated variability is thought to underlie perceptual benefits of attention. However, little is known about neural processes that generate endogenous activity and about mechanisms involved in its control.

To address this question, I collaborate with Nick Steinmetz, a graduate student in the Boahen and Moore labs. Nick records neural activity with laminar array electrodes throughout the cortical depth in the visual area V4 of monkeys engaged in a demanding attentional task. He also simultaneously records in the Frontal Eye Fields (FEF), an area that controls attention and saccadic eye movements. Through analysis of Nick's multi-electrode recordings and computational modeling, I seek to better understand neural mechanisms underlying generation and control of endogenous cortical activity.


Project Status

Hidden Markov Modeling (HMM) of Up-Down transitions. HMM estimates the probabilities of transitioning or remaining in the same state, assuming that transitions happen probabilistically and that each channel emits spikes randomly at different rates in the Up and Down states (upper panel). The model is fit to spiking activity recorded simultaneously at 16-channels of the multi-electrode array (lower left panel). Spikes recorded on each channel are depicted by dots in the corresponding row of the spike raster (lower right panel). Spikes assigned to the Up and Down states are colored green and magenta, respectively.

My analysis of laminar recordings in area V4 showed that population activity in awake behaving monkeys abruptly transitions between intervals of vigorous spiking (Up state) and quiescence (Down state). These transitions are highly synchronized throughout the cortical depth. Up-Down transitions occur spontaneously as well as during visual stimulation and are not locked to any behavioral event.

To extract statistics of Up-Down transitions from population spiking data, I have been using Hidden Markov Modeling. Intriguingly, we established that transition frequency decreases and Up-state duration increases when monkeys attend or prepare a saccade to the receptive field of recorded neurons. This is the first demonstration that internal cortical state can be modulated at a local scale to serve behavioral goals.

As the next step, I will analyze simultaneous V4-FEF recordings to establish how activity in FEF exerts spatially-selective control over the cortical state in V4. In addition, I will build a cortical circuit model to test possible neural network mechanisms by which the cortical state is modulated in the V4-FEF microcircuit.




ID Article Full Text
J42 T Engel, N A Steinmetz, M A Gieselmann, A Thiele, T Moore, and K Boahen, Selective Modulation of Cortical State During Spatial Attention, Science, vol. 354, issue 6316, pp. 1140-1144, December 2, 2016.
Full Text