See google scholar for a full list of publications

Key publications
Sequential spatial reasoning via neural dynamics
This study is a step towards using non-trivial navigation tasks in rodents as a tool to study cognitive processes. Mice can learn to perform a spatial task where they have to form, retain & use hypotheses to identify otherwise ambiguous landmarks that can only be seen one at a time. Neurons in RSC encode the ensuing hypothesis states (states that are informed by prior sensory information & self-motion, and can disambiguate future sensory data) conjunctively with many other variables, and were constrained by stable recurrent dynamics. In collaboration with Ingmar Kanitscheider in Fiete lab we found that this type of recurrent dynamics can in principle solve the task, showing that local circuits in of associative cortex can form, retain, and apply hypotheses. This suggests that general sequential reasoning can be performed via low dimensional recurrent neural dynamics, where hypotheses, motor and sensory data, and their interpretation are co-represented and influence each other.

Spatial reasoning via recurrent neural dynamics in mouse retrosplenial cortex
J Voigts, I Kanitscheider, NJ Miller, EHS Toloza,  JP Newman,  IR Fiete,  MT Harnett, bioRxiv
Dendritic computations during 2D navigation
In this paper we describe the method of allowing mice to rotate their head around the vertical axis during 2-photon imaging and use the method to image dendrites and associated soma of L5 cells in restrosplenial cortex during navigation. We find that the tuning of the activity of dendritic segments differs (albeit not in an obviously systematic way) from the tuning of their parent soma, showing that the conditions for dendritic coincidence detection are met in this context.

Somatic and dendritic encoding of spatial variables in retrosplenial cortex differs during 2d navigation
J Voigts, MT Harnett, Neuron 105 (2), 237-245. e4
Laminar representation of sensory deviations
The central finding in this study is that if we subtly change the activity pattern of L6 neurons while maintaining their mean rate, mice are unable to process sensory deviations, such as small changes in the distance to an object. Other sensory functions however, including threshold level detection or gap-crossing are unaffected. This shows that specific sensory-driven activity patterns in L6 are needed for change detection but not for other sensory function.

Layer 6 ensembles can selectively regulate the behavioral impact and layer-specific representation of sensory deviants
J Voigts, CA Deister, CI Moore, eLife 2020;9:e48957
Gap-crossing as a model of evidence accumulation
These papers deal with how mice integrate evidence about the distance to an object (in this case a platform across a gap). The central idea here is that the pattern of whisker movements is an indicator of the mouse's evolving hypothesis about their distance to objects.

Unsupervised whisker tracking in unrestrained behaving animals
J Voigts, B Sakmann, T Celikel, Journal of neurophysiology 100 (1), 504-515882008

Tactile object localization by anticipatory whisker motion
J Voigts, DH Herman, T Celikel, Journal of neurophysiology 113 (2), 620-632