How does the brain work? What are the kind of computations carried out by neural systems? We try to address these questions by analyzing recordings of neural activity and constructing mathematical models of neural circuits. Projects range from the study of sensory systems whose dynamics are dominated by sensory inputs to the study of cognitive systems (such as the prefrontal cortex) whose dynamics are strongly influenced by internal factors.
Neural mechanisms of working memory
The term "working memory" refers to our ability to keep something in mind for a short period of time and manipulate it. For instance, memorizing a telephone number, dialing it, and then forgetting it -- that is a standard example of working memory in action. In collaboration with Carlos Brody (Princeton) and Ranulfo Romo (UNAM Mexico), we study working memory in the context of a classical psychophysical task ("two-interval-discrimination task") in which a subject observes a stimulus f1, keeps it in mind for a couple of seconds, and then makes a decision by comparing it to a second stimulus f2. Ranulfo Romo has recorded the activities of thousands of neurons in various regions of the primate brain, yielding fascinating snapshots of neural networks performing this simple task. We seek to develop simple neural network models that can reproduce the observed neural activities and that provide insights into some of the principles and mechanisms that underlie working memory.
[1] Machens CK, Brody CD (2008).
Design of continuous attractor networks with monotonic
tuning using a symmetry principle. Neural Comp 20(2):452-485
pdf.
[2] Machens CK, Romo R, Brody CD (2005).
Flexible dynamics of mutual inhibiton: a neural model of two-interval discrimination.
Science 307:1121-1124. pdf.
Neural mechanisms of sensory processing
Sensory systems (such as the auditory, visual, or olfactory systems) help animals to extract behaviorally relevant information from their environment. While their internal architecture relies heavily on lateral and recurrent connections, most of our models do not, but assume instead a feed-forward architecture. We are interested in understanding better the use of these recurrent connections in sensory processing and hope that such an understanding could help to design models of sensory systems that are actually able to predict neural responses to complex, natural stimuli (the type of stimuli that sensory systems are designed to process.) We pursue this research in collaboration with the lab of Axel Borst (motion processing in flies).
[3] Machens CK, Wehr MS, Zador AM (2004). Linearity of cortical receptive fields measured with natural sounds. J Neurosci 24(5):1089-1100. pdf
Theoretical principles of sensory processing
For sensory systems that perform more complicated tasks (e.g. object or sound recognition in the visual or auditory system of mammals) we usually lack computational theories that tell us which set of computations need to be carried out. One of the most successful candidate theories for early sensory processing (across modalities) has been the "efficient coding hypothesis" which states that the coding strategy of neurons matches the stimulus statistics in an animal's natural habitat. We are interested in developing and reformulating such theories of sensory processing in order to relate neural responses to computational principles and the evolutionary constraints that animals face.
[4] Machens CK, Gollisch T, Kolesnikova O, Herz AVM (2005). Testing the efficiency of sensory coding with optimal stimulus ensembles. Neuron 47:447-456.pdf.