Biological and Deep Learning Workshop

July 12, 2018 | 9:00–17:00 | Hörsaal 7, Studenteneingang 3rd Floor, Virchowweg 14

Program

09:45 – 10:00 Matthew Larkum / Jaan Aru – Welcome

10:00 – 10:45 Brice Bathellier

Brice is interested in the computations by which neural populations represent sensory stimuli and link them to behavior. After a PhD, on oscillations and population codes in the olfactory bulb, he focused on the auditory cortex. Currently, he is a CNRS researcher and heads his own lab. He has developed simple extensions of reinforcement learning models which allow precisely fitting the dynamics and variability of sound discrimination learning in mice. Based on these models, and on large scale imaging data and optogenetics, his team has recently shown that unexpected parameters of sound representations, such as the amount of recruited activity, play an important role in sound discrimination learning. He will talk about this exciting study and novel developments in his lab.

10:45 – 11:00 – Coffee Break

11:00  – 11:45 Jasper Poort 

Jasper studied mechanisms of perceptual organization and attention in different visual and frontal areas in the primate visual cortex during his PhD. He then studied how responses of cells in the mouse primary visual cortex change while mice learn a new visual discrimination task (Neuron, 2015), recently also by simultaneously imaging responses of pyramidal cells and PV, SOM and VIP inhibitory interneurons (Nat Neurosci, 2018). He will discuss some of this work as well as new data comparing the effects of perceptual learning and attentional switching on the same populations of cells in visual cortex

11:45 – 12:30 Walter Senn

Walter Senn did his PhD in mathematics with specialty in differential geometry. After a postdoc in neural computation he joined the University of Bern where he is a full professor since 2006. He is well known for his work on the computational models of pyramidal neurons. Recently he has also been interested in the biological implementation of deep learning and has suggested that synaptic learning is driven by a local dendritic prediction error that arises from a failure to predict the top-down input given the bottom-up activities. He will describe this work and his recent efforts to understand the convergence of biological and machine learning.

12:30  13:00 Jaan Aru – Summary of the session, discussion

13:00 – 14:00 Lunch

14:00 – 14:45 Guy Doron / Matthew Larkum

Guy developed during is PhD a novel method for in-vivo single neuron stimulation, termed ‘nanostimulation’. He  applied this technique to study neural coding during behavioral report of single neuron stimulation and found that irregular bursty firing patterns in single neurons are more easily detected compared to regular firing patterns (Neuron, 2014). He is currently leading research in the Larkum lab aimed to examine the neural circuit of associative memory formation in rodents. He will describe his novel findings about how perirhinal cortex modulates learning and memory in the somatosensory cortex.

14:45 – 15:00 – Coffee Break

15:00 – 15:45 Blake Richards

Blake Richards studied synaptic plasticity during his PhD and memory consolidation during his post-doc. Now, in his own lab he is working at the intersection of neuroscience and deep learning. He has been trying to understand how the brain solves the credit assignment problem that is normally solved by the backpropagation-of-error algorithm in machine learning. In particular, he has suggested that the apical compartment of the pyramidal cells could provide a backprop-like teaching signal for the somatic compartment. He will also talk about the insights deep learning could provide for guiding experiments in neuroscience.

15:45 – 16:30 Caspar Schwiedrzik

Caspar has worked on perceptual learning since his PhD. During that time he studied perceptual learning in humans, but then for his post-doc switched to monkeys and has now his own monkey lab in Gottingen. He has a recent first author Neuron paper on statistical learning that provides insights about the “predictive coding” theory. Caspar will present data from both humans and monkeys and from several different experimental paradigms.  

16:30 – 17:00 Matthew Larkum – Summary and general discussion

Participating Institutions