The pages on research, resources and publications will stay up, so that one can find our old work, but no new content will be added.

We are excited about joining the vibrant research community in Tuebingen, with exceptionally strong groups in Machine Learning, AI, as well as in Computational, Experimental, and Clinical Neuroscience!

Check out the Machine Learning Cluster of Excellence, the International Max Planck Research School Intelligent Systems, and the new International Max Planck Research School Mechanisms of Mental Function and Dysfunction!

Check google scholar for new publications, and twitter for updates!

Our paper “Can serial dependencies in choices and neural activity explain choice probabilities?” by Jan-Matthis, Jakob, and Hendrikje Nienborg has been accepted to Journal of Neuroscience and is available online.

Correlations, unexplained by the sensory input, between the activity of sensory neurons and an animal’s perceptual choice (“choice probabilities”) have received attention from both a systems and computational neuroscience perspective. Conversely, while temporal correlations for both spiking activity (“non-stationarities”) and for a subject’s choices in perceptual tasks (“serial dependencies”) have long been established, they have typically been ignored when measuring choice probabilities. Some accounts of choice probabilities incorporating feedback predict that these observations are linked. Here, we explore the extent to which this is the case. We find that, contrasting with these predictions, choice probabilities are largely independent of serial dependencies, which adds new constraints to accounts of choice probabilities that include feedback.

Alexandre’s poster submission “Inferring mesoscopic population models from population spike trains” has been accepted at COSYNE this year. There were 704 submissions to COSYNE this year, of which only 56% were accepted.

How does the interplay of single-neuron dynamics and neural connectivity give rise to the rich dynamical properties of neural populations? To tackle this question, it is desirable to have models which exhibit a wide range of population dynamics but remain interpretable in terms of connectivity and single-neuron dynamics. However, many commonly-used statistical models of neural population dynamics are based on generic models of dynamics (e.g. in Macke et al. 2011). Conversely, it has been challenging to link mechanistic spiking network models to empirical population data. To close this gap, we propose to model such data using mechanistic, but low-dimensional and hence statistically tractable models. We approximate neural populations as being composed of multiple homogeneous `pools’ of neurons, and model the dynamics of the aggregate population activity within each pool. We derive the likelihood of parameters (both single-neuron parameters and inter-pool connectivity) given this activity, which can then be used to either optimize parameters by gradient ascent on the log-likelihood, or to perform Bayesian inference using Markov Chain Monte Carlo (MCMC) sampling.

We illustrate this approach on a model based on generalized integrate-and-fire neurons (Schwalger et al., 2017). Using micro- and mesoscopic simulations of multiple neuron pools, we demonstrate that both single-neuron properties (membrane and adaptation constants) and connectivity-parameters (excitatory vs inhibitory connections and connection strengths) can be recovered on simulated data. Moving beyond point estimates, we compute the Bayesian posterior for combinations of parameters using MCMC sampling. Finally, we investigate how the approximations inherent to a mesoscopic population model impact the accuracy of the inferred single-neuron parameters. Ultimately, our method ensures compatibility between experimental multi-population data and mesoscopic dynamical models, by providing methods for statistical inference of low-dimensional mesoscopic models.

We had three papers accepted at NIPS this year– Jan-Matthis’ and Pedro’s paper Flexible statistical inference for mechanistic models of neural dynamics and Marcel’s paper Extracting low-dimensional dynamics from multiple large-scale neural population recordings by learning to predict correlations were accepted for poster presentations. Artur’s paper Fast amortized inference of neural activity from calcium imaging data with variational autoencoders was selected for a spotlight. There were 3240 submissions to NIPS this year, 678 of which were accepted at the conference, and 112 of which for spotlights.

We put online our new code repository CorBinian for modelling multivariate binary data in Matlab.
It extends and succeeds our previous toolbox, and features new fast MCMC code for regularized fitting of high-dimensional maxEnt models with Rao-Blackwellization, code for minimum probability flow fitting, as well as for computation of specific heat capacity from maxEnt models.

Our abstract ``Flexible Bayesian inference for complex models of single neurons’’ by Pedro J. Goncalves, Jan-Matthis Lueckmann, Giacomo Bassetto and Marcel Nonnenmacher, has been accepted for a talk at the Computational Neuroscience meeting this year.

Registration for this years Cajal Course in Computational Neuroscience has opened. The course will be hosted at the Champalimaud Centre for the Unknown in Lisbon, Portugal and is directed by Gilles Laurent from the MPI for Brain Research in Frankfurt, Christians Machens from the Champalimaud Centre for the Unknown and Jakob Macke from the Research Center caesar.

The course is designed for graduate students and postdoctoral fellows from a variety of disciplines. The application deadline is the 20th of March 2017 at midnight, CET time. More information about the course and how to apply visit the official website. This course is part of the CAJAL Advanced Neuroscience Training Programme - an initiative by FENS, IBRO and The Gatsby Foundation - and is hosted by Champalimaud Foundation, Portugal. We are grateful to Google DeepMind for additional funding.

The annual meeting on Computational and Systems Neuroscience is the most important conference for exchange of experimental and theoretical/computational approaches to problems in systems neuroscience. We are happy that all four of our submissions to the Cosyne Conference were successful this year, and will be presented at the conference. In addition, Artur and Jan-Matthis received travel grants.

On day 1 (February 23), Giacomo Bassetto will present his work on data-efficient receptive field estimation (I-27).

On day 2 (February 24), you are spoilt for choice between two posters from the group: In II-3, ``Flexible Bayesian inference for mechanistic models of neural dynamics’’ by Pedro Goncalves, Jan-Matthis Lueckmann, Giacomo Bassetto and Marcel Nonnenmacher, we are presenting an approach for making statistical inference possible and painless for any neuron model of your choice.

In II-77 ``Can serial dependencies in choices and neural activity explain choice probabilities?’’, Jan-Matthis Lueckmann’s work from our collaboration with Hendrikje Nienborg will investigate whether and how choice probabilities are affected by temporal dependencies.

On day 3 (February 25), check out Artur Speiser’s work with Srinivas Turaga and Evan Archer, on ``Amortized inference for fast spike prediction from calcium imaging data.’’– using deep learning to make generative-model based inference of action potentials much faster.

The International Max Planck Research School for Brain and Behavior is a fully funded graduate program in the neurosciences jointly hosted between caesar and the Max Planck Florida Institute for Neuroscience , USA, and in collaboration with partner universities in Bonn and Florida. The Ph.D. program is open to highly qualified and motivated candidates from all over the world who hold an outstanding diploma or master degree. We will be taking in our second cohort of students this year, the deadline is soon (December 1st)– apply at http://www.imprs-brain-behavior.org! Selection symposium will be in March next year, and positions will start in 2017.

Together with Allie Fletcher (UCLA), Eva Dyer and Konrad Koerding (Northwestern), Jascha Sohl-Dickstein (Google Research) and Joshua Vogelstein (John Hopkins), Jakob is organising a two-day workshop on Brains and Bits: Neuroscience Meets Machine Learning at NIPS 2016. It’s all happening on Friday-Saturday, December 9th-10th, 2016, in Barcelona. We have an exciting line-up of speakers from machine learning, computational neuroscience and computer-science:

• Yoshua Bengio, Université de Montréal
• Adrienne Fairhall, University of Washington
• Demis Hassabis, Google DeepMind
• Christos Papadimitriou,UC Berkeley
• Terry Sejnowski, Salk Institute, UCSD
• Anima Anandkumar, UC Irvine
• Mitya Chklovskii, Simons Foundation
• David Cox, Harvard
• Sophie Denève, ENS
• Emily Fox, University of Washington
• Surya Ganguli, Stanford
• Fred Hamprecht, Heidelberg
• Jonathan Pillow, Princeton
• Maneesh Sahani, Gatsby Unit, University College London

Deadline for submissions is 29/09/2016– we will have both posters and contributed talks!

More instructions and information on the workshop website.

Jan-Matthis, Pedro and Jakob are taking part in the CAJAL Course in Computational Neuroscience from August 7th to 27th at the Champalimaud Centre in Lisbon. Jan-Matthis will attend as a student, and Jakob and Pedro will be teaching. Congratulations to Jan-Matthis for being accepted to the course, and many thanks to the Boehringer Ingelheim Fonds for awarding a travel-grant to Jan-Matthis!

From August 1st to 7th, Jan-Matthis is attending the deep learning summer school in Montreal, organized by the Canadian Institute For Advanced Research.

Today Poornima Ramesh joined our group for an internship, on a DAAD Rise Scholarship. Welcome on board Poornima!

NVIDIA granted us a Titan X GPU for faster training of deep neural networks for analyzing behavioural data. Many thanks to NVIDIA, as well to Jan-Matthis for helping write the application.

Jakob is taking part in the 113th Titsee-conference by the Boehringer Ingelheim Fonds– looking forward to discussing `tools for quantifying brain and behaviour’ with an amazing group of scientists in the black forest.

Artur was accepted into the International Max Planck Research School Brain and Behavior, a partnership between caesar, the Max Planck Florida Institute for Neuroscience, Bonn University and Florida Atlantic University. He was also selected to receive a scholarship for his doctoral research. Congratulations to Artur!

Large-scale recordings of neuronal activity make it possible to gain insights into the collective activity of neural ensembles. It has been hypothesized that neural populations might be optimized to operate at a `thermodynamic critical point’, and that this property has implications for information processing. Support for this notion has come from a series of studies which identified statistical signatures of criticality in the ensemble activity of retinal ganglion cells. What are the underlying mechanisms that give rise to these observations?

We just uploaded a new manuscript that challenges these findings. In the manuscript, we show that signatures of criticality arise even in simple feed-forward models of retinal population activity. In particular, they occur whenever neural population data exhibits correlations, and is randomly sub-sampled during data analysis. These results show that signatures of criticality are not necessarily indicative of an optimized coding strategy, and challenge the utility of analysis approaches based on equilibrium thermodynamics for understanding partially observed biological systems.

The paper Painfree and accurate Bayesian estimation of psychometric functions for (potentially) overdispersed data with Heiko Schuett, Stefan Harmeling and Felix Wichmann was accepted at Vision Research. Psychometric functions play an essential role in much of behavioural neuroscience and psychophysics. In this paper, we describe efficient and accurate methods for estimating psychometric functions using Bayesian inference techniques. In contrast to previous methods, they dont require any parameter tuning or fickly MCMC, as the posterior is integrated numerically. In addition, the model also includes a term that captures `overdispersion’, which can result from fluctuations in the performance of a behaviour. A (hopefully) easy-to-use matlab implementation is available here

Amazon Web Services awarded us an ‘AWS Research Grant’ to develop dimensionality reduction methods for massively parallel neural population data, and to evaluate them on Amazon AWS. Thanks to Hooram and Marcel who helped in preparing the application.

The new International Max Planck Research School for Brain and Behavior is a fully funded graduate program in the neurosciences jointly hosted between caesar and the Max Planck Florida Institute for Neuroscience (MPFI), USA. The Ph.D. program is open to highly qualified and motivated candidates from all over the world who hold an outstanding diploma or master degree. Application deadline is December 1; positions available starting September 2016.

Jakob will be giving a talk about Marcels work at the Group for Neural Theory at Ecole Normale Superieure on Thursday, October 29th.

We are setting up an inter-disciplinary reading group focussed on computational neuroscience and machine learning. Be warned: We will not be afraid to read technical papers. We meet (roughly every second) Thursday morning in caesar near our offices. Everyone that is willing to actively contribute is welcome to attend, please send an email to Jakob to be added to the mailing list.

The paper `Unlocking neural population non-stationarities using hierarchical dynamics models’ by Mijung (Website) and Gergo Bohner was accepted at NIPS. There were 1838 submissions to NIPS this year, 403 of which were accepted at the conference. Pdf and code will follow soon. In this paper, we developed a model and inference procedure for tracking changes in the parameters of population models across trials. To illustrate the method, we used data that was kindly provided to us by Alexander Ecker.

Giacomo, Piotr and Jakob are travelling to Heidelberg next week to attend the Bernstein Conference. Jakob will give a talk in the Workshop `Estimating Parameters and Unobserved State Variables from Neural Data’ on Monday, 14/09/15, and will present Marcel’s work on ‘Correlations and signatures of criticality in neural population models’ in a contributed talk in the main conference on Wednesday 16/09/15. Giacomo will present a poster on our joint work with Alex Ecker on Wednesday evening.

Together with Allie Fletcher (UCSC & Redwood Center & UCLA), Ryan P Adams (Harvard & Twitter Cortex) and Jascha Sohl-Dickstein (Google Research), Jakob is organising a workshop on Statistical Methods for Understanding Neural Systems at NIPS 2015. It’s all happening on Friday, December 11th, 2015, in Montreal, Canada. Deadline for submissions is 10/10/2015.

We need you: We are hiring both graduate students and postdocs to join our group. If you have a background in a quantitative discpline and want to do graduate or postdoctoral research in computational neuroscience and machine learning, check out the advertisement!

Evan Archer (Website), who did his graduate work with Jonathan Pillow at at UT Austin (now Princeton) and with Jakob in Tuebingen, successfully defended his thesis in Tuebingen. In his graduate work, Evan developed Bayesian methods for statistical modelling of neural population spike trains, including very flexible and efficient methods for density estimation, entropy estimation, and system identification both for single neurons and populations of neurons.

Evan is now a postdoc with John Cunningham and Liam Paninski at the Statistics Department and Grossman Center for the Statistics of Mind at Columbia University. Many congratulations to the very first ‘Dr. rer. nat.’ from the lab!

From September 24-26, Visions in Science, the annual interdisciplinary scientific event organized bythe Max Planck PhDnet is taking place at caesar. This year’s conference is themed ‘Branch and Connect’ and it will feature talks on neuroscience and network analysis.

Marcel was accepted into the Advanced Course in Computational Neuroscience, which is part of the CAJAL Advanced Neuroscience Training Programme, and is held at the Champalimaud Foundation in Lisbon, Portugal, 9th-29th August 2015.

Jan-Matthis is going to a summer course in mining and modeling of neuroscience data, which is held at UC Berkeley between 6th and 17th July 2015. The course is organised by the Redwood Center for Theoretical Neuroscience, and is founded through CRCNS (Collaborative Research in Computational Neuroscience).

Today our new website went live. We will continue to update and expand content in the near future, so please bear with us as we are working on it.