poster session
(list not yet complete...)
- 1) Robustness and evolvability in a model of a pattern recognition network (Daesung Cho, University of Göttingen, European Neuroscience Institute, Göttingen, Germany)
abstract…
The large diversity of behaviors even among closely related species indicates the evolvability of the underlying neural circuit. At the same time, the behaviors must be functionally robust, but how systems can be both robust and evolvable is still an open question in neuroscience and evolution. Studying robustness and evolvability requires the mapping between genotypes and phenotypes, which is challenging to obtain experimentally. However, models of neural circuits that generate behavior can be used as a proxy of the biological system, and the mapping between model parameters and model output can be used as a proxy for the genotype-to-phenotype map. Here, we combine Bayesian inference and information theory to quantify robustness and evolvability in circuit models. We test this method using a model of the acoustic pattern recognition circuit in crickets. This circuit consists of linear filters and nonlinearities and can reproduce the full behavioral diversity of song recognition found in crickets. We demonstrate that the method correctly obtains the mapping from the model parameters to the recognition behavior and quantifies the model's evolvability and robustness. The method also identifies directions of sloppiness and stiffness and illustrates how the properties of the parameter map could shape circuit evolution. This approach of characterizing the evolvability and robustness in neural circuit models is applicable to a wide variety of circuits and systems.
- 2) Strategic positioning of visual cortical orientation columns (Luis Peters, Zoe Stawyskyj, Constantin Lührmann, Fred Wolf, University of Göttingen, CIDBN / MPI for Dynamics and Self-Organization Göttingen, Germany)
abstract…
A fundamental property of neurons in the primary visual cortex is their preference for orientation of edges. Recent studies further consolidated the finding that the spatial arrangement of orientation selective neurons in orientation domains and pinwheels is invariant across primates, carnivores and potentially even in some marsupials. No functional advantage favoring this universal common design is currently know. Here we address the question of whether the universal common design can provide a functional advantage over alternative designs. We formulate a mathematical theory formalizing the hypothesis that orientation domains are strategically positioned to provide the maximal amount of predictive information that can be obtained from a finite circular aperture onto an image and its complement. We represent input images as effective contour fields and build on the Gaussian information bottleneck theory to formulate a mathematically tractable predictive information utility function. The steepest asscent maximization of this utility function takes the form of a translation invariant ordered medium. Constrained to layouts with a typical spatial scale the optimal layouts differ strongly when varying the predictive information parameter and resemble observed orientation preference maps for strong and non-local predictive information content. Our results suggest that the experimentally observed universal common design of orientation domains and pinwheels strategically positions orientation columns to optimally perform spatial predictive coding.
- 3) A bottom-up approach to discriminate activity dependent and activity independent synaptic turnover (Aaron Nagel, University of Göttingen, CIDBN, Germany)
abstract…
Activity-dependent synaptic plasticity is widely believed to play the major role in learning and memory. Moreover, the robustness of memories depends on the stability of synapses. Recent studies, however, have shown that synapses exhibit significant volatility which to some extent appears activity- independent [1]. This stochastic turnover therefore can put a challenge on encoding and preserving information in synaptic connectivity [2]. In this study, using in vivo STED nanoscopy, to assess the dynamics of morphological features of excitatory spines in mouse cortical circuits including the dynamics of head size, neck length, and neck width measured over short (hour) and long (days) intervals for up to 30 days, monitoring their changes in time [3,4]. We model two scenarios where the change of spine morphological features was either activity dependent or completely spontaneous.
In the first scenario the synaptic changes solely depend on the timing of discrete ‘learning events’ following a Poisson distribution as the most generic case. In the second scenario we model the spine dynamics based on stochastic dynamics of actin filaments, with the synaptic changes are independent of pre and post synaptic activity.
Comparing theoretical predictions with measured cross-correlation functions of spine features we distinguish distinct roles of activity (in)dependent plasticity in governing synaptic turnover over short and long time intervals. Our results indicate that quenched disorder – the heterogeneity in the stable component of synaptic measures – is necessary to capture the non-vanishing part of cross correlation functions that our data reveal.
- 4) Resonant song recognition in crickets (Winston Mann, University of Göttingen, European Neuroscience Institute Göttingen, Germany)
abstract…
The function and evolution of calling song recognition networks is central to understanding speciation in insects. Evolutionary changes in song recognition networks can compel speciation, so understanding the commonalities in the mechanisms by which song recognition is achieved across species can help define a central or “mother” network architecture that is shared among closely related species in a group. Then, observed differences in phenotype across species can be related to changes within the shared network parameter space, to eventually understand how evolutionary mechanisms give rise to the diversity of observed behaviors. One instance of such a mother network for calling song recognition has been extensively characterized for the cricket species G. Bimaculatus. In crickets, females respond to certain features of male calling song, typically the interpulse period, within a limited range that defines a unimodal phenotype for calling song recognition. A computational model of this proposed network has been shown to recover all single-peaked recognition phenotypes known from crickets. This model consists of five neurons connected in a network with excitatory and inhibitory feedforward connections, adaptation, and linear filters. The novel, multi-peaked response of the Anurogryllus cricket presents a potential challenge to the mother network hypothesis. Here, we characterize this phenotype as resonant and examine the properties of such a resonance type alongside simplified mechanisms of song feature recognition with the aim of expanding the known capabilities of the mother network architecture, and exploring how resonance types may be selectively suppressed and recovered, allowing for fast transitions between tuning types in evolution.
- 5) Statistical Ensemble Analysis: A Comprehensive Investigation of Pattern Equivalence in Orientation Preference Maps (Michael Sternbach, Zoe Stawyskyj, Fred Wolf, University of Göttingen, CIDBN / MPI for Dynamics and Self-Organization Göttingen, Germany)
abstract…
Understanding pattern variation and establishing equivalence between patterns is important to many areas of neuroscience and is made significantly more complex when the full extent of the pattern space is unknown or where it is infinite. Current studies of such patterns therefore often rely on the comparison of low dimensional pattern features to simplify comparison. We present a more comprehensive approach to pattern comparison based on statistical ensembles of circular pattern patches. We apply this approach to orientation preference maps for which previous feature-based comparisons supported the hypothesis of evolutionary invariance. First, using entropy methods, we calculate the number of possible distinguishable patches and the scaling of this quantity with measurement noise. This then provides estimates for the amount of data required to achieve a saturated sample of patches, at a specified noise level. This method successfully distinguishes between ensembles of orientation preference maps obtained from the Moiré interference model and the long-range interaction model which generate noisy hexagonal and quasi-periodic patterns respectively. Our approach can be used for other biological problems can be described as ordered 2D media, determine the consistency of models with data and to describe pattern development in active media.
- 6) Stochastic model for the optimal fusion of social and sensory information in transparent interactions (Selma Kouaiche, University of Göttingen, CIDBN / German Primate Center Göttingen / MPI for Dynamics and Self-Organization Göttingen, Germany)
abstract…
We present a stochastic dynamics theory and analysis framework for studying a continuous time transparent visual game for pairs of agents we call continuous perceptual report (CPR) tasks. The CPR task is a pattern recognition task where subjects view a moving random dot pattern with a time-dependent veridical direction of motion and variable coherence. The task is transparent in that each subject receives information about the current estimate of the other. The difficulty of this type of task depends on the coherence and noise level of the stimulus, on the accessibility of information from the other subject, and on the dynamics with which the veridical direction changes. We construct a general kernel-based linear model that can both be used to simulate the behavior of ideal model subjects and to quantify the behavior of experimentally tested agents. We present a method for estimation of the kernels from measured time series, based on minimizing a cost function that was derived analytically, we split the data set into a training and test set, and we implemented an L1 regularization using the sklearn. linear_model.Lasso() python package. Synthetic time series perfectly recovers the underlying kernel. Even for realistic duration observations kernels are reproducible and correctly predicted as assessed by cross-validation. In addition, we derive an analytical solution to the conditional probability of one or two subjects' stimulus estimates. Using these methods and results we present an assessment of the data demands, and estimation power of this framework and use it to explore stimulus designs.
- 7) The dynamic gain function: Transient neuronal frequency dependence across species (Stefan Pommer, University of Göttingen, CIDBN / University Medical Center Göttingen, Dep. of Neuroanatomy, Germany)
abstract…
Meaningful physiological neuron characterizations need to be designed with the network function in mind, especially when looking for gradients across species. Conventional in vitro neuron characterizations use regular stimuli, i.e. steps or ramps. However, in network models and in vivo, neurons operate in a fluctuation-driven regime. Thus, the respective characteristics are best measured using a statistical quantification of responses to in vivo-like stochastic inputs with fluctuating amplitudes and correlation times. The dynamic gain function (DGF) captures neuronal properties for in-vivo-like operating conditions and reveals properties that are highly relevant for network function (but otherwise hard to assess). (Merino 2021, Zhang 2024). The DGF captures the frequency preference of neurons and their ability to tune-in to rhythmic activity in recurrent networks. Here we employed this tool to compare the features of DGFs in cortical neurons across „Euarchontoglires“: rodents, lemur, monkeys and humans. We found qualitatively and quantitatively similar features across species: 1) The bandwidth of faithfully encoded frequencies was similar, with cut-off frequencies around 400 Hz. High-bandwidth, ultrafast encoding seems to be a general feature and not exclusive to primates. 2) The frequency preference adaptability was almost universal. When slow correlations appear in the input, neurons improve the encoding of high-frequency input components, sometimes remarkably so (Merino et al. 2021). However, we observed a notable exception to this rule in our most diverse dataset (Callithrix jacchus), with a few neurons that showed near constant DGF. In conclusion, this frequency-dependency of neurons could represent correlates of higher brain functions. However, the unique characteristics of higher-order primates seem to be not reflected in a broader functionality of the axo-somatic information processing of individual neurons.
- 8) Contrast-invariant orientation selectivity in a synthetic biology model of the early visual pathway (Julian Vogel, University of Göttingen, CIDBN, Germany)
abstract…
A fundamental property of neurons in the primary visual cortex is their preference for orientation of edges that is invariant under contrast changes. We designed a synthetic hybrid neural circuit to study the emergence of orientation selectivity under different thalamo-cortical connection schemes. To this end, a computational model of the retino-thalamic pathway was combined with an in-vitro model of cortical input layer 4. The latter was either a primary culture of cortical neurons or an acute brain slice of primary visual cortex. The two stages were interfaced optogenetically by expression of channelrhodopsin in the neurons of the in-vitro circuit and holographic stimulation. Neural activity was then monitored either electrophysiologically with multielectrode arrays or by fluorescent imaging. In our model we implemented a connection scheme with thalamocortical input that is retinotopic but unselective for orientation. Interestingly, we measured orientation selective responses in the cortical activity that we reasoned had to be generated intrinsically by the target network. Next, we tested the effect of different contrast levels and found that the orientation selectivity can change for individual neurons but is stable on the population level. In summary, we showed that contrast-invariant orientation preference can emerge from unspecific thalamo-cortical input.