A5052151101- Neural dynamics and brain function
- Visual perception and processing mechanisms
- Domain Adaptation and Few-Shot Learning
- Traffic control and management
- Retinal Development and Disorders
- Transportation Planning and Optimization
- Multimodal Machine Learning Applications
- Complex Network Analysis Techniques
- Face Recognition and Perception
- Neurobiology and Insect Physiology Research
- Model Reduction and Neural Networks
- Advanced Image and Video Retrieval Techniques
- Image Retrieval and Classification Techniques
- Genomics and Phylogenetic Studies
- Neuroscience and Neuropharmacology Research
- Advanced Fluorescence Microscopy Techniques
- Traffic Prediction and Management Techniques
University of Tübingen
2019-2024
University of Göttingen
2021-2024
Max Planck Institute for Intelligent Systems
2022-2024
Bernstein Center for Computational Neuroscience Tübingen
2021
Divisive normalization (DN) is a prominent computational building block in the brain that has been proposed as canonical cortical operation. Numerous experimental studies have verified its importance for capturing nonlinear neural response properties to simple, artificial stimuli, and suggest DN also an important component processing natural stimuli. However, we lack quantitative models of are directly informed by measurements spiking responses applicable arbitrary Here, propose model input...
The retina encodes a broad range of stimuli, adapting its computations to features like brightness, contrast, or motion. However, it is unclear what extent also adapts spatial frequency content - as theories efficient coding would predict for instance, when switching between natural scenes and white noise. To address this, we analyzed neural activity marmoset retinal ganglion cells (RGCs) in response noise naturalistic movie stimuli. We trained linear-nonlinear models on both evaluated their...
Many approaches have been proposed to use diffusion models augment training datasets for downstream tasks, such as classification. However, are themselves trained on large datasets, often with noisy annotations, and it remains an open question which extent these contribute classification performance. In particular, unclear if they generalize enough improve over directly using the additional data of their pre-training process augmentation. We systematically evaluate a range existing methods...
Abstract The diverse nature of visual environments demands that the retina, first stage system, encodes a vast range stimuli with various statistics. retina adapts its computations to some specific features input, such as brightness, contrast or motion. However, it is less clear whether also statistics natural scenes compared white noise, latter which often used infer models retinal computation. To address this question, we analyzed neural activity ganglion cells (RGCs) in response both noise...
More than a dozen excitatory cell types have been identified in the mouse primary visual cortex (V1) based on transcriptomic, morphological and vitro electrophysiological features. However, functional landscape of neurons with respect to their responses stimuli is currently unknown. Here, we combined large-scale two-photon imaging deep learning neural predictive models study organization V1 using digital twins. Digital twins enable exhaustive silico characterization providing bar code...
The neural underpinning of the biological visual system is challenging to study experimentally, in particular as neuronal activity becomes increasingly nonlinear with respect input. Artificial networks (ANNs) can serve a variety goals for improving our understanding this complex system, not only serving predictive digital twins sensory cortex novel hypothesis generation silico, but also incorporating bio-inspired architectural motifs progressively bridge gap between and machine vision. mouse...
Abstract Divisive normalization (DN) is a prominent computational building block in the brain that has been proposed as canonical cortical operation. Numerous experimental studies have verified its importance for capturing nonlinear neural response properties to simple, artificial stimuli, and suggest DN also an important component processing natural stimuli. However, we lack quantitative models of are directly informed by measurements spiking responses applicable arbitrary Here, propose...
Identifying cell types and understanding their functional properties is crucial for unraveling the mechanisms underlying perception cognition. In retina, can be identified by carefully selected stimuli, but this requires expert domain knowledge biases procedure towards previously known types. visual cortex, it still unknown what exist how to identify them. Thus, unbiased identification of in retina new approaches are needed. Here we propose an optimization-based clustering approach using...
Deep predictive models of neuronal activity have recently enabled several new discoveries about the selectivity and invariance neurons in visual cortex. These learn a shared set nonlinear basis functions, which are linearly combined via learned weight vector to represent neuron's function. Such vectors, can be thought as embeddings function, been proposed define functional cell types unsupervised clustering. However, deep usually highly overparameterized, learning problem is unlikely unique...
The capacity of a street segment quantifies the maximal density vehicles before congestion arises. Here we show in simple mathematical model that fluctuations instantaneous number entering are sufficient to induce persistent congestion. Congestion emerges even if average flow is below segment’s where absent without fluctuations. We explain how this fluctuation-induced due self-amplifying reduction vehicle velocities.
Although routing applications increasingly affect individual mobility choices, their impact on collective traffic dynamics remains largely unknown. Smart communication technologies provide accurate data for choosing one route over other alternatives; yet, inherent delays undermine the potential usefulness of such information. Here, we introduce and analyze a simple model dynamics, which results from choice relying outdated We find sufficiently small information that flows are stable against...
Knowledge distillation (KD) is a simple and successful method to transfer knowledge from teacher student model solely based on functional activity. However, current KD has few shortcomings: it recently been shown that this unsuitable inductive biases like shift equivariance, struggles out of domain generalization, optimization time magnitudes longer compared default non-KD training. To improve these aspects KD, we propose Hard Augmentations for Robust Distillation (HARD), generally...
Recent advances in the development of vision-language models (VLMs) are yielding remarkable success recognizing visual semantic content, including impressive instances compositional image understanding. Here, we introduce novel task Visual Data-Type Identification, a basic perceptual skill with implications for data curation (e.g., noisy data-removal from large datasets, domain-specific retrieval) and autonomous vision distinguishing changing weather conditions camera lens staining). We...