Matthias Bethge

openretina provides a standardized framework for training and evaluating retinal models across datasets and species, revealing unexplained variance in current models.

The paper identifies a retinal ganglion cell type that detects visual context changes by responding selectively to chromatic contrast transitions.

The paper investigates how macaque V4 neurons respond to natural images using features from CNNs trained on various tasks, finding V4 is best predicted by semantic features and shows task selectivity.

The paper uses synthetic data to compare two tasks in proprioception modeling, finding that action recognition better explains neural tuning properties than trajectory decoding.

This study improves neural system identification by incorporating efficient coding principles, showing better performance and biologically plausible filters in retinal neuron response prediction.

This study shows that robust deep learning models for object recognition rely on low spatial frequency information, as evidenced by their performance on adversarial attacks and common corruptions.

The paper shows that semantically inconsistent objects attract more fixations than consistent ones, but contextualized meaning maps fail to capture this relationship.

The paper uses digital twins to characterize the functional organization of mouse V1, revealing a combinatorial code of nonlinear response properties across 30 modes.

The paper develops a divisive normalization model informed by V1 neuron responses to natural images, showing orientation-specific normalization within receptive fields.

This study compares meaning maps and saliency models in predicting human fixations, finding that deep neural networks outperform meaning maps in accuracy and responsiveness to meaning changes.

The paper develops a biophysically constrained computational model of inner retinal circuits that performs well on new stimuli and matches connectomics data.

The paper develops a method to image the entire inner plexiform layer of the retina at once, revealing response speed gradients in bipolar cells and isolating experimental batch effects.

The paper updates and provides a step-by-step guide for using DeepLabCut, an open-source toolbox for markerless 3D pose estimation across species and behaviors.

The paper shows that real and model-generated images can be distinguished at V2 scaling, challenging the interpretation of Bouma’s Law in crowding.

This paper uses deep convolutional neural networks to predict spiking activity in primate V1 and finds they outperform traditional models.

The paper shows that moving bars are processed faster than flashed bars in macaque V1, supporting the differential latency model of the flash lag illusion.

This study investigates how attentional fluctuations affect correlated variability in macaque primary visual cortex, finding significant effects at both long and short timescales.

The paper benchmarks spike rate inference algorithms for two-photon calcium imaging data and finds that top-performing algorithms show high correlation despite diverse approaches.

The paper compares texture models using deep CNN features and finds that while CNN models perform well under foveal inspection, no model perfectly matches all textures across viewing conditions.

The paper investigates how visual crowding affects the ability to detect spatial distortions in peripherally presented letters, finding that crowding impairs distortion detection and alters distortion sensitivity tuning.

The paper models how mice detect odors in mixtures using a linear decoder with sparse weights, finding that they generalize poorly without demixing.

The paper benchmarks spike rate inference algorithms using calcium imaging data and introduces a supervised learning method that outperforms existing techniques across diverse datasets.

The paper shows that balanced excitatory and inhibitory inputs in bipolar cells decorrelate visual feature representation in the mammalian inner retina.

The paper analyzes how attentional state fluctuations affect population neural responses in a model where attention modulates neuron gain, showing that attention increases response gain and reduces variability without altering correlation structure.