Konrad Paul Kording

97% of 184 high-impact biomedical AI studies compare cross-validated models with tests assuming fold independence; in 69% of re-analyzable studies a significant comparison does not survive correction.

The paper maps physical cell-cell contacts in mouse visual cortex, revealing a dense contactome that exceeds synaptic connectivity and highlights the role of astrocytes in shaping the neural microenvironment.

This paper introduces a bidirectional recurrent model of attention inspired by biological mechanisms and neural networks, capable of replicating behavioral and neural findings related to visual attention.

This study shows that mid-superior temporal sulcus (mSTS) neurons in macaques encode spatial context and behavioral state during natural, unconstrained exploration.

This study shows humans can implicitly learn and use bimodal priors in complex sensorimotor tasks like returning tennis serves, influenced by visual uncertainty and biomechanical constraints.

This paper examines limitations in interpreting resting-state fMRI correlations, distinguishing measurement distortion from causal non-identifiability and their implications for biological interpretation.

260,000+ primate images from 60+ genera with face boxes and 48/68-point landmarks; detectors and landmark models trained on them work across primates and approach human-trained baselines on human faces.

This paper introduces a free, browser-based tool for automated segmentation of synchrotron-scanned fossils, significantly reducing time and cost compared to manual methods.

The paper investigates how uncertainty, relevance, and learning interact in auditory categorization by modeling human behavior with a Bayesian framework.

The paper studies how individual learning strategies during category learning affect category boundaries in mice.

The paper shows that conclusions about neural network similarity to brain alignment depend heavily on the chosen similarity measure.

The paper demonstrates how causal inference methods like instrumental variables and difference in differences can reduce confounding in optogenetic experiments to estimate effective neural connectivity.

This study identifies neural signatures of natural social behaviors in rhesus macaques using wireless recordings and ethological analysis, revealing how social interactions are encoded in brain activity.

The paper demonstrates that neural activity patterns may not reliably indicate causal contributions of neural units to behavior and introduces a lesioning framework to better infer causal relationships.

The paper shows that gradient descent learning in neural networks creates representations sensitive to common image features, mirroring efficient coding principles observed in animal sensory systems.

The paper shows that mitotic lineage improves predictions of neural connections in C. elegans beyond cell type alone.

This paper tests Bayesian models of batter behavior in professional baseball using a million samples, showing estimates are biased towards priors or likelihoods depending on uncertainty.

The paper investigates whether energy expenditure explains step length-width preferences during walking and finds it is not the sole factor.

This paper develops a DNA-based recording system using TdT to encode environmental signals with minutes resolution.

The paper provides a three-dimensional X-ray microtomography dataset of thalamocortical structures to study neural heterogeneity.

The paper compares linear and nonlinear model performance on brain imaging data and standard datasets, finding linear models outperform nonlinear ones in brain scans despite larger sample sizes.

This paper demonstrates that neuronal spiking allows for unbiased causal influence estimation and gradient descent approximation, unaffected by upstream activity or downstream non-linearities.

The paper investigates how the brain represents preliminary and definitive action plans during natural scene search, identifying two classes of neurons in the frontal eye field.

This paper introduces TURTLES, a DNA-based method to record temporal environmental signals with high resolution, achieving 2-minute accuracy over 60 minutes.

The paper investigates whether learned spatial priors generalize from a sensorimotor task to a perceptual task, finding that priors differ significantly between task modalities.

The paper develops a scalable, reproducible framework for processing large neuroimaging datasets using open-source tools and standardized pipelines.

This study tests whether Bayesian computation is inherent in sensorimotor estimation by comparing children and adults in a position estimation task, finding that Bayesian behavior develops over time.

The paper introduces Linear-Nonlinear-Time-Warp-Poisson models to account for shared temporal variability in neural activity, showing improved predictions in macaque premotor cortex recordings.

This study compares GLMs to machine learning methods for predicting neural spikes and finds that modern methods like XGBoost and ensembles are more accurate and robust.

The paper studies corticocortical signaling in anesthetized mice using optogenetics and electrophysiology, finding rapid and scalable downstream activity driven by upstream activity.