Marcel A. J. van Gerven

This study shows brain alignment in deep neural networks is present early and independent of object classification, suggesting it reflects environmental structure rather than classification strategy.

This study uses Bayesian optimization to find optimal stimulation patterns for cortical neuroprosthetics, improving phosphene perception quality through patient feedback.

This study develops an end-to-end framework to learn safe stimulation parameters for cortical neuroprosthetic vision, improving phosphene interpretability while maintaining image reconstruction performance.

This paper models optimal foraging using a dynamical system framework, showing adaptive foraging behavior and similarities to neural mechanisms in non-human primates.

This paper introduces an artificial neural network with oscillatory dynamics to convert simultaneous inputs into a temporal code, improving classification of competing stimuli.

This paper introduces PAM, a predictive attention mechanism for neural decoding of visual perception, achieving state-of-the-art reconstructions and providing interpretative insights through attention weights and values.

This paper shows that feature-disentangled GAN latents better predict neural responses to visual stimuli than other representations, and can reconstruct visual perception with high fidelity.

This paper presents a differentiable phosphene simulator for optimizing visual cortical prostheses using biologically plausible models of visual perception.

This paper optimizes deep learning models for speech reconstruction from sensorimotor brain activity, achieving high word decoding accuracy and intelligible speech.

This paper shows that predictive coding in neural networks arises from energy efficiency rather than requiring hard-wired hierarchical connectivity.

The paper introduces Brain2Pix, a method to reconstruct naturalistic videos from brain activity using fully convolutional networks and retinotopic mappings, showing significant improvement over existing techniques.

PhenoScore quantifies phenotypic similarity using AI and HPO data to improve genetic variant interpretation and genotype-phenotype studies.

The paper investigates how neural states are organized across the cortical hierarchy and their relation to perceived event boundaries in the human brain.

This paper compares three facial feature extraction methods for diagnosing genetic disorders, finding that VGGFace2 provides the best performance with 0.78 accuracy.

DeNovoCNN is a deep learning tool for de novo variant calling in sequencing data, achieving high recall and precision compared to existing methods.

This paper reconstructs faces from fMRI data using GAN latent spaces, achieving accurate perception reconstructions by aligning neural responses with generator features.

This paper introduces methods to detect neural state transitions linked to event segmentation and shows they outperform existing methods.

The paper introduces a framework for system identification in neuroscience using cortical information flow, demonstrating its ability to estimate meaningful neural computations from fMRI data.

This study uses a neural network model to show how neurons represent and incorporate prior knowledge through environmental regularities, without probabilistic feedback.

The paper provides direct evidence of reversed visual hierarchy activation during mental imagery and reports an 11Hz oscillatory pattern during perception.

This study introduces a recurrent neural network with adaptive time scales, showing improved performance in temporal prediction and memory capacity.

This paper uses deep learning to classify mouse ultrasonic vocalizations by sex, achieving 84% accuracy compared to 64% and 52% with other methods.

This paper introduces a recurrent generative model that mimics human visual attention by minimizing predictive uncertainty through saccadic eye movements.

DeepRF uses deep learning to rapidly estimate population receptive fields, outperforming traditional methods in speed and accuracy.

The paper investigates how mental imagery influences conscious perception by analyzing individual differences in the interaction between imagery and sensory input using a Bayesian hierarchical model.

This paper uses generative adversarial networks to reconstruct natural images from brain activity, achieving 67.2% and 66.4% identification rates in behavioral tests.

The paper introduces two nonparametric Bayesian methods for estimating spike-spike and spike-membrane neuronal connectivity using Gaussian process regression and validates them on simulated and real data.

This study examines the temporal dynamics of visual imagery versus perception, finding distinct patterns in the timing of neural activity.

This paper reviews computational principles for understanding natural intelligence and discusses neural network models for cognitive processes, while outlining remaining challenges for achieving human-like machine intelligence.