3 results match your criteria: "The Netherlands. e.formisano@maastrichtuniversity.nl.[Affiliation]"
Bridging auditory perception and natural language processing with semantically informed deep neural networks.
Sci Rep
September 2024
Department of Cognitive Neuroscience, Faculty of Psychology and Neuroscience, Maastricht University, Maastricht, The Netherlands.
Article Synopsis
- - Human sound recognition is intuitive, while artificial systems struggle; recent advancements using deep neural networks (DNNs), particularly convolutional neural networks (CNNs), have improved sound classification but often ignore the relationships between sound labels.
- - The researchers hypothesize that adding semantic information to DNNs can enhance sound recognition, mimicking how humans use both acoustic and semantic cues; they framed the task as a regression problem, training models to link sound spectrograms to continuous semantic representations.
- - Their findings show that the DNN model utilizing semantic labels (semDNN) outperformed the traditional label model (catDNN), aligning more closely with human similarity ratings in sound recognition, thus highlighting the importance of semantics in improving artificial
Intermediate acoustic-to-semantic representations link behavioral and neural responses to natural sounds.
Nat Neurosci
April 2023
Department of Cognitive Neuroscience, Faculty of Psychology and Neuroscience, Maastricht University, Maastricht, the Netherlands.
Article Synopsis
- The study investigates how the brain transforms sound waves into meaningful representations, focusing on the superior temporal gyrus (STG) and its role in processing auditory information.
- Researchers used a model comparison approach to analyze different representations of sound and found that certain auditory features predict responses in the brain's auditory cortex.
- The results reveal that sound-to-event neural networks are more effective than traditional models in predicting responses in the STG and perceived sound differences, highlighting a complex intermediate level of sound representation.
Seeing patterns through the hemodynamic veil--the future of pattern-information fMRI.
Neuroimage
August 2012
Department of Cognitive Neuroscience, Faculty of Psychology and Neuroscience, Maastricht University, The Netherlands.
Pattern-information fMRI (pi-fMRI) has become a popular method in neuroscience. The technique is motivated by the idea that spatial patterns of fMRI activity reflect the neuronal population codes of perception, cognition, and action. In this commentary, we discuss three fundamental outstanding questions: (1) What is the relationship between neuronal patterns and fMRI patterns? (2) Does pattern-information fMRI benefit from hyperacuity, enabling the investigation of columnar-level neuronal information, even at low resolution? (3) Do high-resolution and high-field fMRI increase sensitivity to pattern information? The empirical answers will enable us to optimize pi-fMRI data acquisition and to understand the ultimate potential and appropriate interpretation of pi-fMRI results.
View Article and Find Full Text PDF