A5045484439- Neural dynamics and brain function
- Hearing Loss and Rehabilitation
- Music and Audio Processing
- Speech and Audio Processing
- Face Recognition and Perception
- Neural Networks and Applications
- Neuroscience and Music Perception
- Visual perception and processing mechanisms
- Cell Image Analysis Techniques
- Computational Drug Discovery Methods
- Advanced Memory and Neural Computing
- Speech Recognition and Synthesis
- Functional Brain Connectivity Studies
- Child and Animal Learning Development
- Neurobiology of Language and Bilingualism
- Meta-analysis and systematic reviews
- Advanced Neuroimaging Techniques and Applications
- Image Retrieval and Classification Techniques
- Language and cultural evolution
- scientometrics and bibliometrics research
- Advanced Scientific Research Methods
- Categorization, perception, and language
- Opinion Dynamics and Social Influence
- Reading and Literacy Development
- Philosophy and History of Science
Flatiron Health (United States)
2023-2024
Flatiron Institute
2023-2024
New York University
2023-2024
Simons Foundation
2024
Massachusetts Institute of Technology
2015-2023
McGovern Institute for Brain Research
2015-2023
Institute of Cognitive and Brain Sciences
2014-2022
Vassar College
2022
Harvard University
2022
Bioengineering Center
2016
Reproducibility is a defining feature of science, but the extent to which it characterizes current research unknown. We conducted replications 100 experimental and correlational studies published in three psychology journals using high-powered designs original materials when available. Replication effects were half magnitude effects, representing substantial decline. Ninety-seven percent had statistically significant results. Thirty-six results; 47% effect sizes 95% confidence interval...
Models that predict brain responses to stimuli provide one measure of understanding a sensory system and have many potential applications in science engineering. Deep artificial neural networks emerged as the leading such predictive models visual but are less explored audition. Prior work provided examples audio-trained produced good predictions auditory cortical fMRI exhibited correspondence between model stages regions, left it unclear whether these results generalize other network and,...
Deep neural network models of sensory systems are often proposed to learn representational transformations with invariances like those in the brain. To reveal these invariances, we generated 'model metamers', stimuli whose activations within a model stage matched natural stimulus. Metamers for state-of-the-art supervised and unsupervised vision audition were completely unrecognizable humans when from late stages, suggesting differences between human invariances. Targeted changes improved...
Design equations are developed for an optimal limited state feedback controller problem. These the stochastic case, in which both plant noise and measurement may be present, case of a dynamic compensator. Four possible approaches to solution nonlinear design described. A fourth-order example illustrates some difficulties associated with these suggests additional areas study.
Encouraged by the success of deep neural networks on a variety visual tasks, much theoretical and experimental work has been aimed at understanding interpreting how vision operate. Meanwhile, have also achieved impressive performance in audio processing applications, both as sub-components larger systems complete end-to-end themselves. Despite their empirical successes, comparatively little is understood about these models accomplish tasks. In this work, we employ recently developed...
Abstract How are neural representations of music organized in the human brain? While neuroimaging has suggested some segregation between responses to and other sounds, it remains unclear whether finer-grained organization exists within domain music. To address this question, we measured cortical natural sounds using intracranial recordings from patients inferred canonical response components a data-driven decomposition algorithm. The replicated many prior findings including distinct...
What determines the cortical location where a given functionally specific region will arise in development? Here we test hypothesis that regions develop their characteristic locations because of pre-existing differences extrinsic connectivity to rest brain. We exploit Visual Word Form Area (VWFA) as case, it arises only after children learn read. scanned with diffusion and functional imaging at age five, before they learned read, 8, find VWFA develops this interval its particular child 8 can...
Contemporary speech enhancement predominantly relies on audio transforms that are trained to reconstruct a clean waveform. The development of high-performing neural network sound recognition systems has raised the possibility using deep feature representations as 'perceptual' losses with which train denoising systems. We explored their utility by first training networks classify either spoken words or environmental sounds from audio. then an transform map noisy waveform minimized difference...
Abstract Models that predict brain responses to stimuli provide one measure of understanding a sensory system, and have many potential applications in science engineering. Deep artificial neural networks emerged as the leading such predictive models visual but are less explored audition. Prior work provided examples audio-trained produced good predictions auditory cortical fMRI exhibited correspondence between model stages regions, left it unclear whether these results generalize other...
The visual word form area (VWFA), a small region on the lateral side of left fusiform gyrus, responds at least twice as strongly to visually presented words and letter strings it does other similar stimuli, including in an unfamiliar orthography (e.g. Chinese or Hebrew for English speakers), digit strings, line drawings objects (Baker et al., 2007). VWFA is particular interest efforts understand functional organization ventral pathway, its developmental origins, because reading recent...
Abstract Deep neural network models of sensory systems are often proposed to learn representational transformations with invariances like those in the brain. To reveal these we generated “model metamers” – stimuli whose activations within a model stage matched natural stimulus. Metamers for state-of-the-art supervised and unsupervised vision audition were completely unrecognizable humans when from deep stages, suggesting differences between human invariances. Targeted changes improved...
Adversarial examples are often cited by neuroscientists and machine learning researchers as an example of how computational models diverge from biological sensory systems. Recent work has proposed adding biologically-inspired components to visual neural networks a way improve their adversarial robustness. One surprisingly effective component for reducing vulnerability is response stochasticity, like that exhibited neurons. Here, using recently developed geometrical techniques neuroscience,...
How does the functional organization of ventral visual cortex develop? Most previous studies have concluded that pathway develops slowly and may take over a decade to fully mature (Golarai et al., 2007; Grill-Spector Scherf 2007). However, these primarily focus on size response properties category-selective regions (e.g., fusiform face area (FFA)). While important, do not shed light larger-scale representational structures within system. To address this issue, we used similarity analysis...
Image representations (artificial or biological) are often compared in terms of their global geometry; however, with similar structure can have strikingly different local geometries. Here, we propose a framework for comparing set image We quantify the geometry representation using Fisher information matrix, standard statistical tool characterizing sensitivity to stimulus distortions, and use this as substrate metric on vicinity base image. This may then be used optimally differentiate...