A5029714016- Neural dynamics and brain function
- Single-cell and spatial transcriptomics
- Adversarial Robustness in Machine Learning
- Cell Image Analysis Techniques
- Advanced Fluorescence Microscopy Techniques
- Explainable Artificial Intelligence (XAI)
- Retinal Development and Disorders
- Advanced Neural Network Applications
- Neural Networks and Applications
- Parkinson's Disease Mechanisms and Treatments
- Gene expression and cancer classification
- Functional Brain Connectivity Studies
- Advanced MRI Techniques and Applications
- Advanced Memory and Neural Computing
- Medical Image Segmentation Techniques
- Neuroscience and Neuropharmacology Research
- Image and Signal Denoising Methods
- Neurological disorders and treatments
- Neuroscience and Neural Engineering
- Advanced NMR Techniques and Applications
- Image Processing Techniques and Applications
- Cerebral Palsy and Movement Disorders
- Time Series Analysis and Forecasting
- Blind Source Separation Techniques
- Medical Imaging Techniques and Applications
University of California, San Francisco
2017-2025
Center for Neurosciences
2022-2023
Universidad Católica de Santa Fe
2023
University of California, Berkeley
2016-2021
Allen Institute for Brain Science
2018-2020
Allen Institute
2018-2020
Flatiron Institute
2016
Flatiron Health (United States)
2016
Sharif University of Technology
2011-2012
Amirkabir University of Technology
2010
Machine-learning models have demonstrated great success in learning complex patterns that enable them to make predictions about unobserved data. In addition using for prediction, the ability interpret what a model has learned is receiving an increasing amount of attention. However, this increased focus led considerable confusion notion interpretability. particular, it unclear how wide array proposed interpretation methods are related and common concepts can be used evaluate them. We aim...
Despite advances in experimental techniques and accumulation of large datasets concerning the composition properties cortex, quantitative modeling cortical circuits under in-vivo-like conditions remains challenging. Here we report publicly release a biophysically detailed circuit model layer 4 mouse primary visual receiving thalamo-cortical inputs. The 45,000-neuron was subjected to battery stimuli, results were compared published work new vivo experiments. Simulations reproduced variety...
Abstract Two-photon fluorescence microscopy has been used extensively to probe the structure and functions of cells in living biological tissue. excitation generates from focal plane, but also outside with out-of-focus increasing as focus is pushed deeper into It postulated that two-photon depth limit, beyond which results become inaccurate, where in-focus are equal, we term balance depth. Calculations suggest should be at ∼600 µm mouse cortex. Neither limit nor have measured brain We found...
Abstract In Parkinson’s disease, imbalances between ‘antikinetic’ and ‘prokinetic’ patterns of neuronal oscillatory activity are related to motor dysfunction. Invasive brain recordings from the network have suggested that medical or surgical therapy can promote a prokinetic state by inducing narrowband gamma rhythms (65–90 Hz). Excessive in cortex promotes dyskinesia rodent models, but relationship humans has not been well established. To assess this relationship, we used sensing-enabled...
Highlights•Cortical gamma oscillations often entrain with pallidal and subthalamic stimulation•Entraining levodopa-induced diminishes their prodyskinetic effect•Levodopa-induced oscillation peak frequency variance reduces when entrainedAbstractBackgroundIn Parkinson's disease, invasive brain recordings show that dopaminergic medication can induce narrowband rhythms in the motor cortex nucleus, which co-fluctuate dyskinesia scores. Deep stimulation these to a subharmonic frequency. However,...
Existing regulatory frameworks for biomedical AI include robustness as a key component but lack detailed implementational guidance. The recent rise of foundation models creates new hurdles in testing and certification given their broad capabilities susceptibility to complex distribution shifts. To balance test feasibility effectiveness, we suggest priority-based, task-oriented approach tailor evaluation objectives predefined specification. We urge concrete policies adopt granular...
Voxelwise encoding models based on convolutional neural networks (CNNs) have emerged as state-of-the-art predictive of brain activity evoked by natural movies. Despite their superior performance, the huge number parameters in CNN-based made them difficult to interpret. Here, we investigate whether model compression can build more interpretable and stable voxelwise while maintaining accuracy. We used multiple techniques prune less important CNN filters connections, a receptive field method...
Abstract Deep neural network models have recently been shown to be effective in predicting single neuron responses primate visual cortex areas V4. Despite their high predictive accuracy, these are generally difficult interpret. This limits applicability characterizing V4 function. Here, we propose the DeepTune framework as a way elicit interpretations of deep network-based neurons area is midtier cortical ventral pathway. Its functional role not yet well understood. Using dataset recordings...
Abstract Quantification of motor symptom progression in Parkinson’s disease (PD) patients is crucial for assessing and optimizing therapeutic interventions, such as dopaminergic medications deep brain stimulation. Cumulative heuristic clinical experience has identified various signs associated with PD severity, but these are neither objectively quantifiable nor robustly validated. Video-based objective quantification enabled by machine learning (ML) introduces a potential solution. However,...
The rapid growth of large-scale spatial gene expression data demands efficient and reliable computational tools to extract major trends in their native context. Here, we used stability-driven unsupervised learning (i.e., staNMF) identify principal patterns (PPs) 3D profiles understand distribution anatomical localization at the whole mouse brain level. Our subsequent correlation analysis systematically compared PPs known regions ontology from Allen Mouse Brain Atlas using neighborhoods. We...
Deep convolutional neural networks (CNNs) have been successful in many tasks machine vision, however, millions of weights the form thousands filters CNNs makes them difficult for human intepretation or understanding science. In this article, we introduce CAR, a greedy structural compression scheme to obtain smaller and more interpretable CNNs, while achieving close original accuracy. The is based on pruning with least contribution classification We demonstrate interpretability CAR-compressed...
A bstract Structural rules underlying functional properties of cortical circuits are poorly understood. To explore these systematically, we integrated information from extensive literature curation and large-scale experimental surveys into a data-driven, biologically realistic model the mouse primary visual cortex. The was constructed at two levels granularity, using either biophysically-detailed or point-neurons, with identical network connectivity. Both variants were compared to each other...
A growing body of work shows that autonomic signals provide a privileged evidence-stream to capture various aspects subjective and neural states. This investigates the potential for markers track effects psychedelics — potent psychoactive drugs with important scientific clinical value. For this purpose, we introduce novel Bayesian framework estimate entropy heart rate dynamics under psychedelics. We also calculate estimates mean variability, investigate how these measures relate reports...
Structural rules underlying functional properties of cortical circuits are poorly understood. To explore these systematically, we integrated information from extensive literature curation and large-scale experimental surveys into data-driven, biologically realistic models the mouse primary visual cortex. The were constructed at two levels granularity, using either biophysically-detailed or point-neuron models, with identical network connectivity. Both compared to each other recordings neural...
Deep convolutional neural networks (CNNs) have been successful in many tasks machine vision, however, millions of weights the form thousands filters CNNs make them difficult for human interpretation or understanding science. In this article, we introduce a greedy structural compression scheme to obtain smaller and more interpretable CNNs, while achieving close original accuracy. The is based on pruning with least contribution classification accuracy lowest Classification Accuracy Reduction...
Convolutional neural networks (CNNs) achieve state-of-the-art performance in a wide variety of tasks computer vision. However, interpreting CNNs still remains challenge. This is mainly due to the large number parameters these networks. Here, we investigate role compression and particularly pruning filters interpretation CNNs. We exploit our recently-proposed greedy structural scheme that prunes trained CNN. In compression, filter importance index defined as classification accuracy reduction...
<h3>Objective:</h3> Determine the relationship of diurnal step count patterns with disability in progressive multiple sclerosis (MS). <h3>Background:</h3> Worsening Expanded Disability Status Scale (EDSS) scores and longer Timed 25-Foot Walk (T25FW) times generally correspond to increased MS correlate lower daily physical activity. Patterns activity progression are understudied. <h3>Design/Methods:</h3> Participants (n=577) SPI2 study MD1003 (high dose biotin) underwent EDSS T25FW every...
Recognition and correction of inhomogeneous displacement caused by patient's movement has been recently discussed as an interesting topic in medical image processing. Considering consistency general structure the during distortion, histogram could be employed a fast implementation method feature domain. Accordingly, attribute vectors defined for each pixel based on spatial features to find corresponding points two images. Consequently point-based non-rigid transformation approach will...
We study the performance of brain computer interface (BCI) systems in a virtual reality (VR) environments and compare it to 2D regular displays. First, we design headset that consists three components: wearable electroencephalography (EEG) device, VR interface. Recordings behavior from human subjects, performing wide variety tasks using our device are collected. The consist object rotation or scaling either mental commands facial expression (smile eyebrow movement). Subjects asked repeat...