A5088549773- Neural dynamics and brain function
- Zebrafish Biomedical Research Applications
- Functional Brain Connectivity Studies
- Human Pose and Action Recognition
- Cell Image Analysis Techniques
- Photoreceptor and optogenetics research
- Bat Biology and Ecology Studies
- Adversarial Robustness in Machine Learning
- Advanced Memory and Neural Computing
- Advanced Fluorescence Microscopy Techniques
- Explainable Artificial Intelligence (XAI)
- CCD and CMOS Imaging Sensors
- Animal Behavior and Welfare Studies
- Wildlife Ecology and Conservation
- Anomaly Detection Techniques and Applications
- Animal Behavior and Reproduction
- Video Surveillance and Tracking Methods
- Neuroendocrine regulation and behavior
- Advanced Vision and Imaging
- Human Motion and Animation
- Artificial Intelligence in Healthcare and Education
- Neuroscience and Neuropharmacology Research
- Educational and Psychological Assessments
- Neuroinflammation and Neurodegeneration Mechanisms
- Neuroscience, Education and Cognitive Function
Columbia University
2018-2024
Columbia University Irving Medical Center
2022
Allen Institute for Brain Science
2020-2021
Center for Theoretical Physics
2020
Recent neuroscience studies demonstrate that a deeper understanding of brain function requires behavior. Detailed behavioral measurements are now often collected using video cameras, resulting in an increased need for computer vision algorithms extract useful information from data. Here we introduce new analysis tool combines the output supervised pose estimation (e.g. DeepLabCut) with unsupervised dimensionality reduction methods to produce interpretable, low-dimensional representations...
Calcium imaging has revolutionized systems neuroscience, providing the ability to image large neural populations with single-cell resolution. The resulting datasets are quite (with scales of TB/hour in some cases), which presented a barrier routine open sharing this data, slowing progress reproducible research. State art methods for analyzing data based on non-negative matrix factorization (NMF); these approaches solve non-convex optimization problem, and highly effective when good...
Abstract Noninvasive behavioral tracking of animals is crucial for many scientific investigations. Recent transfer learning approaches have considerably advanced the state art. Typically these methods treat each video frame and object to be tracked independently. In this work, we improve on (particularly in regime few training labels) by leveraging rich spatiotemporal structures pervasive — specifically, spatial statistics imposed physical constraints (e.g., paw elbow distance), temporal...
Abstract A major goal of computational neuroscience is the development powerful data analyses that operate on large datasets. These form an essential toolset to derive scientific insights from new experiments. Unfortunately, a obstacle currently impedes progress: novel have hidden dependence upon complex computing infrastructure (e.g. software dependencies, hardware), acting as unaddressed deterrent potential analysis users. While existing are increasingly shared open source software, needed...
Ensembling neural networks is an effective way to increase accuracy, and can often match the performance of individual larger models. This observation poses a natural question: given choice between deep ensemble single network with similar one preferable over other? Recent work suggests that ensembles may offer distinct benefits beyond predictive power: namely, uncertainty quantification robustness dataset shift. In this work, we demonstrate limitations these purported benefits, show (but...
Abstract Recent neuroscience studies demonstrate that a deeper understanding of brain function requires behavior. Detailed behavioral measurements are now often collected using video cameras, resulting in an increased need for computer vision algorithms extract useful information from data. Here we introduce new analysis tool combines the output supervised pose estimation (e.g. DeepLabCut) with unsupervised dimensionality reduction methods to produce interpretable, low-dimensional...
Abstract A popular approach to quantifying animal behavior from video data is through discrete behavioral segmentation, wherein frames are labeled as containing one or more classes such walking grooming. Sequence models learn map features extracted behaviors, and both supervised unsupervised methods common. However, each has its drawbacks: require a time-consuming annotation step where humans must hand label the desired behaviors; may fail accurately segment particular behaviors of interest....
Inference-time techniques are emerging as highly effective tools to enhance large language model (LLM) capabilities. However, best practices for developing systems that combine these remain underdeveloped due our limited understanding of the utility individual inference-time and interactions between them. Additionally, efficiently automatically searching space choices, techniques, their compositions is challenging design space. To address challenges, we introduce Archon, a modular framework...
Speech brain-computer interfaces aim to decipher what a person is trying say from neural activity alone, restoring communication people with paralysis who have lost the ability speak intelligibly. The Brain-to-Text Benchmark '24 and associated competition was created foster advancement of decoding algorithms that convert text. Here, we summarize lessons learned ending on June 1, 2024 (the top 4 entrants also presented their experiences in recorded webinar). largest improvements accuracy were...
Classic results establish that encouraging predictive diversity improves performance in ensembles of low-capacity models, e.g. through bagging or boosting. Here we demonstrate these intuitions do not apply to high-capacity neural network (deep ensembles), and fact the opposite is often true. In a large scale study nearly 600 classification ensembles, examine variety interventions trade off component model for diversity. While such can improve small (in line with standard intuitions), they...
Calcium imaging has revolutionized systems neuroscience, providing the ability to image large neural populations with single-cell resolution. The resulting datasets are quite large, which presented a barrier routine open sharing of this data, slowing progress in reproducible research. State art methods for analyzing data based on non-negative matrix factorization (NMF); these approaches solve non-convex optimization problem, and effective when good initializations available, but can break...
A major goal of neuroscience is the development powerful data analyses that process large datasets. Unfortunately, modern have a hidden dependence upon complex computing infrastructure (e.g. software dependencies, hardware), acts as an unaddressed deterrent to analysis users. While existing are increasingly shared open source software, and knowledge needed deploy these efficiently still only accessible minority experts. In this work we develop Neuroscience Cloud Analysis As Service...