A5013816045- Explainable Artificial Intelligence (XAI)
- Adversarial Robustness in Machine Learning
- Machine Learning in Healthcare
- Topic Modeling
- Genetic Associations and Epidemiology
- Genetic and phenotypic traits in livestock
- Generative Adversarial Networks and Image Synthesis
- Genetic Mapping and Diversity in Plants and Animals
- Schizophrenia research and treatment
- RNA Research and Splicing
- RNA modifications and cancer
- Sepsis Diagnosis and Treatment
- Functional Brain Connectivity Studies
- Statistical Methods and Inference
- Medical Imaging Techniques and Applications
- Radiomics and Machine Learning in Medical Imaging
- Fault Detection and Control Systems
- Machine Learning and Data Classification
- Cardiac Arrest and Resuscitation
- Mechanical Circulatory Support Devices
- Gene expression and cancer classification
- Bayesian Modeling and Causal Inference
- RNA and protein synthesis mechanisms
- Obsessive-Compulsive Spectrum Disorders
- Neural Networks and Applications
New York University
2018-2024
Courant Institute of Mathematical Sciences
2020-2024
International Institute of Information Technology Bangalore
2022
Cornell University
2015-2017
The COVID-19 pandemic has challenged front-line clinical decision-making, leading to numerous published prognostic tools. However, few models have been prospectively validated and none report implementation in practice. Here, we use 3345 retrospective 474 prospective hospitalizations develop validate a parsimonious model identify patients with favorable outcomes within 96 h of prediction, based on real-time lab values, vital signs, oxygen support variables. In validation, the achieves high...
Machine learning methods, particularly neural networks trained on large datasets, are transforming how scientists approach scientific discovery and experimental design. However, current state-of-the-art limited by their uninterpretability: Despite excellent accuracy, they cannot describe arrived at predictions. Here, using an "interpretable-by-design" approach, we present a network model that provides insights into RNA splicing, fundamental process in the transfer of genomic information...
Myocardial infarction and heart failure are major cardiovascular diseases that affect millions of people in the USA with morbidity mortality being highest among patients who develop cardiogenic shock. Early recognition shock allows prompt implementation treatment measures. Our objective is to a new dynamic risk score, called CShock, improve early detection cardiac intensive care unit (ICU). We developed externally validated deep learning-based stratification tool, for admitted into ICU acute...
Shapley values are widely used to explain black-box models, but they costly calculate because require many model evaluations. We introduce FastSHAP, a method for estimating in single forward pass using learned explainer model. FastSHAP amortizes the cost of explaining inputs via learning approach inspired by value's weighted least squares characterization, and it can be trained standard stochastic gradient optimization. compare existing estimation approaches, revealing that generates...
Our knowledge of non-linear genetic effects on complex traits remains limited, in part, due to the modest power detect such effects. While kernel-based tests offer a versatile approach test for relationships between sets variants and traits, current approaches cannot be applied Biobank-scale datasets containing hundreds thousands individuals. We propose, FastKAST, that can set quantitative trait. FastKAST provides calibrated hypothesis while enabling analysis with unrelated individuals from...
While the need for interpretable machine learning has been established, many common approaches are slow, lack fidelity, or hard to evaluate. Amortized explanation methods reduce cost of providing interpretations by a global selector model that returns feature importances single instance data. The is trained optimize fidelity interpretations, as evaluated predictor target. Popular learn and in concert, which we show allows predictions be encoded within interpretations. We introduce EVAL-X...
Summary Machine learning methods, particularly neural networks trained on large datasets, are transforming how scientists approach scientific discovery and experimental design. However, current state-of-the-art limited by their uninterpretability: despite excellent accuracy, they cannot describe arrived at predictions. Here, using an “interpretable-by-design” approach, we present a network model that provides insights into RNA splicing, fundamental process in the transfer of genomic...
Predictive modeling often uses black box machine learning methods, such as deep neural networks, to achieve state-of-the-art performance. In scientific domains, the scientist wishes discover which features are actually important for making predictions. These discoveries may lead costly follow-up experiments and it is that error rate on not too high. Model-X knockoffs enable be discovered with control of FDR. However, require rich generative models capable accurately knockoff while ensuring...
Early detection of many life-threatening diseases (e.g., prostate and breast cancer) within at-risk population can improve clinical outcomes reduce cost care. While numerous disease-specific "screening" tests that are closer to Point-of-Care (POC) in use for this task, their low specificity results unnecessary biopsies, leading avoidable patient trauma wasteful healthcare spending. On the other hand, despite high accuracy Magnetic Resonance (MR) imaging disease diagnosis, it is not used as a...
Despite its rich clinical information content, MR has seen limited adoption in population-level screenings, due to concerns about specificity combined with high scan duration and cost. In order begin address such issues, accelerate the entire pipeline from data acquisition diagnosis, we introduce ARMS, an algorithm that learns k-space undersampling patterns maximize accuracy of pathology detection. ARMS detects pathologies directly undersampled data, bypassing explicit image reconstruction....
Robotic arms with varying degrees of freedom are the most prevalent in manufacturing industry. The main objective gesture recognition has always been to reduce barrier between real and digital worlds. In this paper, control a robotic arm using gestures wirelessly is demonstrated. Traditionally, people used flex sensors or accelerometers recognize RF Transmitters accomplish wireless communication. However, Communication, data loss was substantial, precision low. This strategy ineffective for...
Coordinate flying of drones or unmanned aerial vehicles(UAVs) is one the most intricate and vital techniques that many researchers are trying to solve. Drones in coordination can be used applications such as military, rescue operations, package delivery more. Coordination expand a single drone's task coverage radius also enhance success rate execution. In this paper, we address challenges faced communication two parallel methods using overcome difficulty.
Myocardial infarction and heart failure are major cardiovascular diseases that affect millions of people in the US. The morbidity mortality highest among patients who develop cardiogenic shock. Early recognition shock is critical. Prompt implementation treatment measures can prevent deleterious spiral ischemia, low blood pressure, reduced cardiac output due to However, early identification has been challenging human providers' inability process enormous amount data intensive care unit (ICU)...
Background: Classification in psychiatry continues to suffer from challenges validity of the distinctions between its diagnostic categories. The fundamental goal this study is delineate different types psychosis using a stratified approach involving both biological and clinical information. Methods: Patients with psychotic disorder (n = 404) were recruited at 5 sites underwent MRI scans, EEG comprehensive neuropsychological assessments. A machine learning (nonlinear K-means clustering)...