A5014018142- Bayesian Methods and Mixture Models
- Machine Learning in Healthcare
- Artificial Intelligence in Healthcare and Education
- Digital Mental Health Interventions
- Gaussian Processes and Bayesian Inference
- Statistical Methods and Inference
- Sepsis Diagnosis and Treatment
- Topic Modeling
- Mental Health Research Topics
- Advanced Clustering Algorithms Research
- Sparse and Compressive Sensing Techniques
- Complex Network Analysis Techniques
- Domain Adaptation and Few-Shot Learning
- Explainable Artificial Intelligence (XAI)
- Multiple Sclerosis Research Studies
- Algorithms and Data Compression
- Speech and Audio Processing
- Adversarial Robustness in Machine Learning
- Image and Signal Denoising Methods
- Anomaly Detection Techniques and Applications
- Opinion Dynamics and Social Influence
- Ethics in Clinical Research
- Colorectal Cancer Screening and Detection
- Natural Language Processing Techniques
- Ethics and Social Impacts of AI
Google (United States)
2019-2025
Google (United Kingdom)
2025
Duke University
2015-2024
DeepMind (United Kingdom)
2024
Durham Technical Community College
2023
University of California, Berkeley
2022
Carnegie Mellon University
2022
Pain and Rehabilitation Medicine
2018
North Central College
2014
University of Cambridge
2008-2012
ML models often exhibit unexpectedly poor behavior when they are deployed in real-world domains. We identify underspecification as a key reason for these failures. An pipeline is underspecified it can return many predictors with equivalently strong held-out performance the training domain. Underspecification common modern pipelines, such those based on deep learning. Predictors returned by pipelines treated equivalent their domain performance, but we show here that behave very differently...
We present a novel algorithm for agglomerative hierarchical clustering based on evaluating marginal likelihoods of probabilistic model. This has several advantages over traditional distance-based algorithms. (1) It defines model the data which can be used to compute predictive distribution test point and probability it belonging any existing clusters in tree. (2) uses model-based criterion decide merging rather than an ad-hoc distance metric. (3) Bayesian hypothesis testing is merges are...
Background Pythia is an automated, clinically curated surgical data pipeline and repository housing all patient electronic health record (EHR) from a large, quaternary, multisite institute for science initiatives. In effort to better identify high-risk patients complex data, machine learning project trained on was built predict postoperative complication risk. Methods findings A of outcomes created using automated SQL R code that extracted processed clinical across 37 million encounters the...
Abstract Artificial intelligence as a medical device is increasingly being applied to healthcare for diagnosis, risk stratification and resource allocation. However, growing body of evidence has highlighted the algorithmic bias, which may perpetuate existing health inequity. This problem arises in part because systemic inequalities dataset curation, unequal opportunity participate research access. study aims explore standards, frameworks best practices ensuring adequate data diversity...
Successful integrations of machine learning into routine clinical care are exceedingly rare, and barriers to its adoption poorly characterized in the literature.This study aims report a quality improvement effort integrate deep sepsis detection management platform, Sepsis Watch, care.In 2016, multidisciplinary team consisting statisticians, data scientists, engineers, clinicians was assembled by leadership an academic health system radically improve treatment sepsis. This follows framework...
In medicine, both ethical and monetary costs of incorrect predictions can be significant, the complexity problems often necessitates increasingly complex models. Recent work has shown that changing just random seed is enough for otherwise well-tuned deep neural networks to vary in their individual predicted probabilities. light this, we investigate role model uncertainty methods medical domain. Using RNN ensembles various Bayesian RNNs, show population-level metrics, such as AUC-PR, AUC-ROC,...
Abstract For many biological applications, exploration of the massive parametric space a mechanism-based model can impose prohibitive computational demand. To overcome this limitation, we present framework to improve efficiency by orders magnitude. The key concept is train neural network using limited number simulations generated mechanistic model. This small enough such that be completed in short time frame but large enable reliable training. trained then used explore much larger space. We...
Determine if deep learning detects sepsis earlier and more accurately than other models. To evaluate model performance using implementation-oriented metrics that simulate clinical practice.We trained internally temporally validated a (multi-output Gaussian process recurrent neural network [MGP-RNN]) to detect encounters from adult hospitalized patients at large tertiary academic center. Sepsis was defined as the presence of 2 or systemic inflammatory response syndrome (SIRS) criteria, blood...
Machine learning (ML) approaches have demonstrated promising results in a wide range of healthcare applications. Data plays crucial role developing ML-based systems that directly affect people's lives. Many the ethical issues surrounding use ML stem from structural inequalities underlying way we collect, use, and handle data. Developing guidelines to improve documentation practices regarding creation, maintenance datasets is therefore critical importance. In this work, introduce Healthsheet,...
Abstract The variability in the visual interpretation of cardiotocograms (CTGs) poses substantial challenges obstetric care. Despite recent strides automated CTG for early detection fetal hypoxia, comparative efficacy objective versus subjective ground truth labels and robustness to temporal distribution shifts remains underexplored. Using a published convolutional neural network (CNN), we predict compromise from recordings, incorporating pre-processing hyperparameter tuning. We use an...
Background Recently, artificial intelligence-powered devices have been put forward as potentially powerful tools for the improvement of mental healthcare. An important question is how these impact physician-patient interaction. Aims Aifred an clinical decision support system (CDSS) treatment major depression. Here, we explore use a simulation centre environment in evaluating usability Aifred, particularly its on physician–patient Method Twenty psychiatry and family medicine attending staff...
We present a scalable end-to-end classifier that uses streaming physiological and medication data to accurately predict the onset of sepsis, life-threatening complication from infections has high mortality morbidity. Our proposed framework models multivariate trajectories continuous-valued time series using multitask Gaussian processes, seamlessly accounting for uncertainty, frequent missingness, irregular sampling rates typically associated with real clinical data. The process is directly...
Although interest in ketamine use during electroconvulsive therapy (ECT) has increased, studies have been equivocal with regard to its efficacy. The aims of this clinical trial were evaluate ketamine's antidepressive effects ECT as a primary anesthetic, determine tolerability when compared standard anesthesia, and if plasma brain-derived neurotrophic factor (BDNF) is necessary for treatment response.Adults meeting criteria treatment-resistant depression undergoing index course received...