A5012700366- Parallel Computing and Optimization Techniques
- Artificial Intelligence in Healthcare and Education
- Distributed and Parallel Computing Systems
- Embedded Systems Design Techniques
- Machine Learning in Healthcare
- Cloud Computing and Resource Management
- Topic Modeling
- Advanced Data Storage Technologies
- Distributed systems and fault tolerance
- Software Testing and Debugging Techniques
- Security and Verification in Computing
- Network Security and Intrusion Detection
- Natural Language Processing Techniques
- Graph Theory and Algorithms
- Explainable Artificial Intelligence (XAI)
- Real-time simulation and control systems
- Radiomics and Machine Learning in Medical Imaging
- Computational Fluid Dynamics and Aerodynamics
- Advanced MRI Techniques and Applications
- Recommender Systems and Techniques
- Advanced Malware Detection Techniques
- Scientific Computing and Data Management
- Bone health and osteoporosis research
- Advanced Neuroimaging Techniques and Applications
- Clinical Reasoning and Diagnostic Skills
Princeton University
2020-2024
New York University
2024
Massachusetts General Hospital
2023
Abstract Recent advances in large language models (LLMs) have demonstrated remarkable successes zero- and few-shot performance on various downstream tasks, paving the way for applications high-stakes domains. In this study, we systematically examine capabilities limitations of LLMs, specifically GPT-3.5 ChatGPT, performing zero-shot medical evidence summarization across six clinical We conduct both automatic human evaluations, covering several dimensions summary quality. Our study...
Abstract Recent studies indicate that Generative Pre-trained Transformer 4 with Vision (GPT-4V) outperforms human physicians in medical challenge tasks. However, these evaluations primarily focused on the accuracy of multi-choice questions alone. Our study extends current scope by conducting a comprehensive analysis GPT-4V’s rationales image comprehension, recall knowledge, and step-by-step multimodal reasoning when solving New England Journal Medicine (NEJM) Image Challenges—an imaging quiz...
Recent advances in large language models (LLMs) have demonstrated remarkable successes zero- and few-shot performance on various downstream tasks, paving the way for applications high-stakes domains. In this study, we systematically examine capabilities limitations of LLMs, specifically GPT-3.5 ChatGPT, performing zero-shot medical evidence summarization across six clinical We conduct both automatic human evaluations, covering several dimensions summary quality. Our study has that metrics...
Systematic literature review is essential for evidence-based medicine, requiring comprehensive analysis of clinical trial publications. However, the application artificial intelligence (AI) models medical mining has been limited by insufficient training and evaluation across broad therapeutic areas diverse tasks. Here, we present LEADS, an AI foundation model study search, screening, data extraction from literature. The trained on 633,759 instruction points in LEADSInstruct, curated 21,335...
The promise of automatic parallelization, freeing programmers from the error-prone and time-consuming process making efficient use parallel processing resources, remains unrealized. For decades, imprecision memory analysis limited applicability non-speculative parallelization. introduction speculative parallelization overcame these limitations, but, even in case no misspeculation, techniques exhibit high communication bookkeeping costs for validation commit. This paper presents Perspective,...
Program analysis determines the potential dataflow and control flow relationships among instructions so that compiler optimizations can respect these to transform code correctly. Since many of rarely or never occur, speculative assert they do not exist while optimizing code. To preserve correctness, add validation checks activate recovery when assertions prove untrue. This approach results in missed opportunities because program thus other remain unaware full impact dynamically-enforced...
Modern and emerging architectures demand increasingly complex compiler analyses transformations. As the emphasis on infrastructure moves beyond support for peephole optimizations extraction of instruction-level parallelism, compilers should custom tools designed to meet these demands with higher-level analysis-powered abstractions functionalities wider program scope. This paper introduces NOELLE, a robust open-source domain-independent compilation layer built upon LLVM providing this...
A compiler's intermediate representation (IR) defines a program's execution plan by encoding its instructions and their relative order. Compiler optimizations aim to replace given with semantically-equivalent one that increases the performance for target architecture. Alternative representations of an IR, like Program Dependence Graph (PDG), aid this process capturing minimum set constraints plans must satisfy. Parallel programming OpenMP extends sequential adding possibility running in...
Type-safe languages improve application safety by eliminating whole classes of vulnerabilities–such as buffer overflows–by construction. However, this sometimes comes with a performance cost. As result, many modern type-safe provide escape hatches that allow developers to manually bypass them. The relative value and the degree obtained depends upon context, including user goals hardware which is be executed. Since libraries may used in different contexts, library cannot make...
Recent studies indicate that Generative Pre-trained Transformer 4 with Vision (GPT-4V) outperforms human physicians in medical challenge tasks. However, these evaluations primarily focused on the accuracy of multi-choice questions alone. Our study extends current scope by conducting a comprehensive analysis GPT-4V's rationales image comprehension, recall knowledge, and step-by-step multimodal reasoning when solving New England Journal Medicine (NEJM) Image Challenges - an imaging quiz...
Memory profiling captures programs’ dynamic memory behavior, assisting programmers in debugging, tuning, and enabling advanced compiler optimizations like speculation-based automatic parallelization. As each use case demands its unique program trace summary, various profiler types have been developed. Yet, designing practical profilers often requires extensive expertise, adeptness optimization, significant implementation effort. This results a void where aspirations for fast robust remain...
Motivation: Current understanding of dolichoectasia has largely been drawn from patients with clinical need for brain imaging in the cross-sectional settings. Goal(s): To characterize longitudinal changes MRA geometry vessel metrics and their associations demographic variables, biomarkers cognitive performance. Approach: Basic information were compared between two groups w/wo MRI metric measurement change using two-sample t-tests. The decline or incidence dementia tested logistic regression....
While holding great promise for improving and facilitating healthcare, large language models (LLMs) struggle to produce up-to-date responses on evolving topics due outdated knowledge or hallucination. Retrieval-augmented generation (RAG) is a pivotal innovation that improves the accuracy relevance of LLM by integrating LLMs with search engine external sources knowledge. However, quality RAG can be largely impacted rank density key information in retrieval results, such as...
Manually writing parallel programs is difficult and error-prone. Automatic parallelization could address this issue, but profitability can be limited by not having facts known only to the programmer. A parallelizing compiler that collaborates with programmer increase coverage performance of while reducing errors overhead associated manual parallelization. Unlike collaboration involving analysis tools report program properties or make suggestions programmer, decompiler-based leverage strength...
Memory profiling captures programs' dynamic memory behavior, assisting programmers in debugging, tuning, and enabling advanced compiler optimizations like speculation-based automatic parallelization. As each use case demands its unique program trace summary, various profiler types have been developed. Yet, designing practical profilers often requires extensive expertise, adeptness optimization, significant implementation efforts. This results a void where aspirations for fast robust remain...