醫療保健中的 MLOPS 和 AI 治理：普羅維登斯的用例.pdf

《醫療保健中的 MLOPS 和 AI 治理：普羅維登斯的用例.pdf》由會員分享，可在線閱讀，更多相關《醫療保健中的 MLOPS 和 AI 治理：普羅維登斯的用例.pdf（14頁珍藏版）》請在三個皮匠報告上搜索。

1、2024 Databricks Inc.All rights reserved1MLOpsMLOps and AI and AI Governance in Governance in HealthcareHealthcareLindsay Mico and Vivek TomerLindsay Mico and Vivek TomerJune 12,2024June 12,20242024 Databricks Inc.All rights reserved2024 Databricks Inc.All rights reserved“Transformation and collabora

2、tion will be a consistent theme across health care in 2024.Though health care has been talking about disrupting itself for years,this will be a year of accelerated change and new innovations adopted at scale.The explosion of generative AI will be one of the major drivers of transformation,and well s

3、ee health systems partner with the tech sector to responsibly usher in new innovations.”-Rod Hochman,President and CEO of Providence32024 Databricks Inc.All rights reserved4Driving Transformation with AIDriving Transformation with AI2024 Databricks Inc.All rights reservedPriority AreasPriority Areas

4、Clinical:Workflow optimization and support in documentation/charting and in-basket managementPatient/Consumer:Personalization and navigation for better self-serviceBack Office:Automation of processes that are not patient-facing/differentiatingWorkforce/Administrative:Augment our workforce to support

5、 productivity and reduce burden2024 Databricks Inc.All rights reservedA Tsunami of OpportunityA Tsunami of Opportunity77%12%8%0%50%100%PendingDevelopmentProductionAI Pipeline2024 Databricks Inc.All rights reservedA Flywheel of TransformationA Flywheel of TransformationPrioritizeExecuteMeasureFund202

6、4 Databricks Inc.All rights reserved82024 Databricks Inc.All rights reservedCheckpoint 1B is initial validation of the proof of concept(POC)for the internal and vendor AI/ML models.The advertised AI/ML performance metrics of the POC are validated and documented for future use.Checkpoint 2 fully vali

7、dates the AI/ML models after model development and before model deployment.Checkpoint 2 has three outcomes:the model is approved,the model needs improvement,and the model is not approved for deployment.Continuous model monitoring is performed after the model is in production to detect drift in model

8、 performance.Checkpoint 1BCheckpoint 2Model MonitoringCheckpoint 1BCheckpoint 2Model Monitoring9Checkpoints in AI Guardrails ProcessCheckpoints in AI Guardrails ProcessThree Components of Model Risk Management(MRM)Three Components of Model Risk Management(MRM)2024 Databricks Inc.All rights reserved

9、Objective of MLOps Platform:Create a robust platform for developing,validating,and deploying a large inventory of AI/ML models at scale.Four Pillars of MLOps:Model development,model validation,model deployment,and model monitoring with strategic partnership with Databricks for all four pillars.10Dev

10、elopment of the MLOps PlatformDevelopment of the MLOps PlatformObjectives and MLOps PillarsObjectives and MLOps Pillars2024 Databricks Inc.All rights reserved Databricks Serverless adds to the growing API ecosystem at Providence and extends the infrastructure needed to build AI Applications of the f

11、uture.Databricks Serverless allows us to convert our homebrewed traditional machine learning models and fine-tuned Large Language Models(LLMs)to API and share them across the enterprise.Databricks Serverless enables the creation of API endpoints for open-source LLMs,thereby creating a low-cost alter

12、native to OpenAI models.11Databricks Serverless at ProvidenceDatabricks Serverless at Providence2024 Databricks Inc.All rights reserved12Centralize AI Deployment2024 Databricks Inc.All rights reserved13Model Fairness&SafetyModel Fairness&Safety2024 Databricks Inc.All rights reserved14Continuous Model MonitoringContinuous Model Monitoring