A Continuously Trained
BERT Financial Sentiment Model
Friday, December 4, 2020 @ 11 AM PST
Presented by Jagane Sundar of InfinStor
and Gita Seshadri of Multilex
Leverage the power of InfinStor’s storage innovations on AWS to create and continuously improve a BERT based sentiment analysis model for financial news. InfinSnap fine grained snapshot is used for the initial training run, and subsequent improvements are made using slices of data accessed using InfinSlice. InfinStor Scheduled Run Transforms, developed using AWS EventBridge and AWS Lambda, is used in lieu of cron jobs for collecting data. InfinStor’s MLflow as a Service, built to cloud scale using AWS DynamoDB and AWS Lambda, is used to track the machine learning project lifecycle.