Retrieval Augmented Forecasting for Sequential Data
Thursday, June 6, 2024
Forecasting sequential data is a popular problem in the machine learning world, with several applications in the AIOps, Financial, Healthcare and manufacturing domains. Moving average based techniques have been traditionally used here, but they have limitations in their ability to predict abnormal patterns. Deep Learning using sequence models has been lately popular, but while they are powerful, they are also expensive for continuous forecasting. In this session, we discuss a new technique called retrieval augmented forecasting. This technique combines the power of historical data with deep learning models to provide a cheap, yet powerful forecasting solution that can continuously learn and handle a variety of situations. This is an application of Generative AI to the numeric data world. We will discuss a specific implementation for forecasting lag in Apache Kafka consumers. This session would be useful for Data scientists and IT engineers, who work in the domain of sequence forecasting.