site-logo

JAWS PANKRATION 2024

site-logo
HomeNewsTimetableCfPCommitteePromotionFollow UpPrivacy Policy

Not only Claude 3 and Amazon Forecast! Get the Future by "Chronos" of Amazon's Time series FM

Lv300

Lv300

2024/8/25 11:40 (JST)

セッション情報

While working with past and present time series data is easy on AWS, forecasting is difficult.

Now, we can analyze by Amazon Forecast, ML with SageMaker and LLM but it takes some time and effort, accuracy problems.

This session introduces Amazon's Time Series Foundation Model "Chronos", well suited to this problem.

Although Time Series FM is still in its infancy and there is no managed service yet, I will present it as a different GenAI from LLM.

Kohei  MATSUSHITA

Kohei MATSUSHITA

- AWS Heroes -



セッションカテゴリ
Analysis
IoT


関連AWSサービス

セッション資料

セッションアーカイブ

セッションサマリ(by Amazon Bedrock)
    The speaker, Matsushita (nicknamed MAX), introduces Amazon Chronos, a time series data service for easy AI-based future predictions. He explains the current AI trends, emphasizing the popularity of chat interfaces and pre-trained models. The presentation focuses on time series data, which is common in various business applications. While traditional use of time series data involves visualization and notifications, the speaker emphasizes the importance of predicting future outcomes. He discusses different methods for time series prediction, including multimodal language models and Amazon SageMaker, before introducing Amazon Chronos. Chronos is a specialized foundation model for time series data, developed by Amazon's research division. The speaker demonstrates Chronos' capabilities using a simple dataset of sequential numbers from 1 to 29. He shows how to make predictions and visualize the results using just a few lines of Python code. Chronos provides not only point predictions but also probability distributions, allowing for confidence interval calculations. The presentation then moves on to practical applications, showcasing how to integrate Chronos predictions into a web API using FastAPI. This allows for easy integration with other systems. Finally, the speaker discusses deploying the API using AWS Lambda with a container adapter, demonstrating how to containerize the application and deploy it on AWS. The presentation concludes by emphasizing the importance of choosing appropriate foundation models for specific data types and encouraging exploration of various options beyond just Amazon Chronos.

©JAWS-UG (AWS User Group - Japan). All rights reserved.