Scribd Logo
Upload

Welcome to Scribd!

  • 53 views

    IDE204 - TimeGPT Generative AI For Time Series

    Uploaded by

    Khoi Nguyen Nguyen

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    IDE204 - TimeGPT Generative AI For Time Series

    Uploaded by

    Khoi Nguyen Nguyen
    53 views36 pages

    Original Title

    IDE204_TimeGPT-Generative-AI-for-time-series

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    53 views36 pages

    IDE204 - TimeGPT Generative AI For Time Series

    Uploaded by

    Khoi Nguyen Nguyen

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    of 36

    © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.

    IDE204

    TimeGPT: Generative AI
    for time series

    Azul Garza
    (she/her)
    CTO and Co-Founder
    Nixtla

    © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
    Chatgpt by

    stable diffusion
    by

    TimeGPT by
    Businesses don’t speak in
    images or language; they
    speak time series
    We are building foundational
    AI to understand, classify, and
    predict time series
    Time series research is
    the systematic approach
    to unveil the future
    It enables anticipating
    ocean tides or predicting
    closing values of the
    Dow Jones
    Time series are the DNA
    of finance, commerce,
    energy, and technology
    Nixtla created the most
    comphrehensive time series
    ecosystem in the world
    We learned a lot.
    But one insight
    stood out.
    Forecasting can be
    extremely hard,
    confusing, and
    expensive
    We like big problems
    We love big solutions
    Introducing TimeGPT-1
    Generative AI for
    Temporal Data
    The first foundation
    model for time series
    TimeGPT-1
    TimeGPT is more accurate
    Orders of magnitude faster
    Fine tunable in your
    enterprise data
    No machine learning
    expertise needed
    Simple
    The future is for everybody
    We launched the beta
    8 weeks ago
    More than 5,000 companies
    signed up
    Sectors: Tech, Consultancies,
    Retail, Manufacturing,
    Finance, Energy, Healthcare.
    Some use cases:
    Demand forecasting,
    inventory management, waste
    reduction, anomaly detection,
    healthcare, iot, predictive
    maintenance, finance, …
    We got some unexpected
    things too
    Predicting circadian rhythms
    of cattle and WWE audience
    Let’s compare TimeGPT to
    VertexAI
    NYC Bike Trips: 383 time series
    Performance Comparison
    Model MSE MAE
    TimeGPT (Nixtla) 24.68 20.40
    Deep learning model 25.36 21.18
    Statistical model 29.16 24.13
    Conclusions
    TimeGPT:

    is 5% more accurate than deep learning model and 20% more accurate
    than statistical model.

    took 1.6 minutes against 4 hours of deep learning model .

    deployment took less than an hour against 8 hours of deep learning model .

    only takes two lines of code to be deployed in production.


    Thank you! Please complete the session
    survey in the mobile app

    Go to nixtla.io/#timegpt and submit interest

    © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.

    You might also like