gdg devfest xiamen 2017

Download GDG DevFest Xiamen 2017

Post on 21-Jan-2018

85 views

Category:

Engineering

1 download

Embed Size (px)

TRANSCRIPT

  1. 1. Taegyun Jeon GDG DevFest Xiamen 2017 Time Series Analysis using TensorFlow
  2. 2. Taegyun Jeon (South Korea) Google Developer Expert - Machine Learning (2017) PhD (Machine Learning) Speaker
  3. 3. Deep Learning Applications for TSA Time Series Analysis (TSA) TensorFlow API for Time Series Contents
  4. 4. Deep Learning Applications
  5. 5. Deep Learning Applications (TSA) Finance Speech Recognition Natural Language Processing / Translation Medicine Weather Forecasting Sales Forecasting
  6. 6. Time Series Classification for Finance https://cloud.google.com/solutions/financial-services/#development_guides
  7. 7. Time Series Classification for Finance https://cloud.google.com/solutions/financial-services/#development_guides
  8. 8. Speech Recognition Speech Recognition API Google Home, Assistant, Nest and Cast
  9. 9. Natural Language Translation
  10. 10. Natural Language Translation
  11. 11. Cardiogram https://cardiogr.am
  12. 12. Cardiogram
  13. 13. EEG Classification (Conv-FC)
  14. 14. EEG Classification (LSTM)
  15. 15. Weather Forecasting MeteoSWISS
  16. 16. Time Series Analysis
  17. 17. Time Series Time Series Analysis Models for Time Series Analysis: AR, MA, ARMA, ARIMA TensorFlow TimeSeries API (TFTS)
  18. 18. Time Series Analysis Time Series Analysis Models for Time Series Analysis: AR, MA, ARMA, ARIMA TensorFlow TimeSeries API (TFTS)
  19. 19. Time Series Analysis
  20. 20. Time Series Data
  21. 21. Time Series Data Stock values Economic variables Weather Sensor: Internet-of-Things Energy demand Signal processing Sales forecasting
  22. 22. Problems on Time Series Data Standard Supervised Learning IID assumption Same distribution for training and test data Distributions fixed over time (stationarity) Time Series Not applicable
  23. 23. Models for Time Series Analysis Time Series Analysis Models for Time Series Analysis: AR, MA, ARMA, ARIMA, Recurrent Neural Networks TensorFlow TimeSeries API (TFTS)
  24. 24. Autoregressive (AR) Models AR(p) model : Linear generative model based on the pth order Markov assumption : zero mean uncorrelated random variables with variance : autoregressive coefficients : observed stochastic process
  25. 25. Moving Average (MA) MA(q) model : Linear generative model for noise term on the qth order Markov assumption : moving average coefficients
  26. 26. ARMA Model ARMA(p,q) model : generative linear model that combines AR(p) and MA(q) models
  27. 27. Stationarity Definition: a sequence of random variables is stationary if its distribution is invariant to shifting in time.
  28. 28. Lag Operator Definition: Lag operator is defined by ARMA model in terms of the lag operator: Characteristic polynomial can be used to study properties of this stochastic process.
  29. 29. ARIMA Model Definition: Non-stationary processes can be modeled using processes whose characteristic polynomial has unit roots. Characteristic polynomial with unit roots can be factored: ARIMA(p, D, q) model is an ARMA(p,q) model for
  30. 30. Other Extensions Further variants: Models with seasonal components (SARIMA) Models with side information (ARIMAX) Models with long-memory (ARFIMA) Multi-variate time series model (VAR) Models with time-varing coefficients other non-linear models
  31. 31. Recurrent Neural Networks
  32. 32. Recurrent Neural Networks
  33. 33. Recurrent Neural Networks
  34. 34. Recurrent Neural Networks
  35. 35. Recurrent Neural Networks
  36. 36. Recurrent Neural Networks
  37. 37. Recurrent Neural Networks
  38. 38. Recurrent Neural Networks
  39. 39. Recurrent Neural Networks
  40. 40. TensorFlow API for Time Series
  41. 41. TensorFlow API for Time Series Time Series Analysis Models for Time Series Analysis: AR, MA, ARMA, ARIMA TensorFlow TimeSeries API (TFTS)
  42. 42. TensorFlow TimeSeries tf.contrib.timeseries Classic model (state space, autoregressive) Flexible infrastructure Data management Chunking Batching Saving model Truncated backpropagation
  43. 43. EXAMPLES 1. Probabilistic Forecasts 2. Known Anomaly 3. Multivariate Forecasting / Anomaly Detection 4. Custom Model Building TensorFlow TimeSeries
  44. 44. EXAMPLES 1. Probabilistic Forecasts 2. Known Anomaly 3. Multivariate Forecasting / Anomaly Detection 4. Custom Model Building TensorFlow TimeSeries
  45. 45. EXAMPLES 1. Probabilistic Forecasts 2. Known Anomaly 3. Multivariate Forecasting / Anomaly Detection 4. Custom Model Building TensorFlow TimeSeries
  46. 46. EXAMPLES 1. Probabilistic Forecasts 2. Known Anomaly 3. Multivariate Forecasting / Anomaly Detection 4. Custom Model Building TensorFlow TimeSeries
  47. 47. EXAMPLES 1. Probabilistic Forecasts 2. Known Anomaly 3. Multivariate Forecasting / Anomaly Detection 4. Custom Model Building TensorFlow TimeSeries
  48. 48. Deep Learning Applications for Time Series Analysis Time Series Analysis TensorFlow API for Time Series Summary