Corpus ID: 29546424

Time-series Extreme Event Forecasting with Neural Networks at Uber

  title={Time-series Extreme Event Forecasting with Neural Networks at Uber},
  author={Nikolay Pavlovich Laptev and Jason Yosinski and Li Erran Li and Slawek Smyl},
Accurate time-series forecasting during high variance segments (e.g., holidays), is critical for anomaly detection, optimal resource allocation, budget planning and other related tasks. At Uber accurate prediction for completed trips during special events can lead to a more efficient driver allocation resulting in a decreased wait time for the riders. State of the art methods for handling this task often rely on a combination of univariate forecasting models (e.g., Holt-Winters) and machine… Expand

Figures from this paper

Modeling approaches for time series forecasting and anomaly detection
Accurate time series forecasting is critical for business operations for optimal resource allocation, budget planning, anomaly detection and tasks such as predicting customer growth, or understandingExpand
Time Series Forecasting with Recurrent Neural Networks in Presence of Missing Data
A novel model that utilize Dilated RNN(DRNN) and a modified attention mechanism is proposed, focusing on the problem of time series forecasting with missing data, which outperforms existing models such as AutoRegressive Integrated Moving Average (ARIMA) and Gated Recurrent Unit (GRU), with respect to the forecast accuracy on benchmark datasets. Expand
Deep and Confident Prediction for Time Series at Uber
  • Lingxue Zhu, N. Laptev
  • Physics, Mathematics
  • 2017 IEEE International Conference on Data Mining Workshops (ICDMW)
  • 2017
A novel end-to-end Bayesian deep model is proposed that provides time series prediction along with uncertainty estimation at Uber and is successfully applied to large-scale time series anomaly detection at Uber. Expand
RECOWNs: Probabilistic Circuits for Trustworthy Time Series Forecasting
Recurrent Neural Networks are shown to be accurate and trustworthy time series predictors, able to “know when they do not know”, and a Log-Likelihood Ratio Score is formulated as better estimation of uncertainty that is tailored to time series and Whittle likelihoods. Expand
Time Series Forecasting Using Sequence-to-Sequence Deep Learning Framework
Through the air quality multivariate time series forecasting experiments, the proposed model has better forecasting performance than classic shallow learning and baseline deep learning models and the predicted PM2.5 value can be well matched with the ground truth value under single timestep and multi-timestep forward forecasting conditions. Expand
Recurrent neural networks for time series prediction
This thesis proposes a novel approach to infer accurate and reliable predictive distributions from scarce time series data through a recurrent neural network approach that infers a joint feature representation over the space of quantised time series by means of a compilation of auxiliary datasets. Expand
Few-shot Learning for Time-series Forecasting
A few-shot learning method that forecasts a future value of a time- series in a target task given a few time-series in the target task by minimizing an expected test error of forecasting next timestep values. Expand
LSTM Algorithm For Forecasting Events In Changing Electric Consumption
The industrial sections have always been thirsty to predict events and be aware of future sudden changes. Where using time-series forecasting based on deep learning techniques can have a significantExpand
This paper shows for the first time how time-series representations learned with Long Short Term Memory on large-scale datasets can be efficiently transferred to other time- series forecasting tasks with limited amount of training data and validates that the transferred representation leads to significantly improved forecasting results. Expand
Deep Learning for Non-stationary Multivariate Time Series Forecasting
This paper uses the heterogeneous data streams that generated from a limited number of existing Internet of Things devices which monitor different parking lots, to construct the parking availability data and employs deep Gated Recurrent Units (GRUs) to forecast multivariate time series in parking lots setting. Expand


A new boosting algorithm for improved time-series forecasting with recurrent neural networks
The overall results obtained through this ensemble method are more accurate than those obtained through the standard method, backpropagation through time, on these datasets and perform significantly better even when long-range dependencies play an important role. Expand
Automatic Time Series Forecasting: The forecast Package for R
Two automatic forecasting algorithms that have been implemented in the forecast package for R, based on innovations state space models that underly exponential smoothing methods, are described. Expand
Large-Scale Unusual Time Series Detection
This work computes a vector of features on each time series, measuring characteristics of the series, and uses various bivariate outlier detection methods applied to the first two principal components to identify servers that are behaving unusually. Expand
Time series shapelets: a new primitive for data mining
A new time series primitive, time series shapelets, is introduced, which can be interpretable, more accurate and significantly faster than state-of-the-art classifiers. Expand
Forecast verification for extreme value distributions with an application to probabilistic peak wind prediction
Predictions of the uncertainty associated with extreme events are a vital component of any prediction system for such events. Consequently, the prediction system ought to be probabilistic in nature,Expand
Nonlinear Systems Identification Using Deep Dynamic Neural Networks
It is demonstrated that deep neural networks are effective model estimators from input-output data and associated characteristics of the underlying dynamical systems. Expand
Quantile Regression Forests
It is shown here that random forests provide information about the full conditional distribution of the response variable, not only about the conditional mean, in order to be competitive in terms of predictive power. Expand
Uncertainty in Deep Learning
This work develops tools to obtain practical uncertainty estimates in deep learning, casting recent deep learning tools as Bayesian models without changing either the models or the optimisation, and develops the theory for such tools. Expand
Modeling asymptotically independent spatial extremes based on Laplace random fields
We tackle the modeling of threshold exceedances in asymptotically independent stochastic processes by constructions based on Laplace random fields. These are defined as Gaussian random fields scaledExpand
Bootstrapping time series models
This paper surveys recent development in bootstrap methods and the modifications needed for their applicability in time series models and argues that bootstrapping residuals is the preferable approach. Expand