# Deep and Confident Prediction for Time Series at Uber

**Authors:** Lingxue Zhu, Nikolay Laptev  
**arXiv:** [1709.01907](https://arxiv.org/abs/1709.01907)  
**Date:** 2017-09-06  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/1709.01907v1/

## Key points

- Reliable uncertainty estimation is crucial in time series prediction across various fields such as physics, biology, and manufacturing.
- Probabilistic time series forecasting at Uber plays a key role in predicting trip numbers during special events, allocating driver incentives, and detecting real-time anomalies.
- Traditional time series models have challenges in tuning, scalability, and incorporating exogenous variables.
- The authors propose an innovative end-to-end Bayesian deep model based on LSTM networks for accurate time series prediction and robust uncertainty estimation.
- Detailed experiments on completed trips data demonstrate the effectiveness of the proposed solution in large-scale time series anomaly detection at Uber.

## Summary

In their paper titled "Deep and Confident Prediction for Time Series at Uber," authors Lingxue Zhu and Nikolay Laptev highlight the importance of reliable uncertainty estimation in time series prediction across various fields such as physics, biology, and manufacturing. At Uber, probabilistic time series forecasting plays a crucial role in predicting the number of trips during special events, allocating driver incentives, and detecting real-time anomalies across a multitude of metrics. Traditional time series models are commonly paired with a probabilistic framework to estimate uncertainty; however, these models often present challenges in terms of tuning, scalability, and incorporating exogenous variables. Motivated by the resurgence of Long Short Term Memory networks (LSTM), the authors propose an innovative end-to-end Bayesian deep model that not only offers accurate time series prediction but also provides robust uncertainty estimation. Through detailed experiments conducted on completed trips data, the authors demonstrate the effectiveness of their proposed solution in large-scale time series anomaly detection at Uber. This novel approach addresses key limitations of traditional models and showcases the potential for leveraging advanced deep learning techniques to enhance predictive capabilities while effectively managing uncertainty in time series forecasting applications.