Deep and Confident Prediction for Time Series at Uber

By Lingxue Zhu, Nikolay Pavlovich Laptev. 2017 IEEE International Conference on Data Mining Workshops (ICDMW), 2017

373 citations recorded by the source corpus.

Reliable uncertainty estimation for time series prediction is critical in many fields, including physics, biology, and manufacturing. At Uber, probabilistic time series forecasting is used for robust prediction of number of trips during special events, driver incentive allocation, as well as real-time anomaly detection across millions of metrics. Classical time series models are often used in conjunction with a probabilistic formulation for uncertainty estimation. However, such models are hard to tune, scale, and add exogenous variables to. Motivated by the recent resurgence of Long Short Term Memory networks, we propose a novel end-to-end Bayesian deep model that provides time series prediction along with uncertainty estimation. We provide detailed experiments of the proposed solution on completed trips data, and successfully apply it to large-scale time series anomaly detection at Uber.

Subfields: Artificial Intelligence, Signal Processing

View Paper: https://www.semanticscholar.org/paper/23afea0a4d9f598e6e63846d0ee2c2a046528708

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Signal Processing

Citations in the corpus

Deep Learning for Time Series Forecasting: Tutorial and Literature Survey (2020, methodology): “In [205], the authors use an LSTM with Monte Carlo dropout as both the encoder and decoder.”
Deep Learning on Traffic Prediction: Methods, Analysis, and Future Directions (2021, methodology): “In RNN-based sequence learning, a special network structure known as encoder-decoder has been applied for trafﬁc prediction ( [53], [58], [61]–[66], [71]–[79] ).”
On Last-Layer Algorithms for Classification: Decoupling Representation from Uncertainty Estimation (2020, methodology): “This technique, often named Monte-Carlo Dropout (MC-Dropout), is widely used in practical applications [ZL17, LAA + 17, NPAA18] due to its simplicity and good performance.”
On Hallucination and Predictive Uncertainty in Conditional Language Generation (2021, positive): “Other applications of uncertainty quantification have been explored in the context of time series predictions (Zhu and Laptev, 2017), natural language processing tasks (Xiao and Wang, 2019), etc.”
Anomaly Detection for IoT Time-Series Data: A Survey (2020, methodology): “Bayesian Neural Networks are investigated in [77], by using an LSTM-based auto-encoder to perform prediction for a number of steps ahead, followed by an MLP to perform the ﬁnal prediction steps, this construction provides not only a prediction for later values but a level of certainty in that…”
What Clinicians Want: Contextualizing Explainable Machine Learning for Clinical End Use (2019, background): “ods (Zhu and Laptev, 2017) and to the best of our knowledge remains mostly unexplored in clinical healthcare employing deep learning methods.”
Single-Model Uncertainties for Deep Learning (2018, background): “As argued by Begoli et al. [5], uncertainty quantiﬁcation is a problem of paramount importance when deploying machine learning models in sensitive domains such as information security [72], engineering [82], transportation [87], and medicine [5], to name a few.”
JHU-CROWD++: Large-Scale Crowd Counting Dataset and A Benchmark Method (2020, methodology): “Inspired by uncertainty estimation in CNNs [39, 40, 41, 74], we aim to model pixel-wise aleatoric uncertainty of the residuals estimated by CBi.”
MOrdReD: Memory-based Ordinal Regression Deep Neural Networks for Time Series Forecasting (2018, background): “Indeed, recent work looks to apply these approaches to time series forecasting [34], however still making the tacit assumption that the predictive distribution is uni-modal Gaussian.”
Autoregressive Denoising Diffusion Models for Multivariate Probabilistic Time Series Forecasting (2021, neutral): “Bayesian neural networks can also be used to provide epistemic uncertainty in fore-casts as well as detect distributional shifts (Zhu & Laptev, 2018), although these methods often do not perform as well empirically (Wenzel et al., 2020).”
Deep Learning for Time Series Anomaly Detection: A Survey (2022, neutral): “Uber uses an end-to-end neural network architecture for uncertainty estimation [197].”
Short-Term Load Forecasting With Deep Residual Networks (2018, methodology): “According to [37], σ 2 can be estimated using an independent validation dataset.”