A5048713693- Time Series Analysis and Forecasting
- Forecasting Techniques and Applications
- Stock Market Forecasting Methods
- Anomaly Detection Techniques and Applications
- Gaussian Processes and Bayesian Inference
- Data Stream Mining Techniques
- Algorithms and Data Compression
- Neural Networks and Applications
- Computational Physics and Python Applications
- Machine Learning and Algorithms
- Bayesian Methods and Mixture Models
- Natural Language Processing Techniques
- Network Security and Intrusion Detection
- Complex Systems and Time Series Analysis
- Machine Learning and Data Classification
- Model Reduction and Neural Networks
- Energy Load and Power Forecasting
- Advanced Graph Neural Networks
- Machine Learning in Healthcare
- Data Analysis with R
- Data Management and Algorithms
- Machine Learning in Bioinformatics
- Bayesian Modeling and Causal Inference
- Advanced Multi-Objective Optimization Algorithms
- Advanced Chemical Sensor Technologies
Amazon (Germany)
2017-2022
Universität Bayern
2022
Amazon (United States)
2018-2021
University College London
2008-2017
Gatsby Charitable Foundation
2010
Oxford Centre for Computational Neuroscience
2008
Probabilistic forecasting, i.e., estimating a time series' future probability distribution given its past, is key enabler for optimizing business processes. In retail businesses, example, probabilistic demand forecasts are crucial having the right inventory available at and in place. This paper proposes DeepAR, methodology producing accurate forecasts, based on training an autoregressive recurrent neural network model large number of related series. We demonstrate how application deep...
We present a platform built on large-scale, data-centric machine learning (ML) approaches, whose particular focus is demand forecasting in retail. At its core, this enables the training and application of probabilistic models, provides convenient abstractions support functionality for problems. The comprises complex end-to-end system Apache Spark, which includes data preprocessing, feature engineering, distributed learning, as well evaluation, experimentation ensembling. Furthermore, it...
There is a large body of research on scalable machine learning (ML). Nevertheless, training ML models large, continuously evolving datasets still difficult and costly undertaking for many companies institutions. We discuss such challenges derive requirements an industrial-scale platform. Next, we describe the computational model behind Amazon SageMaker, which designed to meet challenges. SageMaker platform provided as part Web Services (AWS), supports incremental training, resumable elastic...
We introduce Neural Contextual Anomaly Detection (NCAD), a framework for anomaly detection on time series that scales seamlessly from the unsupervised to supervised setting, and is applicable both univariate multivariate series. This achieved by combining recent developments in representation learning series, with techniques deep originally developed computer vision we tailor setting. Our window-based approach facilitates boundary between normal anomalous classes injecting generic synthetic...
We propose an unbounded-depth, hierarchical, Bayesian nonparametric model for discrete sequence data. This can be estimated from a single training sequence, yet shares statistical strength between subsequent symbol predictive distributions in such way that performance generalizes well. The builds on specific parameterization of unbounded-depth hierarchical Pitman-Yor process. introduce analytic marginalization steps (using coagulation operators) to reduce this one represented time and space...
Probabilistic forecasting, i.e. estimating the probability distribution of a time series' future given its past, is key enabler for optimizing business processes. In retail businesses, example, forecasting demand crucial having right inventory available at place. this paper we propose DeepAR, methodology producing accurate probabilistic forecasts, based on training an auto regressive recurrent network model large number related series. We demonstrate how by applying deep learning techniques...
We introduce Gluon Time Series (GluonTS, available at https://gluon-ts.mxnet.io), a library for deep-learning-based time series modeling. GluonTS simplifies the development of and experimentation with models common tasks such as forecasting or anomaly detection. It provides all necessary components tools that scientists need quickly building new models, efficiently running analyzing experiments evaluating model accuracy.
Producing probabilistic forecasts for large collections of similar and/or dependent time series is a practically relevant and challenging task. Classical models fail to capture complex patterns in the data, multivariate techniques struggle scale problem sizes. Their reliance on strong structural assumptions makes them data-efficient, allows provide uncertainty estimates. The converse true based deep neural networks, which can learn dependencies given enough data. In this paper, we propose...
Predicting the dependencies between observations from multiple time series is critical for applications such as anomaly detection, financial risk management, causal analysis, or demand forecasting. However, computational and numerical difficulties of estimating time-varying high-dimensional covariance matrices often limits existing methods to handling at most a few hundred dimensions requires making strong assumptions on dependence series. We propose combine an RNN-based model with Gaussian...
The prevalence of approaches based on gradient boosted trees among the top contestants in M5 competition is potentially most eye-catching result. Tree-based methods out-shone other solutions, particular deep learning-based solutions. winners both tracks heavily relied them. This even more remarkable given dominance literature and M4 competition. article tries to explain why tree-based were so widely used We see possibilities for future improvements models then distill some learnings...
Time series forecasting is a key ingredient in the automation and optimization of business processes: retail, deciding which products to order where store them depends on forecasts future demand different regions; cloud computing, estimated usage services infrastructure components guides capacity planning; workforce scheduling warehouses, call centers, factories requires workload. Recent years have witnessed paradigm shift techniques applications, from computer-assisted model-...
Probabilistic models of sequences play a central role in most machine translation, automated speech recognition, lossless compression, spell-checking, and gene identification applications to name but few. Unfortunately, real-world sequence data often exhibit long range dependencies which can only be captured by computationally challenging, complex models. Sequence arising from natural processes also exhibits power-law properties, yet common do not capture such properties. The memoizer is new...
Time series forecasting is a key ingredient in the automation and optimization of business processes: retail, deciding which products to order where store them depends on forecasts future demand different regions; cloud computing, estimated usage services infrastructure components guides capacity planning; workforce scheduling warehouses factories requires workload. Recent years have witnessed paradigm shift techniques applications, from computer-assisted model- assumption-based data-driven...
In this work we describe a sequence compression method based on combining Bayesian nonparametric model with entropy encoding. The model, hierarchy of Pitman-Yor processes unbounded depth previously proposed by Wood et al. [16] in the context language modelling, allows modelling long-range dependencies allowing conditioning contexts length. We show that incremental approximate inference can be performed thereby it to used text setting. resulting compressor reliably outperforms several PPM...
Time series forecasting is a key ingredient in the automation and optimization of business processes: retail, deciding which products to order where store them depends on forecasts future demand different regions; cloud computing, estimated usage services infrastructure components guides capacity planning; workforce scheduling warehouses factories requires workload. Recent years have witnessed paradigm shift techniques applications, from computer-assisted model- assumption-based data-driven...