3CS Algorithm for Efficient Gaussian Process Model Retrieval

Berns, Fabian; Schmidt, Kjeld Willy; Bracht, Ingolf; Beecks, Christian

doi:10.1109/icpr48806.2021.9412805

Cited by 5 publications

(5 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Furthermore, as mentioned in Sect. 2, different kernel search algorithms exist [6,8,14,15]. For this proof of concept, we included a hyperparameter optimization (no change of the GP's kernel expression) after a detected change point as a comparison partner (CPD HPO).…”

Section: Discussionmentioning

confidence: 99%

“…An appropriate kernel expression can be chosen by an expert based on previous knowledge about the data. Without such expert knowledge, automatic kernel search algorithms can be employed to find an optimal fit for the given data [6,8,14,15]. In 2013, Duvenaud et al [8] introduced such an algorithm for the first time, i.e.…”

Section: Related Workmentioning

confidence: 99%

“…The problem of scalability was addressed in different ways: Kim et al [14] introduced the Scalable Kernel Composition (SKC) in 2018, which performs model selection via lower and upper bounds for the GPs' marginal likelihood instead of the exact evaluations. Berns et al developed new approaches that use a prior segmentation of the data to accelerate the search [5] or perform the segmentation themselves [6]. We aim to expand this idea by performing a dynamic segmentation via online CPD during the modelling process.…”

Section: Related Workmentioning

confidence: 99%

See 2 more Smart Citations

Dynamically Self-adjusting Gaussian Processes for Data Stream Modelling

Huwel

Haselbeck

Grimm

et al. 2022

Lecture Notes in Computer Science

Self Cite

View full text Add to dashboard Cite

One of the major challenges in time series analysis are changing data distributions, especially when processing data streams. To ensure an up-to-date model delivering useful predictions at all times, model reconfigurations are required to adapt to such evolving streams. For Gaussian processes, this might require the adaptation of the internal kernel expression. In this paper, we present dynamically self-adjusting Gaussian processes by introducing Event-Triggered Kernel Adjustments in Gaussian process modelling (ETKA), a novel data stream modelling algorithm that can handle evolving and changing data distributions. To this end, we enhance the recently introduced Adjusting Kernel Search with a novel online change point detection method. Our experiments on simulated data with varying change point patterns suggest a broad applicability of ETKA. On real-world data, ETKA outperforms comparison partners that differ regarding the model adjustment and its refitting trigger in nine respective ten out of 14 cases. These results confirm ETKA’s ability to enable a more accurate and, in some settings, also more efficient data stream processing via Gaussian processes.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Dynamically Self-adjusting Gaussian Processes for Data Stream Modelling

Huwel

Haselbeck

Grimm

et al. 2022

Lecture Notes in Computer Science

Self Cite

View full text Add to dashboard Cite

show abstract

“…The Gaussian process model is widely regarded as a prominent tool for capturing the inherent characteristics of data [1]. The Gaussian process model is a Bayesian machine learning model that has been widely applied in the pattern recognition domain due to its ability to infer from unreliable, noisy or highly idiosyncratic data [2]. Studies related to this Gaussian process have been found in various studies, for example [2][3][4].…”

Section: Introductionmentioning

confidence: 99%

“…The Gaussian process model is a Bayesian machine learning model that has been widely applied in the pattern recognition domain due to its ability to infer from unreliable, noisy or highly idiosyncratic data [2]. Studies related to this Gaussian process have been found in various studies, for example [2][3][4]. In [3], a two-phase Gaussian process degradation model with a change point, consisting of a change point model based on the Wiener process as a special case, is proposed to describe the degradation pathway with a two-phase pattern.…”

Section: Introductionmentioning

confidence: 99%

Signal Modeling with IG Noise and Parameter Estimation Based on RJMCMC

Fauzy¹,

Suparman²,

Supandi³

2022

View full text Add to dashboard Cite

Piecewise constant (PC) is a stochastic model that can be applied in various fields such as engineering and ecology. The stochastic model contains a noise. The accuracy of the stochastic model in modeling a signal is influenced by the type of noise. This paper aims to propose inverse-gamma noise in the PC model and the procedure for estimating the model parameters. The model parameters are estimated using the Bayes approach. Model parameters have a variable dimension space so that the Bayesian estimator cannot be determined analytically. Therefore, the Bayesian estimator is calculated using the reversible jump Markov Chain Monte Carlo (RJMCMC) algorithm. The performance of the RJMCMC algorithm is validated using data synthesis. The finding is a new PC model in which the noise has an inverse-gamma distribution. In addition, this paper also proposes a parameter estimation procedure for the model based on an RJMCMC. The simulation study shows that the model parameter estimators generated by this algorithm are close to the model parameter values. This paper concludes that inverse gamma noise can be used as an alternative noise in the PC model. The RJMCMC is categorized as a valid algorithm and can estimate the PC model parameters where the noise has an inverse-gamma distribution. The novelty in this paper is the development of a new stochastic model and the procedure for estimating the model parameters. In application, the findings in this paper have the potential to improve the suitability of the stochastic model to the signal.

show abstract