Incremental on-line learning: A review and comparison of state of the art algorithms

Losing, Viktor; Hammer, Barbara; Wersing, Heiko

doi:10.1016/j.neucom.2017.06.084

Cited by 298 publications

(206 citation statements)

References 59 publications

Supporting

Mentioning

181

Contrasting

Unclassified

Order By: Relevance

“…2) Real-world Data Streams: There is a consensus among researchers that the locations and/or the presence of concept drift in the ELECTRICITY, FOREST COVERTYPE, and POKER HAND data streams are not known [4], [9], [20], [28], [36]. This implies, in turn, that the drift detection delay as well as the false positive and false negative rates cannot be determined since the knowledge of the drift locations is necessary in order to evaluate these quantities.…”

Section: Experiments and Discussionmentioning

confidence: 99%

McDiarmid Drift Detection Methods for Evolving Data Streams

Pesaranghader

Viktor

Paquet

2018

2018 International Joint Conference on Neural Networks (IJCNN)

View full text Add to dashboard Cite

Increasingly, Internet of Things (IoT) domains, such as sensor networks, smart cities, and social networks, generate vast amounts of data. Such data are not only unbounded and rapidly evolving. Rather, the content thereof dynamically evolves over time, often in unforeseen ways. These variations are due to so-called concept drifts, caused by changes in the underlying data generation mechanisms. In a classification setting, concept drift causes the previously learned models to become inaccurate, unsafe and even unusable. Accordingly, concept drifts need to be detected, and handled, as soon as possible. In medical applications and emergency response settings, for example, change in behaviours should be detected in near real-time, to avoid potential loss of life. To this end, we introduce the McDiarmid Drift Detection Method (MDDM), which utilizes McDiarmid's inequality [1] in order to detect concept drift. The MDDM approach proceeds by sliding a window over prediction results, and associate window entries with weights. Higher weights are assigned to the most recent entries, in order to emphasize their importance. As instances are processed, the detection algorithm compares a weighted mean of elements inside the sliding window with the maximum weighted mean observed so far. A significant difference between the two weighted means, upper-bounded by the McDiarmid inequality, implies a concept drift. Our extensive experimentation against synthetic and real-world data streams show that our novel method outperforms the state-of-the-art. Specifically, MDDM yields shorter detection delays as well as lower false negative rates, while maintaining high classification accuracies.

show abstract

Section: Experiments and Discussionmentioning

confidence: 99%

McDiarmid Drift Detection Methods for Evolving Data Streams

Pesaranghader

Viktor

Paquet

2018

2018 International Joint Conference on Neural Networks (IJCNN)

View full text Add to dashboard Cite

show abstract

“…The modulation function is denoted as f ± here, omitting its arguments. Terms of order O(1/N ) have been discarded; note that (ξ µ ) 2 = O(N ) according to (4).…”

Section: Mathematical Analysismentioning

confidence: 99%

“…Similarly, in technical contexts, training data is often available in the form of non-stationary data streams, e.g. [1,4,5,6,7]. Two major types of nonstationary environments have been discussed in the literature: In virtual drifts, statistical properties of the available training data are time-dependent, while the actual target task remains unchanged.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Prototype-Based Classifiers in the Presence of Concept Drift: A Modelling Framework

Biehl

Abadi

Göpfert

et al. 2019

Advances in Intelligent Systems and Computing

Self Cite

View full text Add to dashboard Cite

We present a modelling framework for the investigation of prototype-based classifiers in non-stationary environments. Specifically, we study Learning Vector Quantization (LVQ) systems trained from a stream of high-dimensional, clustered data. We consider standard winnertakes-all updates known as LVQ1. Statistical properties of the input data change on the time scale defined by the training process. We apply analytical methods borrowed from statistical physics which have been used earlier for the exact description of learning in stationary environments. The suggested framework facilitates the computation of learning curves in the presence of virtual and real concept drift. Here we focus on timedependent class bias in the training data. First results demonstrate that, while basic LVQ algorithms are suitable for the training in non-stationary environments, weight decay as an explicit mechanism of forgetting does not improve the performance under the considered drift processes.

show abstract

“…Incremental learning and support vector regression is the most popular combined techniques. Incremental support vector machine uses partial support vectors as the "candidate vectors", while the final support vectors are selected depending on the newly-added samples [36]. Incremental learning is useful for support vector regression method, especially giving the consideration that the computational memory of the SVR increases significantly for a large dataset.…”

Section: Model Updates (B3)mentioning

confidence: 99%

The Application of Cyber Physical System for Thermal Power Plants: Data-Driven Modeling

Yang

et al. 2018

Energies

View full text Add to dashboard Cite

Optimal operation of energy systems plays an important role to enhance their lifetime security and efficiency. The determination of optimal operating strategies requires intelligent utilization of massive data accumulated during operation or prediction. The investigation of these data solely without combining physical models may run the risk that the established relationships between inputs and outputs, the models which reproduce the behavior of the considered system/component in a wide range of boundary conditions, are invalid for certain boundary conditions, which never occur in the database employed. Therefore, combining big data with physical models via cyber physical systems (CPS) is of great importance to derive highly-reliable and -accurate models and becomes more and more popular in practical applications. In this paper, we focus on the description of a systematic method to apply CPS to the performance analysis and decision making of thermal power plants. We proposed a general procedure of CPS with both offline and online phases for its application to thermal power plants and discussed the corresponding methods employed to support each sub-procedure. As an example, a data-driven model of turbine island of an existing air-cooling based thermal power plant is established with the proposed procedure and demonstrates its practicality, validity and flexibility. To establish such model, the historical operating data are employed in the cyber layer for modeling and linking each physical component. The decision-making procedure of optimal frequency of air-cooling condenser is also illustrated to show its applicability of online use. It is concluded that the cyber physical system with the data mining technique is effective and promising to facilitate the real-time analysis and control of thermal power plants.

show abstract

Incremental on-line learning: A review and comparison of state of the art algorithms

Cited by 298 publications

References 59 publications

McDiarmid Drift Detection Methods for Evolving Data Streams

McDiarmid Drift Detection Methods for Evolving Data Streams

Prototype-Based Classifiers in the Presence of Concept Drift: A Modelling Framework

The Application of Cyber Physical System for Thermal Power Plants: Data-Driven Modeling

Contact Info

Product

Resources

About