An Appraisal of Incremental Learning Methods

Abstract

As a special case of machine learning, incremental learning can acquire useful knowledge from incoming data continuously while it does not need to access the original data. It is expected to have the ability of memorization and it is regarded as one of the ultimate goals of artificial intelligence technology. However, incremental learning remains a long term challenge. Modern deep neural network models achieve outstanding performance on stationary data distributions with batch training. This restriction leads to catastrophic forgetting for incremental learning scenarios since the distribution of incoming data is unknown and has a highly different probability from the old data. Therefore, a model must be both plastic to acquire new knowledge and stable to consolidate existing knowledge. This review aims to draw a systematic review of the state of the art of incremental learning methods. Published reports are selected from Web of Science, IEEEXplore, and DBLP databases up to May 2020. Each paper is reviewed according to the types: architectural strategy, regularization strategy and rehearsal and pseudo-rehearsal strategy. We compare and discuss different methods. Moreover, the development trend and research focus are given. It is concluded that incremental learning is still a hot research area and will be for a long period. More attention should be paid to the exploration of both biological systems and computational models.

Identifying Minimum Information Requirements to Improve Integrated Modeling Capabilities: Lessons Learned From Dynamic Adaptive Policy Pathways

“…This is aligned with the idea of incremental learning, notably emphasized in the ML field, and identifying fit-for-purpose solutions for solving incremental learning tasks (Luo et al, 2020). Such incremental learning assumes ongoing adaptation of models to learning (Luo et al, 2020) and allows for rapid feedback when evaluating planning assumptions and the identification of potential areas to explore prior to considering further investments (Fu et al, 2020;Zare et al, 2020). In addition, it assumes a flexible structure when considering solutions, with implications for model interoperability (Madni and Sievers, 2014), especially when considering implementationbased solutions.…”

“…The use of incremental solutions of varying complexity encourages an ordering of solutions that tries to meet user needs firstly by model reuse, progressing to more costly changes to model structure and behavior only when the need for the change is established. This is aligned with the idea of incremental learning, notably emphasized in the ML field, and identifying fit-for-purpose solutions for solving incremental learning tasks (Luo et al, 2020). Such incremental learning assumes ongoing adaptation of models to learning (Luo et al, 2020) and allows for rapid feedback when evaluating planning assumptions and the identification of potential areas to explore prior to considering further investments (Fu et al, 2020;Zare et al, 2020).…”

The Ensemble of Unsupervised Incremental Learning Algorithm for Time Series Data

“…A number of papers are published on incremental learning. According to literature [1][2][3], it is a process of adopting new behavior on arrival of an incoming data stream. It becomes important to improve the accuracy of the model prediction in interactive environments with the help of human feedback in a timely fashion as described Rico-Juan and Inesta [4].…”

An Online Security Prediction and Control Framework for Modern Power Grids

“…The incremental Naïve Bayes algorithm is used to realize the online classification model in this paper. The basic idea of the incremental Naive Bayes algorithm is to calculate a posterior probability based on the prior probability and new data [51]. The ability of the incremental Naive Bayes classification algorithm to support online learning is due to its leverage on and exploitation of prior information of the datasets.…”

“…However, only a few tools support online applications. Weka software is one of the few tools that provides an online machine learning application development environment [51]. In particularly demanding real-world applications like security state prediction for the grid operators, the Weka environment can be used to produce real-time predictions.…”