This paper proposes an assessor-guided learning strategy for continual learning where an assessor guides the learning process of a base learner by controlling the direction and pace of the learning process thus allowing an efficient learning of new environments while protecting against the catastrophic interference problem. The assessor is trained in a meta-learning manner with a meta-objective to boost the learning process of the base learner. It performs a soft-weighting mechanism of every sample accepting positive samples while rejecting negative samples. The training objective of a base learner is to minimize a meta-weighted combination of the cross entropy loss function, the dark experience replay (DER) loss function and the knowledge distillation loss function whose interactions are controlled in such a way to attain an improved performance. A compensated over-sampling (COS) strategy is developed to overcome the class imbalanced problem of the episodic memory due to limited memory budgets. Our approach, Assessor-Guided Learning Approach (AGLA), has been evaluated in the class-incremental and task-incremental learning problems. AGLA achieves improved performances compared to its competitors while the theoretical analysis of the COS strategy is offered. Source codes of AGLA, baseline algorithms and experimental logs are shared publicly in https://github.com/anwarmaxsum/AGLA for further study.

Evolving classifiers and especially evolving fuzzy classifiers have been established as a prominent technique for addressing the recent demands in building classifiers in an incremental open-loop manner, e.g. for the purpose of processing data streams online. So far, the focus lies on classifiers which are obtained based on the input and/or feedback in the form of target labels provided by a single user/expert. In this paper, we propose three variants of evolving multi-user fuzzy classifier systems (EFCS-MU), where multiple users may provide their label feedback: i) ensembled single-user classifiers system, which allows a separate classifier training per user and embeds an advanced aggregation strategy (→ ensembling on a model level), ii) consensus all-user classifier system, where a joint classifier is established for all users based on consensus labelings (→ ensembling on a label level), iii) shift-work all-user classifier system, where a joint classifier is established for all users based on the classical shift-work concept. The classifiers are incrementally evolved by a single-pass learning approach embedding the autonomous evolution of new rules on demand; it integrates an unsupervised evolving clustering technique for rule partitioning, thus the same partition is established in all single-user classifiers, only the consequents in the form of class confidence vectors typically differ among the users due to their different labelings. This offers direct explainability of the varying users’ annotation behaviors. The possible different experience levels of the users in relation to the process behind and possible ambiguities among the provided users’ labels are handled by the proper integration of uncertainty levels into the update of the classifier(s). Furthermore, a concept is presented as to how to adequately integrate possibly available expert rules for a particular newly (on-the-fly) arising class (or in advance for several classes). Finally, an on-line active learning (oAL) strategy is demonstrated, to select only the most important samples to be labelled and thus reduce users’ labeling costs, ensuring economically practicable applicability. The approach was successfully evaluated on two real-world application scenarios, one stemming for a visual inspection scenario, where four users check the quality of the imprint of compact discs and are affected by different experience levels, and one from a nursery school employment ranking application, where a new class was introduced later. The results provide insights into the performance behavior of the three different multi-user classifier variants under different circumstances (with and without expert rules, uncertainty integration, different labelling budgets etc.), including comparisons based on on-line accuracy trends versus the economy of the labeling effort.

Multi-label classification has attracted much attention in the machine learning community to address the problem of assigning single samples to more than one (not necessarily non-overlapping) class at the same time. We propose an evolving multi-label fuzzy classifier (EFC-ML) which is able to self-adapt and self-evolve its structure with new incoming multi-label samples in an incremental, single-pass manner. It is based on a multi-output Takagi–Sugeno type architecture, where for each class a separate consequent hyper-plane is defined, which yields flexibility for partially approximating the respective classes in a binary [0,1]-regression context. The learning procedure embeds a locally weighted incremental correlation-based algorithm combined with (conventional) recursive fuzzily weighted least squares and Lasso-based regularization. Locality is important to avoid the out-masking effect of single class labels in one or more rules; the correlation-based part ensures that the interrelations between class labels, a specific well-known property in multi-label classification for improved performance, are preserved properly; the Lasso-based regularization reduces the curse of dimensionality effects in the case of a higher number of inputs. Antecedent learning is achieved by product-space clustering and conducted for all class labels together, which yields a single rule base (opposed to related techniques such as one-versus-rest or classifier chaining, achieving multiple different rule bases, one per class), allowing a compact knowledge view and thus enabling better interpretable insights. Furthermore, our approach comes with an online active learning (AL) strategy for updating the classifier on just a (smaller) number of selected samples, which in turn makes the approach applicable for scarcely labelled streams in applications, where the annotation effort is typically expensive. It is based on three essential concepts: novelty content in the antecedent space, uncertainty due to ambiguity in the consequent (output) space and parameter instability reduction, and these in combination with an upper-allowed selection budget (which could be predefined by a user). Our approach was evaluated on several data sets from the MULAN repository and showed significantly improved classification accuracy and average precision trend lines compared to (evolving) one-versus-rest or classifier chaining concepts. A significant result was that, due to the online AL method, a 90% reduction in the number of samples used for classifier updates had little effect on the accumulated accuracy trend lines compared to a full update in most data set cases.

During the last 15 to 20 years, evolving (neuro-) fuzzy systems (E(N) FS) have enjoyed more and more attraction in the context of data stream mining and modeling processes. This is because they can be updated on the fly in a single-pass sample-wise manner and are able to perform autonomous changes of the models on structural level in order to react onto process drifts. A wide variety of evolving (neuro-) fuzzy systems approaches have been proposed in order to handle data stream mining and modeling processes by dynamically updating the rule structure and antecedents. The current denominator in the update of the consequent (output weight) parameters is the usage of the recursive (fuzzily weighted) least squares estimator (R(FW) LS), as being applied in almost all E(N) FS approaches. In this paper, we propose and examine alternative variants for consequent parameter updates, namely multi-innovation RFWLS, recursive correntropy and especially recursive weighted total least squares (RWTLS). Multi-innovation RFWLS guarantees more stability in the update whenever structural changes (i.e. changes in the antecedents) in the E(N) FS are performed. This is because rule membership degrees are actualized on (a portion of) past samples and properly integrated in each update step. Recursive correntropy addresses the problematic of outliers by down-weighing the influence of higher errors in the parameter updates. Recursive weighted total least squares also takes into account a possible noise level in the input variables (and not solely in the target variable as done in RFWLS). The approaches are compared with standard RFWLS i.) on three data stream regression problems from practical applications, which are affected by noise levels and where one embeds a known drift, and ii.) on a time-series based forecasting problem. The results based on accumulated prediction error trends over time indicate that RFWLS can be largely outperformed by the proposed alternative variants, and this with even lower sensitivity on various data noise levels. So, the proposed variants could be worth of being further considered as promising and serious alternatives.

Evolving fuzzy systems (EFS) have received increased attention from the community for the purpose of data stream modeling in an incremental, single-pass and transparent manner. To date, a wide variety of EFS approaches have been developed and successfully used in real-world applications which address structural evolution and parameter adaptation in single EFS models. We propose a specific ensemble scheme of EFS to increase their robustness in predictive performance on new stream samples. Our approach relies on an online variant of bagging in which various EFS ensemble members are generated from online bags, that is, the members are updated based on a specific probabilistic online sampling technique, and this with guaranteed convergence to classical sampling in batch bagging. The autonomous pruning of ensemble members is undertaken to omit undesired members with atypically higher errors than other members. We propose two variants, hard pruning where undesired members are deleted forever from the ensemble, and soft pruning where members receive weights to calculate the overall ensemble prediction, according to their single performance; thus, members who are undesired at a certain point of time may be dynamically recalled at a later stage. The autonomous evolution of new ensemble members is carried out whenever a drift in the stream is detected, based on a significantly worsening performance indicator, measured in terms of the Hoeffding inequality. Newer members typically represent the drifted state better and are thus up-weighed compared to older members within an advanced (weighted) calculation of the overall ensemble prediction. The new approach termed online bagged EFS (OB-EFS) was successfully evaluated and compared with single EFS models and related SoA approaches on four data streams from real-world applications (containing various noise levels, drifts and new operating conditions) and showed significantly lower prediction error trend lines.

Transferring knowledge across many streaming processes remains an uncharted territory in the existing literature and features unique characteristics: no labelled instance of the target domain, covariate shift of source and target domain, different period of drifts in the source and target domains. Autonomous transfer learning (ATL) is proposed in this paper as a flexible deep learning approach for the online unsupervised transfer learning problem across many streaming processes. ATL offers an online domain adaptation strategy via the generative and discriminative phases coupled with the KL divergence based optimization strategy to produce a domain invariant network while putting forward an elastic network structure. It automatically evolves its network structure from scratch with/without the presence of ground truth to overcome independent concept drifts in the source and target domain. Rigorous numerical evaluation has been conducted along with comparison against recently published works. ATL demonstrates improved performance while showing significantly faster training speed than its counterparts.