Abstract This work addresses the challenges of learning from Evolving Streaming Time Series, a scenario defined in streaming contexts by the need for continuous learning, managing temporal dependence, reacting to concept drifts, and avoiding catastrophic forgetting. Streaming Continual Learning (SCL) is an emerging paradigm specifically aimed at tackling these issues. To this end, SCL combines elements of Streaming Machine Learning, Continual Learning, and Time Series Analysis. The core idea of SCL is to maintain a progressively enriched representation that reflects everything learned so far. When a drift occurs, the model integrates newly emerging information and also selectively draws on earlier knowledge. In this context, the present work proposes Dynamic Continuous Progressive Neural Networks (DYNcPNN), a comprehensive and pioneering SCL embodiment built on our previously introduced cPNN architecture. DYNcPNN is a dynamic variant designed to adapt more effectively to concept drifts. Its key contribution is a mechanism that dynamically determines when to expand the architecture to incorporate new knowledge, reducing unnecessary complexity while preserving high performance. DYNcPNN also introduces a strategy to prevent forgetting when, after a concept drift, the model continues learning without expanding, thereby risking overwriting the existing knowledge. The model integrates an automatic concept drift detection system that enables adaptation without manual intervention. Experimental results show that DYNcPNN consistently outperforms traditional SML models, cPNN, and a continuously trained LSTM. It adapts more quickly to concept drifts, effectively mitigates catastrophic forgetting, and optimizes memory usage. Moreover, the results highlight the limitations of SML models, which are unable to account for temporal dependence.

Concept drift, temporal dependence, and catastrophic forgetting represent major challenges when learning from data streams. While Streaming Machine Learning and Continual Learning (CL) address these issues separately, recent efforts in Streaming Continual Learning (SCL) aim to unify them. In this work, we introduce MAGIC Net, a novel SCL approach that integrates CL-inspired architectural strategies with recurrent neural networks to tame temporal dependence. MAGIC Net continuously learns, looks back at past knowledge by applying learnable masks over frozen weights, and expands its architecture when necessary. It performs all operations online, ensuring inference availability at all times. Experiments on synthetic and real-world streams show that it improves adaptation to new concepts, limits memory usage, and mitigates forgetting.

Internet of Things (IoT) Analytics often involves applying machine learning (ML) models on data streams. In such scenarios, traditional ML paradigms face obstacles related to continuous learning while dealing with concept drifts, temporal dependence, and avoiding forgetting. Moreover, in IoT, different edge devices build up a network. When learning models on those devices, connecting them could be useful in improving performance and reusing others’ knowledge. This work proposes Mutual Assisted Learning, a learning paradigm grounded on Vygotsky’s popular Sociocultural Theory of Cognitive Development. Each device is autonomous and does not need a central orchestrator. Whenever it degrades its performance due to a concept drift, it asks for assistance from others and decides whether their knowledge is useful for solving the new problem. This way, the number of connections is drastically reduced compared to the classical Federated Learning approaches, where the devices communicate at each training round. Every device is equipped with a Continuous Progressive Neural Network (cPNN) to handle the dynamic nature of data streams. We call this implementation Mutual Assisted cPNN (MAcPNN). To implement it, we allow cPNNs for single data point predictions and apply quantization to reduce the memory footprint. Experimental results prove the effectiveness of MAcPNN in boosting performance on synthetic and real data streams.

In batch learning, it is commonly assumed that samples are independent and identically distributed (i.i.d.). However, this assumption does not hold in dynamic environments where data streams are not identically distributed due to concept drifts. Furthermore, while most Streaming Machine Learning (SML) literature assumes independence among examples, data streams often have important temporal components that learning should adequately consider. Neglecting this temporal dependence can lead to significant misguidance in designing and evaluating SML models. To support our thesis, we propose Tenet, a novel benchmarking framework designed to evaluate data stream classifiers in non-i.i.d. scenarios comparatively. Tenet consists of a data stream generator and a baseline. The data stream generator introduces temporal dependence into the data streams commonly used for evaluating SML algorithms. The baseline is a continuous version of the Long Short-Term Memory algorithm called cLSTM. Extensive experiments using Tenet demonstrate that cLSTM consistently outperforms state-of-the-art SML classifiers when learning from data streams with temporal dependence. This result is a call to action for the SML and the Deep Learning communities to investigate classifiers in the time-dependent streaming scenario and makes Tenet the first publicly available benchmark to support this research.

Abstract Dealing with an unbounded data stream involves overcoming the assumption that data is identically distributed and independent. A data stream can, in fact, exhibit temporal dependencies (i.e., be a time series), and data can change distribution over time (concept drift). The two problems are deeply discussed, and existing solutions address them separately: a joint solution is absent. In addition, learning multiple concepts implies remembering the past (a.k.a. avoiding catastrophic forgetting in Neural Networks’ terminology). This work proposes Continuous Progressive Neural Networks (cPNN), a solution that tames concept drifts, handles temporal dependencies, and bypasses catastrophic forgetting. cPNN is a continuous version of Progressive Neural Networks, a methodology for remembering old concepts and transferring past knowledge to fit the new concepts quickly. We base our method on Recurrent Neural Networks and exploit the Stochastic Gradient Descent applied to data streams with temporal dependencies. Results of an ablation study show a quick adaptation of cPNN to new concepts and robustness to drifts.