2008–, with V. Tresp, A. Bifet and H. M. Gomes.
If deductive stream reasoning asks what can be derived from a stream given what we already know, the inductive side asks the opposite question: what can be learned from the stream itself, while it flows.
The line began with early work with Volker Tresp on combining stream reasoning and machine learning, applied to traffic forecasting and social media analytics, and grew into Streaming Machine Learning with Albert Bifet and Heitor Murilo Gomes — models that train incrementally, one sample at a time, and must cope with a world that changes underneath them. It also covers Time Series Analytics, time-evolving analytics and Edge AI, where the constraint is not only that data arrives continuously but that the model has to run where the data is produced. The work on temporal dependence in data streams that started here is what eventually opened the Streaming Continual Learning line.
Selected works
-
Tracking Adaptation Time: Metrics for Temporal Distribution Shift
Lorenzo Iovine, Giacomo Ziffer, Emanuele Della Valle
1st Streaming Continual Learning Bridge at AAAI 2026, Singapore, 2026Proposes metrics to measure how long a model takes to recover after a temporal distribution shift, so that adaptation speed can be tracked and compared, not only final accuracy.
-
Tenet: Benchmarking Data Stream Classifiers in Presence of Temporal Dependence — Best Student Paper Runner-Up Award, IEEE BigData 2024
Giacomo Ziffer, Federico Giannini, Emanuele Della Valle
IEEE International Conference on Big Data, BigData 2024, Washington, DC, USA, December 15-18, 2024, 2024Shows that ignoring temporal dependence can mislead both the design and the evaluation of Streaming Machine Learning classifiers. Tenet injects temporal dependence into the streams commonly used as benchmarks and provides a continuous LSTM baseline, which consistently outperforms state-of-the-art streaming classifiers; it is the first public benchmark for this setting.
-
Choosing the Right Time to Learn Evolving Data Streams
Alessio Bernardo, Emanuele Della Valle, Albert Bifet
IEEE International Conference on Big Data, BigData 2023, Sorrento, Italy, December 15-18, 2023, 2023Asks whether a streaming learner really needs every sample to update its model. It proposes OE-SPL, an ensemble meta-strategy that combines online ensembles with the Spaced Learning heuristic to learn concepts without using all samples: it matches state-of-the-art ensembles in accuracy while recovering from repeated concept drifts faster and with less time and memory.
-
Towards time-evolving analytics: Online learning for time-dependent evolving data streams
Giacomo Ziffer, Alessio Bernardo, Emanuele Della Valle, Vítor Cerqueira, Albert Bifet
Data Sci., 2023A position paper arguing that Streaming Machine Learning and Time Series Analytics each solve only half of the problem: the former relaxes the identical-distribution assumption, the latter the independence one. It lays the foundations of Time-Evolving Analytics, which keeps models relevant in real time when data are both evolving and temporally dependent.
-
An extensive study of C-SMOTE, a Continuous Synthetic Minority Oversampling Technique for Evolving Data Streams
Alessio Bernardo, Emanuele Della Valle
Expert Syst. Appl., 2022An extensive experimental study of C-SMOTE, the continuous version of SMOTE that rebalances imbalanced data streams on the fly, so that streaming learners keep their performance on the minority class even when the stream evolves.
-
BOTTARI: An augmented reality mobile application to deliver personalized and location-based recommendations by continuous analysis of social media streams
Marco Balduini, Irene Celino, Daniele Dell’Aglio, Emanuele Della Valle, Yi Huang, Tony Kyung-il Lee, Seon-Ho Kim, Volker Tresp
J. Web Semant., 2012BOTTARI, an augmented-reality mobile application for the Insadong district of Seoul that recommends places by continuously analysing the social media stream about them, combining deductive stream reasoning over an ontology of points of interest with inductive analysis of what people say. It won the Semantic Web Challenge 2011.
-
Semantic Traffic-Aware Routing Using the LarKC Platform
Emanuele Della Valle, Irene Celino, Daniele Dell’Aglio, Ralph Grothmann, Florian Steinke, Volker Tresp
IEEE Internet Comput., 2011Early work with V. Tresp in the LarKC project: semantic, traffic-aware routing in Milan that combines stream reasoning with machine-learning forecasts of urban traffic.
-
Deductive and Inductive Stream Reasoning for Semantic Social Media Analytics
Davide Francesco Barbieri, Daniele Braga, Stefano Ceri, Emanuele Della Valle, Yi Huang, Volker Tresp, Achim Rettinger, Hendrik Wermser
IEEE Intell. Syst., 2010Early work with V. Tresp that combines deductive stream reasoning and inductive machine learning to analyse social media streams, showing how the two forms of reasoning complement each other.