Proceedings Paper
Computer Science, Artificial Intelligence
Zach Wood-Doughty, Isabel Cachola, Mark Dredze
Summary: Research has shown that methods for interpreting black-box models through explaining predictions one-by-one can be slow and inconsistent, and ill-suited for time series applications. This study introduces a proxy model approach that is faster, faithful to the original model, and globally consistent in its explanations, showing improvements over existing methods in political event forecasting.
20TH IEEE INTERNATIONAL CONFERENCE ON MACHINE LEARNING AND APPLICATIONS (ICMLA 2021)
(2021)
Article
Economics
Dilini Rajapaksha, Christoph Bergmeir, Rob J. Hyndman
Summary: Global forecasting models (GFMs) have shown superior results compared to univariate forecasting approaches, but they lack interpretability, which reduces stakeholders' trust and confidence in the predictions. To address this issue, a novel local model-agnostic interpretability approach is proposed to explain the forecasts from GFMs.
INTERNATIONAL JOURNAL OF FORECASTING
(2023)
Proceedings Paper
Computer Science, Information Systems
Leonid Schwenke, Martin Atzmueller
Summary: This paper addresses the issue of enhancing interpretability and explainability of Multi-Headed Attention in Transformer models for time series classification. Introducing a method for constructing global coherence representations and providing abstraction and interpretation methods lead to intuitive visualizations of attention patterns. The proposed approach and methods were evaluated on various datasets, demonstrating their effectiveness.
2021 IEEE 8TH INTERNATIONAL CONFERENCE ON DATA SCIENCE AND ADVANCED ANALYTICS (DSAA)
(2021)
Article
Chemistry, Analytical
Dominique Mercier, Andreas Dengel, Sheraz Ahmed
Summary: The TimeREISE method is a model-agnostic attribution method for time series classification that outperforms existing methods in terms of different established measurements. It shows impressive results in deletion and insertion tests, infidelity, and sensitivity while maintaining the correctness of the attribution map.
Article
Computer Science, Artificial Intelligence
Michael Franklin Mbouopda, Engelbert Mephu Nguifo
Summary: This paper proposes a time series classification method using shapelets, which exploit the shared characteristics among members of the same class to improve the computational efficiency. Experimental results show that the proposed method achieves higher accuracy and scalability compared to the state of the art Shapelet Transform algorithm.
PATTERN RECOGNITION
(2024)
Article
Computer Science, Interdisciplinary Applications
Manjunatha Veerappa, Mathias Anneken, Nadia Burkart, Marco F. Huber
Summary: This paper introduces the use of explainable artificial intelligence (XAI) methods in ship type classification task, which provides explanations in terms of the features contributing the most towards the prediction along with their corresponding time intervals to support decision-making.
JOURNAL OF COMPUTATIONAL SCIENCE
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Raneen Younis, Sergej Zerr, Zahra Ahmadi
Summary: This paper proposes a method to interpret the outputs of CNN models by extracting and clustering activated time series sequences, improving the interpretability of black-box model predictions. Experimental results confirm significant improvements in the interpretability of network predictions and relation identification.
2022 IEEE 9TH INTERNATIONAL CONFERENCE ON DATA SCIENCE AND ADVANCED ANALYTICS (DSAA)
(2022)
Article
Computer Science, Information Systems
Pablo Zinemanas, Martin Rocamora, Marius Miron, Frederic Font, Xavier Serra
Summary: Deep learning models have advanced in many research areas, but their black-box structure limits understanding of their inner workings and predictions. A new interpretable deep learning model is proposed for automatic sound classification, explaining predictions based on similarities to learned prototypes and leveraging domain knowledge. The model achieves comparable results to state-of-the-art methods in speech, music, and environmental audio classification tasks, with automatic pruning methods available for interpretability.
Article
Computer Science, Artificial Intelligence
Antoine Pirovano, Hippolyte Heuberger, Sylvain Berlemont, SaId Ladjal, Isabelle Bloch
Summary: This paper addresses the issue of interpretability in deep learning methods for medical applications, proposing a novel approach for computing interpretable slide-level heat maps to enhance overall classification performance. Through validation and research, a corrective approach based on activation colocalization further improves the performance and stability of the proposed method.
MACHINE LEARNING AND KNOWLEDGE EXTRACTION
(2021)
Article
Computer Science, Artificial Intelligence
Kevin Fauvel, Elisa Fromont, Veronique Masson, Philippe Faverdin, Alexandre Termier
Summary: This paper presents XEM, an eXplainable-by-design Ensemble method for Multivariate time series classification. XEM combines an explicit boosting-bagging approach and an implicit divide-and-conquer approach to handle the bias-variance trade-off and individualize classifier errors. The evaluation shows that XEM outperforms state-of-the-art MTS classifiers and exhibits robust performance when facing challenges from continuous data collection.
DATA MINING AND KNOWLEDGE DISCOVERY
(2022)
Article
Mathematics
Kevin Fauvel, Tao Lin, Veronique Masson, Elisa Fromont, Alexandre Termier
Summary: This paper introduces XCM, a new compact convolutional neural network for MTS classification, which not only achieves good generalization performance on both large and small datasets, but also provides faithful explanations by precisely identifying important input data variables and timestamps.
Proceedings Paper
Computer Science, Artificial Intelligence
Zhendong Wang, Isak Samsten, Rami Mochaourab, Panagiotis Papapetrou
Summary: Counterfactual explanations can improve the interpretability of models and generate valid counterfactual explanations closer to the decision boundary in time series classification.
DISCOVERY SCIENCE (DS 2021)
(2021)
Proceedings Paper
Computer Science, Artificial Intelligence
Tilman Rauker, Anson Ho, Stephen Casper, Dylan Hadfield-Menell
Summary: The field of machine learning has made significant progress in the past decade, with the deployment of deep neural networks in real-world applications. However, the lack of understanding of the internal workings of these networks raises concerns. Tools for interpreting neural networks are important for improving trust in AI and enhancing our understanding of their behavior.
2023 IEEE CONFERENCE ON SECURE AND TRUSTWORTHY MACHINE LEARNING, SATML
(2023)
Proceedings Paper
Automation & Control Systems
Matthieu Bellucci, Nicolas Delestre, Nicolas Malandain, Cecilia Zanni-Merk
Summary: Explainable Artificial Intelligence (XAI) has gained popularity in recent years due to new legislations promoting the right to explanation. Despite the development of various methods to understand black-box models, there is still a lack of clarity on the definition of explanation, leading to a lack of consensus that hinders the field's progress.
KNOWLEDGE-BASED AND INTELLIGENT INFORMATION & ENGINEERING SYSTEMS (KSE 2021)
(2021)
Article
Computer Science, Artificial Intelligence
Hasan A. Bedel, Irmak Sivgin, Onat Dalmaz, Salman U. H. Dar, Tolga Cukur
Summary: BolT is a blood-oxygen-level-dependent Transformer model for analyzing multi-variate fMRI time series. It captures contextual representations across diverse time scales using a cascade of Transformer encoders equipped with a novel fused window attention mechanism. The comprehensive experiments on large-scale public datasets demonstrate the superior performance of BolT against state-of-the-art methods.
MEDICAL IMAGE ANALYSIS
(2023)
Article
Computer Science, Artificial Intelligence
Antonis Matakos, Sijing Tu, Aristides Gionis
KNOWLEDGE AND INFORMATION SYSTEMS
(2020)
Article
Computer Science, Artificial Intelligence
Polina Rozenshtein, Nikolaj Tatti, Aristides Gionis
Summary: This paper investigates the problem of determining entity activity based on interactions, proposing two formulations and efficient algorithms for untangling networks. While the sum problem is shown to be NP-hard, the max problem can be solved optimally in linear time. In cases of multiple activity intervals per entity, both formulations are proved to be inapproximable but efficient algorithms based on alternative optimization are proposed. Evaluation on synthetic and real-world datasets supports the validity of concepts and performance of algorithms.
DATA MINING AND KNOWLEDGE DISCOVERY
(2021)
Article
Computer Science, Artificial Intelligence
Antonis Matakos, Aristides Gionis
Summary: Online social networks offer numerous benefits such as establishing new connections, gaining knowledge about the world, exposure to diverse viewpoints, and access to previously inaccessible information. This research focuses on leveraging the triadic closure principle to develop methods that foster new connections and improve the flow of information in the network.
DATA MINING AND KNOWLEDGE DISCOVERY
(2022)
Article
Computer Science, Artificial Intelligence
Bruno Ordozgoiti, Ananth Mahadevan, Antonis Matakos, Aristides Gionis
Summary: When searching for information in a data collection, it is often important to not only find relevant items but also assemble a diverse set to explore different concepts in the data. This paper addresses the problem of finding a diverse set of items when item relatedness is measured by a similarity function. The authors propose a new minimization objective and employ a randomized rounding strategy to find good solutions efficiently. They also introduce a novel bound for the ratio of Poisson-Binomial densities, which has applications beyond this problem. The proposed algorithm outperforms greedy approaches commonly used in the literature according to experiments on benchmark datasets.
DATA MINING AND KNOWLEDGE DISCOVERY
(2022)
Article
Computer Science, Artificial Intelligence
Guangyi Zhang, Nikolaj Tatti, Aristides Gionis
Summary: Submodular maximization is fundamental in many important machine learning problems and has various applications. However, the study of maximizing submodular functions has often been limited to selecting a set of items, while many real-world applications require a ranking solution. This paper introduces a novel formulation for ranking items with submodular valuations and budget constraints, and proposes practical algorithms with approximation guarantees for different types of budget constraints. The empirical evaluation shows that the proposed algorithms outperform strong baselines.
DATA MINING AND KNOWLEDGE DISCOVERY
(2022)
Article
Computer Science, Artificial Intelligence
Antonis Matakos, Cigdem Aslay, Esther Galbrun, Aristides Gionis
Summary: Social-media platforms have provided new ways for citizens to participate in public debates and stay informed. This paper proposes a novel approach to maximize the diversity of exposure in a social network, ensuring citizens are exposed to diverse viewpoints for a healthy information sharing environment.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2022)
Proceedings Paper
Computer Science, Information Systems
Martino Ciaperoni, Aristides Gionis, Athanasios Katsamanis, Panagiotis Karras
Summary: This paper presents an algorithm called SIEVE, which is an improvement on the Viterbi algorithm to address the issue of its space complexity growing with the number of observations. SIEVE improves space efficiency by discarding and recomputing parts of the DP solution, without incurring a time complexity overhead.
PROCEEDINGS OF THE 2022 INTERNATIONAL CONFERENCE ON MANAGEMENT OF DATA (SIGMOD '22)
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Suhas Thejaswi, Bruno Ordozgoiti, Aristides Gionis
Summary: The study introduces a novel problem of diversity-aware clustering, where potential cluster centers belong to groups defined by protected attributes. It shows that the diversity-aware k-median problem is NP-hard in general cases but approximation algorithms can be obtained when facility groups are disjoint. Experimentally, approximation methods are evaluated for tractable cases, and a relaxation-based heuristic is provided for theoretically intractable scenarios.
MACHINE LEARNING AND KNOWLEDGE DISCOVERY IN DATABASES, ECML PKDD 2021: RESEARCH TRACK, PT II
(2021)
Proceedings Paper
Computer Science, Information Systems
Cigdem Aslay, Martino Ciaperoni, Aristides Gionis, Michael Mathioudakis
Summary: Bayesian networks are probabilistic models capturing dependencies among variables, with Variable Elimination being a fundamental algorithm for probabilistic inference. This paper proposes a novel materialization method to enhance efficiency in processing inference queries. Experimental results show that moderate materialization can significantly improve query running time.
2021 IEEE 37TH INTERNATIONAL CONFERENCE ON DATA ENGINEERING (ICDE 2021)
(2021)
Article
Computer Science, Artificial Intelligence
Gennady Andrienko, Natalia Andrienko, Chiara Boldrini, Guido Caldarelli, Paolo Cintia, Stefano Cresci, Angelo Facchini, Fosca Giannotti, Aristides Gionis, Riccardo Guidotti, Michael Mathioudakis, Cristina Ioana Muntean, Luca Pappalardo, Dino Pedreschi, Evangelos Pournaras, Francesca Pratesi, Maurizio Tesconi, Roberto Trasarti
Summary: The exponential growth of large-scale mobility data has led to the vision of smart cities but also raised privacy concerns. Research communities and industrial stakeholders show strong interest in building knowledge discovery pipelines over these data sources.
INTERNATIONAL JOURNAL OF DATA SCIENCE AND ANALYTICS
(2021)
Proceedings Paper
Computer Science, Information Systems
Aristides Gionis, Antonis Matakos, Bruno Ordozgoiti, Han Xiao
WWW'20: COMPANION PROCEEDINGS OF THE WEB CONFERENCE 2020
(2020)
Proceedings Paper
Computer Science, Artificial Intelligence
Ryuta Matsuno, Aristides Gionis
PROCEEDINGS OF THE 2020 SIAM INTERNATIONAL CONFERENCE ON DATA MINING (SDM)
(2020)
Proceedings Paper
Computer Science, Artificial Intelligence
Suhas Thejaswi, Aristides Gionis
PROCEEDINGS OF THE 2020 SIAM INTERNATIONAL CONFERENCE ON DATA MINING (SDM)
(2020)
Proceedings Paper
Computer Science, Artificial Intelligence
Guangyi Zhang, Aristides Gionis
PROCEEDINGS OF THE 2020 SIAM INTERNATIONAL CONFERENCE ON DATA MINING (SDM)
(2020)
Proceedings Paper
Computer Science, Information Systems
Han Xiao, Bruno Ordozgoiti, Aristides Gionis
WEB CONFERENCE 2020: PROCEEDINGS OF THE WORLD WIDE WEB CONFERENCE (WWW 2020)
(2020)