Machine learning and data mining

Welcome to the Machine Learning and Data Mining research group at IBM Research – Ireland. We carry out both fundamental and applied research in the fields of machine learning and data mining.

The broad spectrum of our activities ranges from theoretical analysis of machine learning methods, algorithmic development, and implementation of large scale systems. We are applying and developing techniques for multivariate regression, forecasting, clustering, classification, ranking, anomaly and change point detection problems.

We have significant experience in analyzing and extracting value from data originated in various domains, including electricity demand, water quality, transit, financial time series, images, pharmacology, social networks, and recommender systems.

Regularization techniques

We study machine learning techniques based on regularization and their application to a variety of problems, ranging from dynamical system identification [2014-10], medical image classification [2014-6], water quality data analysis [2013-2], and electric load forecasting [2012-1]. We are also interested in analyzing properties of the regularized solution [2014-1], and developing efficient optimization techniques to solve them on a large scale. We are particularly interested in developing multi-task and transfer learning techniques based on regularization, as well as methods to combine multiple heterogeneous datasets.

Electricity demand data analysis

An application area of particular interest to us is analysis, modeling and forecasting of electricity demand. We develop approaches for understanding demand at various disaggregation levels, including the identification of different customer behavior through clustering methods [2013-1]. Another line of work is on online learning algorithms to track trends and dynamic changes in demand patterns in real-time [2012-1]. Moreover, we study methods for detecting anomalies in demand patterns, e.g., through bootstrap confidence intervals [2014-2]. Ongoing work is on modeling uncertainty in demand, feature selection, and multitask learning to facilitate analysis of massive amounts of demand time series.

Clustering and dimensionality reduction

We have been applying clustering, feature analysis and dimensionality reduction techniques to several applications and domains. These include specific problems in smarter energy [2013-1], urban dynamics [2013-8] and biomedical data [2014-11]. The contributions are in the areas of semi-supervised learning [2014-8], feature selection and clustering stability [2014-7] and denoising [2014-11; 2013-8]. We focus our interest in developing predictive models for unsupervised and semi-supervised learning based on kernels with applicability to large-scale data. Ongoing work includes clustering large graphs coming from the natural language processing domain.

Time series analysis and management

From a practical point of view, we are interested in methods for analyzing time series that are (i) robust w.r.t. noise; (ii) computationally efficient; (iii) distribution-free; (iv) scalable to massive amounts of time series [2012-1; 2012-3; 2013-4; 2014-2; 2014-3]. A key focus is on non-linear regression models and their application to data from energy, transport and other smarter cities domains [2012-1; 2013-4; 2014-2]. On the theoretical side, we are working on understanding fundamental properties of Conditional Markov Chains, which are useful to model latent states of a system conditional on a sequence of exogenous variables [2012-2; 2013-9]. Another line of work is on efficient representations and scalable algorithms for mining streaming data [2013-6; 2013-7].

Mining graphs and social media

Analytics methods for the social web are becoming increasingly necessary to make sense out of the huge amount of user generated content and reduce the complexity for human user understanding. Current research greatly benefits from cross-disciplines, including machine learning, recommender systems, and computational social science. The integration of interdisciplinary evidence also represents an important ingredient of our work, and our research on mining massive graphs, collaborative filtering in social media streams, and collective intelligence, expands IBM capabilities to transform the huge mass of social media data into useful and powerful insight that will allow us to build more intelligent systems for the benefit of society and our customers.

References

[ 2014 | 2013 | 2012 ]

2014

[1] A. Argyriou and F. Dinuzzo
"A unifying view of representer theorems"
Proceedings of the 31th International Conference on Machine Learning (ICML 2014).

[2] B. Chen, M. Sinn, J. Ploennigs and A. Schumann
"Statistical anomaly detection in mean and variation of energy consumption"
International Conference on Pattern Recognition (ICPR), 2014.

[3] J. Choi, B. Chen and B. Abraham
"Simulated maximum likelihood in autoregressive model with stochastic volatility errors"
Applied Stochastic Models in Business and Industry, accepted.

[4] X. Dong, D. Mavroeidis, F. Calabrese and P. Frossard
"Multiscale event detection in social media"
arXivNeuroImage: Clinical, Volume 4, 2014.

[5] L. Drumond, E. Diaz-Aviles, L. Schmidt-Thieme and W. Nejdl
"Optimizing multi-relational factorization models for multiple target relations"
International Conference on Information and Knowledge Management (CIKM 2014).

[6] J.-B. Fiot, H. Raguet, L. Risser, L. D. Cohen, J. Fripp and F.-X. Vialard
"Longitudinal deformation models, spatial regularizations and learning strategies to quantify Alzheimer's disease progression"
NeuroImage: Clinical, Volume 4, 2014.

[7] D. Mavroeidis and E. Marchiori
"Feature selection for k-means clustering stability: theoretical analysis and an algorithm"
Data Mining and Knowledge Discovery 28(4): 918-960 (2014).

[8] S. Mehrkanoon, C. Alzate, R. Mall, R. Langone and J.A.K. Suykens
"Multiclass semi-supervised learning based upon kernel spectral clustering"
IEEE Transactions on Neural Networks and Learning Systems, 2014.

[9] C. Orellana-Rodriguez, E. Diaz-Aviles, W. Nejdl and I. Sengör Altingövde
"Learning to rank for joy"
WWW (Companion Volume) 2014.

[10] G. Pillonetto, F. Dinuzzo, T. Chen, G. De Nicolao and L. Ljung
"Kernel methods in system identification, machine learning and function estimation: A survey"
Automatica 50 (3), 657-682, 2014.

[11] C. Varon, C. Alzate and J.A.K. Suykens
"Noise level estimation for model selection in kernel PCA denoising"
IEEE Transactions on Neural Networks and Learning Systems, 2014.

2013

[1] C. Alzate and M Sinn
"Improved electricity load forecasting via kernel spectral clustering of smart meters"
IEEE 13th International Conference on Data Mining (ICDM), 943-948, 2013.

[2] A. Ba and S. McKenna
"Time-variant regularization in affine projection algorithms"
Allerton, 2013.

[3] M. Berlingerio, F. Pinelli and F. Calabrese
"ABACUS: frequent pAttern mining-BAsed Community discovery in mUltidimensional networkS"
Data Mining and Knowledge Discovery 27(3), 294-320, 2013.

[4] B. Chen, F. Pinelli, M. Sinn, A. Botea and F. Calabrese
"Uncertainty in urban mobility: Predicting waiting times for shared bicycles and parking lots"
International Conference on Intelligent Transportation System (ITSC), 2013.

[6] A. Marascu, P. Pompey, E. Bouillet, O. Verscheure, M. Wurst, M. Grund and P. Cudre-Mauroux
"MiSTRAL: An architecture for low-latency analytics on MasSive time series"
IEEE International Conference on Big Data, 2013.

[7] K. Mirylenka, A. Marascu, T. Palpanas, M. Fehr, S. Jank, G. Welde and D. Groeber
"Envelope-based anomaly detection for high-speed manufacturing processes"
European Advanced Process Control and Manufacturing Conference (APCM), 2013.

[8] F. Pinelli, F. Calabrese and E.P. Bouillet
"A methodology for denoising and generating bus infrastructure data"
Intelligent Transportation Systems-(ITSC), 2013.

[9] M. Sinn and B. Chen
"Central limit theorems for conditional Markov chains"
16th International Conference on Artificial Intelligence and Statistics (AISTATS 2013).


2012

[1] A. Ba, M. Sinn, Y. Goude and P. Pompey
"Adaptive learning of smoothing functions: Application to electricity load forecasting"
Advances in Neural Information Processing Systems (NIPS 2012).

[2] M. Sinn and B. Chen
"Mixing properties of conditional Markov chains with unbounded feature functions"
Advances in Neural Information Processing Systems (NIPS 2012).

[3] M. Sinn, A. Ghodsi and K. Keller
"Detecting change-points in time series by maximum mean discrepancy of ordinal pattern distributions"
28th Conference on Uncertainty in Artificial Intelligence (UAI 2012).