Ruy Luiz Milidiú

Wearable computing: accelerometers' data classification of body postures and movements

During the last 5 years, research on Human Activity Recognition (HAR) has reported on systems showing good overall recognition performance. As a consequence, HAR has been considered as a potential technology for e-health systems. Here, we propose a machine learning based HAR classifier. We also provide a full experimental description that contains the HAR wearable devices setup and a public domain dataset comprising 165,633 samples. We consider 5 activity classes, gathered from 4 subjects wearing accelerometers mounted on their waist, left thigh, right arm, and right ankle. As basic input features to our classifier we use 12 attributes derived from a time window of 150ms. Finally, the classifier uses a committee AdaBoost that combines ten Decision Trees. The observed classifier accuracy is 99.4%.

Time-series forecasting through wavelets transformation and a mixture of expert models

Abstract This paper describes a system formed by a mixture of expert models (MEM) for time-series forecasting. We deal with several different competing models, such as partial least squares, K-nearest neighbours and carbon copy. The input space, after changing its base using the Haar wavelets transform, is partitioned into disjoint regions by a clustering algorithm. For each region, a benchmark is performed among the different competing models aiming at selecting the most adequate one. MEM has improved the forecast performance when compared with the single models as experimentally demonstrated through two different time series: laser data and exchange rate data.

Portuguese Part-of-Speech Tagging Using Entropy Guided Transformation Learning

Entropy Guided Transformation Learning (ETL) is a new machine learning strategy that combines the advantages of Decision Trees (DT) and Transformation Based Learning (TBL). In this work, we apply the ETL framework to Portuguese Part-of-Speech Taggging. We use two different corpora: Mac-Morpho and Tycho Brahae. ETL achieves the best results reported so far for Machine Learning based POS tagging of both corpora. ETL provides a new training strategy that accelerates transformation learning. For the Mac-Morpho corpus this corresponds to a factor of three speedup. ETL shows accuracies of 96.75% and 96.64% for Mac-Morpho and Tycho Brahae, respectively.

Data stream anomaly detection through principal subspace tracking

We consider the problem of anomaly detection in multiple co-evolving data streams. In this paper, we introduce FRAHST (Fast Rank-Adaptive row-Householder Subspace Tracking). It automatically learns the principal subspace from N numerical data streams and an anomaly is indicated by a change in the number of latent variables. Our technique provides state-of-the-art estimates for the subspace basis and has a true dominant complexity of only 5Nr operations while satisfying all desirable streaming constraints. FRAHST successfully detects subtle anomalous patterns and when compared against four other anomaly detection techniques, it is the only with a consistent F1 ≥ 80% in the Abilene datasets as well as in the ISP datasets introduced in this work.

Towards Gazetteer Integration through an Instance-based Thesauri Mapping Approach

A gazetteer is a database that stores information about a set of geographic features, classified using terms taken from a given feature type thesaurus. A geographic information system may integrate one or more gazetteers to create a consolidated information source about the data the system stores, for example [9]. However, as in a data-warehouse creation process, gazetteer integration requires aligning feature type thesauri, which is the central question we address in this paper.

Efficient Implementation of the WARM-UP Algorithm for the Construction of Length-Restricted Prefix Codes

Given an alphabet Σ = {a1, ..., an} with a corresponding list of positive weights {w1, ..., wn} and a length restriction L, the length-restricted prefix code problem is to find, a prefix code that minimizes Σni=1 wili, where li, the length of the codeword assigned to ai, cannot be greater than L, for i = 1, ..., n. In this paper, we present an efficient implementation of the WARM-UP algorithm, an approximative method for this problem. The worst-case time complexity of WARMUP is O(n log n + n log wn), where wn is the greatest weight. However, some experiments with a previous implementation of WARM-UP show that it runs in linear time for several practical cases, if the input weights are already sorted. In addition, it often produces optimal codes. The proposed implementation combines two new enhancements to reduce the space usage of WARM-UP and to improve its execution time. As a result, it is about ten times faster than the previous implementation of WARM-UP and overcomes the LRR Package Method, the faster known exact method.

Latent trees for coreference resolution

We describe a structure learning system for unrestricted coreference resolution that explores two key modeling techniques: latent coreference trees and automatic entropy-guided feature induction. The latent tree modeling makes the learning problem computationally feasible because it incorporates a meaningful hidden structure. Additionally, using an automatic feature induction method, we can efficiently build enhanced nonlinear models using linear model learning algorithms. We present empirical results that highlight the contribution of each modeling technique used in the proposed system. Empirical evaluation is performed on the multilingual unrestricted coreference CoNLL-2012 Shared Task datasets, which comprise three languages: Arabic, Chinese and English. We apply the same system to all languages, except for minor adaptations to some language-dependent features such as nested mentions and specific static pronoun lists. A previous version of this system was submitted to the CoNLL-2012 Shared Task closed track, achieving an official score of 58.69, the best among the competitors. The unique enhancement added to the current system version is the inclusion of candidate arcs linking nested mentions for the Chinese language. By including such arcs, the score increases by almost 4.5 points for that language. The current system shows a score of 60.15, which corresponds to a 3.5% error reduction, and is the best performing system for each of the three languages.

Entropy Guided Transformation Learning

This chapter details the entropy guided transformation learning algorithm [8, 23]. ETL is an effective way to overcome the transformation based learning bottleneck: the construction of good template sets. In order to better motivate and describe ETL, we first provide an overview of the TBL algorithm in Sect. 2.1. Next, in Sect. 2.2, we explain why the manual construction of template sets is a bottleneck for TBL. Then, in Sect. 2.3, we detail the entropy guided template generation strategy employed by ETL. In Sect. 2.3, we also present strategies to handle high dimensional features and to include the current classification feature in the generated templates. In Sects. 2.4–2.6 we present some variations on the basic ETL strategy. Finally, in Sect. 2.7, we discuss some related works.

Bounding the Inefficiency of Length-Restricted Prefix Codes

Abstract. We consider an alphabet Σ= {a1,\ldots,an} with corresponding symbol probabilities p1,\ldots,pn . For each prefix code associated to Σ , let li be the length of the codeword associated to ai . The average code length c is defined by c=\sumi=1n pi li . An optimal prefix code for Σ is one that minimizes c . An optimal L -restricted prefix code is a prefix code that minimizes c constrained to li ≤ L for i=1,\ldots,n . The value of the length restriction L is an integer no smaller than \lceil log n \rceil . Let A be the average length of an optimal prefix code for Σ . Also let AL be the average length of an optimal L -restricted prefix code for Σ . The average code length difference ɛ is defined by ɛ=AL -A . Let ψ be the golden ratio 1.618. In this paper we show that ɛ < 1/ψL-\lceil\log (n+\lceil\log n\rceil-L)\rceil-1 when L > \lceil log n \rceil . We also prove the sharp bound ɛ < \lceil log n \rceil -1 , when L = \lceil log n \rceil . By showing the lower bound 1/(ψL-\lceil\log n\rceil+2+\lceil\log (n/(n-L))\rceil-1) on the maximum value of ɛ , we guarantee that our bound is asymptotically tight in the range \lceil log n \rceil < L ≤ n/2 . When L\geq \lceil log n \rceil +11 , the bound guarantees that ɛ < 0.01 . From a practical point of view, this is a negligible loss of compression efficiency. Furthermore, we present an O(n) time and space 1/ψL-\lceil\log (n+\lceil\log n\rceil-L)\rceil-1 -approximative algorithm to construct L -restricted prefix codes, assuming that the given probabilities are already sorted. The results presented in this paper suggest that one can efficiently implement length restricted prefix codes, obtaining also very effective codes.

The WARM-UP Algorithm: A Lagrangian Construction of Length Restricted Huffman Codes

Given an alphabet {a,1, . . . ,an} with the corresponding list of weights [w1, . . . ,wn], and a number