Abstract Emails constitute an important genre of online communication. Many of us are often faced with the daunting task of sifting through increasingly large amounts of emails on a daily basis. Keywords extracted from emails can help us combat such information overload by allowing a systematic exploration of the topics contained in emails. Existing literature on keyword extraction has not covered the email genre, and no human-annotated gold standard datasets are currently available. In this paper, we introduce a new dataset for keyword extraction from emails, and evaluate supervised and unsupervised methods for keyword extraction from emails. The results obtained with our supervised keyword extraction system (38.99% F-score) improve over the results obtained with the best performing systems participating in theSemEval2010 keyword extraction task.

Power trace of a smartphone, as time series data, carries important information of the system behavior and is useful for many applications, such as energy management [1-3], software optimization [4-6] and anomaly detection [7, 8]. However, the power trace measured from the battery terminals include the power consumption by all the hardware components and thus describes the activity of the whole system. Yet modern smartphones are multiprocessing, i.e., multiple applications can be running simultaneously in the same system. Our goal is to answer the following question: “Can we identify smartphone app by power trace?” That is, whether the power trace of a smartphone can be different by running different applications.

The electrocardiogram (ECG) is an emerging biometric modality that has seen about 13 years of development in peer-reviewed literature, and as such deserves a systematic review and discussion of the associated methods and findings. In this paper, we review most of the techniques that have been applied to the use of the electrocardiogram for biometric recognition. In particular, we categorize the methodologies based on the features and the classification schemes. Finally, a comparative analysis of the authentication performance of a few of the ECG biometric systems is presented, using our inhouse database. The comparative study includes the cases where training and testing data come from the same and different sessions (days). The authentication results show that most of the algorithms that have been proposed for ECG-based biometrics perform well when the training and testing data come from the same session. However, when training and testing data come from different sessions, a performance degradation occurs. Multiple training sessions were incorporated to diminish the loss in performance. That notwithstanding, only a few of the proposed ECG recognition algorithms appear to be able to support performance improvement due to multiple training sessions. Only three of these algorithms produced equal error rates (EERs) in the single digits, including an EER of 5.5% using a method proposed by us.

We consider the trade-off between identification and secret-key binding capacities in biometric systems. We propose the use of linear codes to achieve optimal trade-off for general source and noise models. We relate the problem to classical point to point communication problems, and two system designs based on linear codes are discussed. One design is proven to achieve optimal trade-off by using good linear codes and good decoders. The other design can be proven to achieve slightly suboptimal performance with lower decoder complexity, which is proper for many practical situations. Based on these results, we are also able to extend previous achievable rate region results to general source and noise models that have dependency within and across source and noise elements.

In real life, one often encounters situations where one needs to infer a structural relationship among data points based on an incomplete dataset. Stemmatology and phylogenetics are two classes of such problems where partial text scripts or genome sequences are available and the goal is to reconstruct the copying history of scripts or evolutionary relations among species. In this paper, we study the potential applications of minimum description length (MDL) concepts to the structural inference problem, particularly focusing on stemmatology where in addition to missing data points, the available data points have missing values. We offer new insights on how to handle these issues, especially missing values. We develop a general algorithm based on MDL insights that is simple to implement and can be used along with other existing algorithms, and propose a generic MDL encoder with minimal assumptions made about the data. In simulations, our method performs reasonably well on a simple dataset and outperforms major existing methods in a larger and much more realistic dataset. We discuss directions and ongoing efforts to further improve performance.

A novel approach for remotely sensing mechanical cardiovascular activity for use as a biometric marker is proposed. Laser Doppler Vibrometry (LDV) is employed to sense vibrations on the surface of the skin above the carotid artery related to arterial wall movements associated with the central blood pressure pulse. Carotid LDV signals are recorded using noncontact methods and the resulting unobtrusiveness is a major benefit of this technique. Several recognition methods are proposed that use the temporal and/or spectral information in the signal to assess biometric performance both on an intrasession basis, and on an intersession basis where LDV measurements were acquired from the same subjects after delays ranging from one week to six months. A waveform decomposition method that utilizes principal component analysis is used to model the signal in the time domain. Authentication testing for this approach produces an equal-error rate of 0.5% for intrasession testing. However, performance degrades substantially for intersession testing, requiring a more robust approach to modeling. Improved performance is obtained using techniques based on time-frequency decomposition, incorporating a method for extracting informative components. Biometric fusion methods including data fusion and information fusion are applied to train models using data from multiple sessions. As currently implemented, this approach yields an intersession equal-error rate of 6.3%.

We propose an information-theoretic clustering framework for density-based clustering and similarity or distance-based clustering with objective functions of clustering performance derived from stochastic complexity and minimum description length (MDL) arguments. Under this framework, the number of clusters and parameters can be determined in a principled way without prior knowledge from users. We show that similarity-based clustering can be viewed as combinatorial optimization on graphs. We propose two clustering algorithms, one of which relies on a minimum arborescence tree algorithm which returns optimal clustering under the proposed MDL objective function for similarity-based clustering. We demonstrate clustering performance on synthetic data.

Pattern recognition systems based on compressed patterns and compressed sensor measurements can be designed using low-density matrices. We examine truncation encoding where a subset of the patterns and measurements are stored perfrectly while the rest is discarded. We also examine the use of LDPC parity check matrices for compressing measurements and patterns. We show how more general ensembles of good linear codes can be used as the basis for pattern recognition system design, yielding system design strategies for more general noise models.

Pattern recognition systems may be designed to recognize an exponentially large number of objects from potentially noisy measurements. We propose a design based on storing compressed representations of binary patterns corresponding to objects of interest. Sensor measurements are similarly compressed and recognition proceeds by comparing the compressed sensor measurements to the compressed representations of the objects. Parity check matrices corresponding to low density parity check codes are used for the compression. This design yields an ensemble of systems such that the probability of error goes to zero as the length of the patterns grows