Factors influencing robustness and effectiveness of conditional random fields in active learning frameworks

Active learning approaches reduce the annotation cost required by traditional supervised approaches to reach the same effectiveness by actively selecting informative instances during the learning phase. However, effectiveness and robustness of the learnt models are influenced by a number of factors. In this paper we investigate the factors that affect the effectiveness, more specifically in terms of stability and robustness, of active learning models built using conditional random fields (CRFs) for information extraction applications. Stability, defined as a small variation of performance when small variation of the training data or a small variation of the parameters occur, is a major issue for machine learning models, but even more so in the active learning framework which aims to minimise the amount of training data required. The factors we investigate are a) the choice of incremental vs. standard active learning, b) the feature set used as a representation of the text (i.e., morphological features, syntactic features, or semantic features) and c) Gaussian prior variance as one of the important CRFs parameters. Our empirical findings show that incremental learning and the Gaussian prior variance lead to more stable and robust models across iterations. Our study also demonstrates that orthographical, morphological and contextual features as a group of basic features play an important role in learning effective models across all iterations.

Pre-processing for approximate Bayesian computation in image analysis

Most of the existing algorithms for approximate Bayesian computation (ABC) assume that it is feasible to simulate pseudo-data from the model at each iteration. However, the computational cost of these simulations can be prohibitive for high dimensional data. An important example is the Potts model, which is commonly used in image analysis. Images encountered in real world applications can have millions of pixels, therefore scalability is a major concern. We apply ABC with a synthetic likelihood to the hidden Potts model with additive Gaussian noise. Using a pre-processing step, we fit a binding function to model the relationship between the model parameters and the synthetic likelihood parameters. Our numerical experiments demonstrate that the precomputed binding function dramatically improves the scalability of ABC, reducing the average runtime required for model fitting from 71 hours to only 7 minutes. We also illustrate the method by estimating the smoothing parameter for remotely sensed satellite imagery. Without precomputation, Bayesian inference is impractical for datasets of that scale.

Effects of different surfaces on the perception of prey-generated noise by the Indian false vampire bat Megaderma lyra.

The low- and high-frequency components of a rustling sound, created when prey (freshly killed frog) was jerkily pulled on dry and wet sandy floors and asbestos, were recorded and played back to individual Indian false vampire bats (Megaderma lyra). Megaderma lyra responded with flight toward the speakers and captured dead frogs, that were kept as reward. The spectral peaks were at 8.6, 7.1 and 6.8 kHz for the low-frequency components of the sounds created at the dry, asbestos and wet floors, respectively. The spectral peaks for the high-frequency sounds created on the respective floors were at 36.8,27.2 and 23.3 kHz. The sound from the dry floor was more intense than that of from the other two substrata. Prey movements that generated sonic or ultrasonic sounds were both sufficient and necessary for the bats to detect and capture prey. The number of successful prey captures was significantly greater for the dry floor sound, especially to its high-frequency components. Bat-responses were low to the wet floor and moderate to the asbestos floor sounds. The bats did not respond to the sound of unrecorded parts of the tape. Even though the bats flew toward the speakers when the prey generated sounds were played back and captured the dead frogs we cannot rule out the possibility of M. lyra using echolocation to localize prey. However, the study indicates that prey that move on dry sandy floor are more vulnerable to predation by M. lyra.

Modeling kinect sensor noise for improved 3D reconstruction and tracking

We contribute an empirically derived noise model for the Kinect sensor. We systematically measure both lateral and axial noise distributions, as a function of both distance and angle of the Kinect to an observed surface. The derived noise model can be used to filter Kinect depth maps for a variety of applications. Our second contribution applies our derived noise model to the KinectFusion system to extend filtering, volumetric fusion, and pose estimation within the pipeline. Qualitative results show our method allows reconstruction of finer details and the ability to reconstruct smaller objects and thinner surfaces. Quantitative results also show our method improves pose estimation accuracy. © 2012 IEEE.

Using extended random set to find specific patterns

With the overwhelming increase in the amount of data on the web and data bases, many text mining techniques have been proposed for mining useful patterns in text documents. Extracting closed sequential patterns using the Pattern Taxonomy Model (PTM) is one of the pruning methods to remove noisy, inconsistent, and redundant patterns. However, PTM model treats each extracted pattern as whole without considering included terms, which could affect the quality of extracted patterns. This paper propose an innovative and effective method that extends the random set to accurately weigh patterns based on their distribution in the documents and their terms distribution in patterns. Then, the proposed approach will find the specific closed sequential patterns (SCSP) based on the new calculated weight. The experimental results on Reuters Corpus Volume 1 (RCV1) data collection and TREC topics show that the proposed method significantly outperforms other state-of-the-art methods in different popular measures.

Decision making in social neurobiological systems modeled as transitions in dynamic pattern formation

Extant models of decision making in social neurobiological systems have typically explained task dynamics as characterized by transitions between two attractors. In this paper, we model a three-attractor task exemplified in a team sport context. The model showed that an attacker–defender dyadic system can be described by the angle x between a vector connecting the participants and the try line. This variable was proposed as an order parameter of the system and could be dynamically expressed by integrating a potential function. Empirical evidence has revealed that this kind of system has three stable attractors, with a potential function of the form V(x)=−k1x+k2ax2/2−bx4/4+x6/6, where k1 and k2 are two control parameters. Random fluctuations were also observed in system behavior, modeled as white noise εt, leading to the motion equation dx/dt = −dV/dx+Q0.5εt, where Q is the noise variance. The model successfully mirrored the behavioral dynamics of agents in a social neurobiological system, exemplified by interactions of players in a team sport.

Catching drink drivers in Queensland through Random Breath Testing : what does the last decade of data tell us

Background Random Breath Testing (RBT) has proven to be a cornerstone of enforcement attempts to deter (as well as apprehend) motorists from drink driving in Queensland (Australia) for decades. However, scant published research has examined the relationship between the frequency of implementing RBT activities and subsequent drink driving apprehension rates across time. Aim This study aimed to examine the prevalence of apprehending drink drivers in Queensland over a 12 year period. It was hypothesised that an increase in breath testing rates would result in a corresponding decrease in the frequency of drink driving apprehension rates over time, which would reflect general deterrent effects. Method The Queensland Police Service provided RBT data that was analysed. Results Between the 1st of January 2000 and 31st of December 2011, 35,082,386 random breath tests (both mobile and stationary) were conducted in Queensland, resulting in 248,173 individuals being apprehended for drink driving offences. A total of 342,801 offences were recorded during this period, representing an intercept rate of .96. Of these offences, 276,711 (80.72%) were recorded against males and 66,024 (19.28%) offences committed by females. The most common drink driving offence was between 0.05 and 0.08 BAC limit. The largest proportion of offences was detected on the weekends, with Saturdays (27.60%) proving to be the most common drink driving night followed by Sundays (21.41%). The prevalence of drink driving detection rates rose steadily across time, peaking in 2008 and 2009, before slightly declining. This decline was observed across all Queensland regions and any increase in annual figures was due to new offence types being developed. Discussion This paper will further outline the major findings of the study in regards to tailoring RBT operations to increase detection rates as well as improve the general deterrent effect of the initiative.

Corner detection in images under different noise levels

Corner detection has shown its great importance in many computer vision tasks. However, in real-world applications, noise in the image strongly affects the performance of corner detectors. Few corner detectors have been designed to be robust to heavy noise by now, partly because the noise could be reduced by a denoising procedure. In this paper, we present a corner detector that could find discriminative corners in images contaminated by noise of different levels, without any denoising procedure. Candidate corners (i.e., features) are firstly detected by a modified SUSAN approach, and then false corners in noise are rejected based on their local characteristics. Features in flat regions are removed based on their intensity centroid, and features on edge structures are removed using the Harris response. The detector is self-adaptive to noise since the image signal-to-noise ratio (SNR) is automatically estimated to choose an appropriate threshold for refining features. Experimental results show that our detector has better performance at locating discriminative corners in images with strong noise than other widely used corner or keypoint detectors.

An external field prior for the hidden Potts model with application to cone-beam computed tomography

In images with low contrast-to-noise ratio (CNR), the information gain from the observed pixel values can be insufficient to distinguish foreground objects. A Bayesian approach to this problem is to incorporate prior information about the objects into a statistical model. A method for representing spatial prior information as an external field in a hidden Potts model is introduced. This prior distribution over the latent pixel labels is a mixture of Gaussian fields, centred on the positions of the objects at a previous point in time. It is particularly applicable in longitudinal imaging studies, where the manual segmentation of one image can be used as a prior for automatic segmentation of subsequent images. The method is demonstrated by application to cone-beam computed tomography (CT), an imaging modality that exhibits distortions in pixel values due to X-ray scatter. The external field prior results in a substantial improvement in segmentation accuracy, reducing the mean pixel misclassification rate for an electron density phantom from 87% to 6%. The method is also applied to radiotherapy patient data, demonstrating how to derive the external field prior in a clinical context.

A modified Density-Based Scanning Algorithm with Noise for spatial travel pattern analysis from Smart Card AFC data

Smart Card Automated Fare Collection (AFC) data has been extensively exploited to understand passenger behavior, passenger segment, trip purpose and improve transit planning through spatial travel pattern analysis. The literature has been evolving from simple to more sophisticated methods such as from aggregated to individual travel pattern analysis, and from stop-to-stop to flexible stop aggregation. However, the issue of high computing complexity has limited these methods in practical applications. This paper proposes a new algorithm named Weighted Stop Density Based Scanning Algorithm with Noise (WS-DBSCAN) based on the classical Density Based Scanning Algorithm with Noise (DBSCAN) algorithm to detect and update the daily changes in travel pattern. WS-DBSCAN converts the classical quadratic computation complexity DBSCAN to a problem of sub-quadratic complexity. The numerical experiment using the real AFC data in South East Queensland, Australia shows that the algorithm costs only 0.45% in computation time compared to the classical DBSCAN, but provides the same clustering results.

Real-time assessment of GNSS observation noise with single receivers

Stochastic modelling is critical in GNSS data processing. Currently, GNSS data processing commonly relies on the empirical stochastic model which may not reflect the actual data quality or noise characteristics. This paper examines the real-time GNSS observation noise estimation methods enabling to determine the observation variance from single receiver data stream. The methods involve three steps: forming linear combination, handling the ionosphere and ambiguity bias and variance estimation. Two distinguished ways are applied to overcome the ionosphere and ambiguity biases, known as the time differenced method and polynomial prediction method respectively. The real time variance estimation methods are compared with the zero-baseline and short-baseline methods. The proposed method only requires single receiver observation, thus applicable to both differenced and un-differenced data processing modes. However, the methods may be subject to the normal ionosphere conditions and low autocorrelation GNSS receivers. Experimental results also indicate the proposed method can result on more realistic parameter precision.

High order unconditionally stable difference schemes for the Riesz space-fractional telegraph equation

In this paper, a class of unconditionally stable difference schemes based on the Pad´e approximation is presented for the Riesz space-fractional telegraph equation. Firstly, we introduce a new variable to transform the original dfferential equation to an equivalent differential equation system. Then, we apply a second order fractional central difference scheme to discretise the Riesz space-fractional operator. Finally, we use (1, 1), (2, 2) and (3, 3) Pad´e approximations to give a fully discrete difference scheme for the resulting linear system of ordinary differential equations. Matrix analysis is used to show the unconditional stability of the proposed algorithms. Two examples with known exact solutions are chosen to assess the proposed difference schemes. Numerical results demonstrate that these schemes provide accurate and efficient methods for solving a space-fractional hyperbolic equation.

Noise Filtering of Process Execution Logs based on Outliers Detection

This paper presents a technique for the automated removal of noise from process execution logs. Noise is the result of data quality issues such as logging errors and manifests itself in the form of infrequent process behavior. The proposed technique generates an abstract representation of an event log as an automaton capturing the direct follows relations between event labels. This automaton is then pruned from arcs with low relative frequency and used to remove from the log those events not fitting the automaton, which are identified as outliers. The technique has been extensively evaluated on top of various auto- mated process discovery algorithms using both artificial logs with different levels of noise, as well as a variety of real-life logs. The results show that the technique significantly improves the quality of the discovered process model along fitness, appropriateness and simplicity, without negative effects on generalization. Further, the technique scales well to large and complex logs.