Relative Performance Improvement Research Articles

Context-augmented convolutional neural networks for twitter sarcasm detection

Abstract Sarcasm detection on twitter has received increasing research in recent years. However, existing work has two limitations. First, existing work mainly uses discrete models, requiring a large number of manual features, which can be expensive to obtain. Second, most existing work focuses on feature engineering according to the tweet itself, and does not utilize contextual information regarding the target tweet. However, contextual information (e.g. a conversation or the history tweets of the target tweet author) may be available for the target tweet. To address the above two issues, we explore the neural network models for twitter sarcasm detection. Based on convolutional neural network, we propose two different context-augmented neural models for this task. Results on the dataset show that neural models can achieve better performance compared to state-of-the-art discrete models. Meanwhile, the proposed context-augmented neural models can effectively decode sarcastic clues from contextual information, and give a relative improvement in the detection performance.

A scaling law for short term load forecasting on varying levels of aggregation

This paper proposes a simple empirical scaling law that describes load forecasting accuracy at varying levels of aggregation. We show that for many forecasting methods, aggregating more customers improves the relative forecasting performance up to specific point. Beyond this point, no more improvement in relative performance can be obtained. A benchmarking procedure for applying the scaling law to different forecasting models is presented. The aggregation model is evaluated with year long consumption profiles of over 180 thousand Pacific Gas & Electric customers. A theoretical model based on a bias variance decomposition of the forecast error is used to model the Aggregation Error Curves (AECs) that are empirically explored.

Synthetic speech detection using fundamental frequency variation and spectral features

Recent works on the vulnerability of automatic speaker verification (ASV) systems confirm that malicious spoofing attacks using synthetic speech can provoke significant increase in false acceptance rate. A reliable detection of synthetic speech is key to develop countermeasure for synthetic speech based spoofing attacks. In this paper, we targeted that by focusing on three major types of artifacts related to magnitude, phase and pitch variation, which are introduced during the generation of synthetic speech. We proposed a new approach to detect synthetic speech using score-level fusion of front-end features namely, constant Q cepstral coefficients (CQCCs), all-pole group delay function (APGDF) and fundamental frequency variation (FFV). CQCC and APGDF were individually used earlier for spoofing detection task and yielded the best performance among magnitude and phase spectrum related features, respectively. The novel FFV feature introduced in this paper to extract pitch variation at frame-level, provides complementary information to CQCC and APGDF. Experimental results show that the proposed approach produces the best stand-alone spoofing detection performance using Gaussian mixture model (GMM) based classifier on ASVspoof 2015 evaluation dataset. An overall equal error rate of 0.05% with a relative performance improvement of 76.19% over the next best-reported results is obtained using the proposed method. In addition to outperforming all existing baseline features for both known and unknown attacks, the proposed feature combination yields superior performance for ASV system (GMM with universal background model/i-vector) integrated with countermeasure framework. Further, the proposed method is found to have relatively better generalization ability when either one or both of copy-synthesized data and limited spoofing data are available a priori in the training pool.

Non-intrusive load monitoring by using active and reactive power in additive Factorial Hidden Markov Models

Non-intrusive load monitoring (NILM) is the task of determining the appliances individual contributions to the aggregate power consumption by using a set of electrical parameters measured at a single metering point. NILM allows to provide detailed consumption information to the users, that induces them to modify their habits towards a wiser use of the electrical energy. This paper proposes a NILM algorithm based on the joint use of active and reactive power in the Additive Factorial Hidden Markov Models framework. In particular, in the proposed approach, the appliance model is represented by a bivariate Hidden Markov Model whose emitted symbols are the joint active-reactive power signals. The disaggregation is performed by means of an alternative formulation of the Additive Factorial Approximate Maximum a Posteriori (AFAMAP) algorithm for dealing with the bivariate HMM models. The proposed solution has been compared to the original AFAMAP algorithm based on the active power only and to the seminal approach proposed by Hart (1992), based on finite state machine appliance models and which employs both the active and reactive power. Hart’s algorithm has been improved for handling the occurrence of multiple solutions by means of a Maximum A Posteriori technique (MAP). The experiments have been conducted on the AMPds dataset in noised and denoised conditions and the performance evaluated by using the F1-Measure and the normalized disaggregation metrics. In terms of F1-Measure, the results showed that the proposed approach outperforms AFAMAP, Hart’s algorithm, and Hart’s with MAP respectively by +14.9%, +21.8%, and +2.5% in the 6 appliances denoised case study. In the 6 appliances noised case study, the relative performance improvement is +25.5%, +51.1%, and +6.7%.

Text-independent speaker identification using robust statistics estimation

It is well-known that the performance of Gaussian mixture model-based text-independent speaker identification systems deteriorates significantly with the presence of noise and spectral distortion in the training and testing utterances. In this paper, we propose a novel GMM-based speaker identification system based on two robust-statistics estimation methods: the minimum volume ellipsoid method, and the minimum covariance determinant method. Compared to the traditional maximum likelihood estimation method, the proposed methods are less sensitive to outliers in the feature-vector space caused by additive noise and spectral distortion. Moreover, in the testing phase, we propose a simple distance metric to be used for comparing the unknown testing utterance against the speakers’ models. Furthermore, we derive a more robust version of the i-vector extractor, named robust i-vector, which utilizes our proposed robust estimation methods for estimating the parameters of the base universal background model. The proposed classification system has been applied to the NIST 2000 speaker recognition evaluation and the COSINE database. It has also been compared against state-of-the-art techniques such as the GMM/UBM method, the super-vectors method, and the i-vector methods. Experimental results show that the proposed classification system provides up to 16% relative improvement in the identification performance over the i-vector methods for short utterances in the NIST 2000 database and up to 8% when the utterances of the NIST 2000 database are contaminated by different types of artificial noise for signal-to-noise ratio ranging from 0 to 20 dB. For the COSINE database, the robust i-vector estimation provides an absolute improvement of up to 8%. Finally, the real time factor of the proposed distance metric for testing is 55% higher than the RT of the regular ML scoring.

Efficient Light Management by Textured Nanoimprinted Layers for Perovskite Solar Cells

Inorganic–organic perovskites like methylammonium-lead-iodide have proven to be an effective class of materials for fabricating efficient solar cells. To improve their performance, light management techniques using textured surfaces, similar to those used in established solar cell technologies, should be considered. Here, we apply a light management foil created by UV nanoimprint lithography on the glass side of an inverted (p-i-n) perovskite solar cell with 16.3% efficiency. The obtained 1 mA cm–2 increase in the short-circuit current density translates to a relative improvement in cell performance of 5%, which results in a power conversion efficiency of 17.1%. Optical 3D simulations based on experimentally obtained parameters were used to support the experimental findings. A good match between the simulated and experimental data was obtained, validating the model. Optical simulations reveal that the main improvement in device performance is due to a reduction in total reflection and that relative improv...

A Spatial Layout and Scale Invariant Feature Representation for Indoor Scene Classification

Unlike standard object classification, where the image to be classified contains one or multiple instances of the same object, indoor scene classification is quite different since the image consists of multiple distinct objects. Further, these objects can be of varying sizes and are present across numerous spatial locations in different layouts. For automatic indoor scene categorization, large scale spatial layout deformations and scale variations are therefore two major challenges and the design of rich feature descriptors which are robust to these challenges is still an open problem. This paper introduces a new learnable feature descriptor called "spatial layout and scale invariant convolutional activations" to deal with these challenges. For this purpose, a new Convolutional Neural Network architecture is designed which incorporates a novel 'Spatially Unstructured' layer to introduce robustness against spatial layout deformations. To achieve scale invariance, we present a pyramidal image representation. For feasible training of the proposed network for images of indoor scenes, the paper proposes a new methodology which efficiently adapts a trained network model (on a large scale data) for our task with only a limited amount of available training data. Compared with existing state of the art, the proposed approach achieves a relative performance improvement of 3.2%, 3.8%, 7.0%, 11.9% and 2.1% on MIT-67, Scene-15, Sports-8, Graz-02 and NYU datasets respectively.

Does “6-Clicks” Day 1 Postoperative Mobility Score Predict Discharge Disposition After Total Hip and Knee Arthroplasties?

BackgroundThe use of inpatient rehabilitation services after total joint arthroplasty (TJA) is an important driver of episode-of-care costs. We determined the utility of a new standardized instrument collected during the immediate postoperative period, the Activity Measure for Post-Acute Care (AM-PAC) “6-Clicks” Mobility score, in predicting discharge disposition after TJA and its accuracy in estimating prolonged hospital stay, readmissions, and emergency department (ED) visits. MethodsUsing our institutional database, we retrospectively reviewed 744 patients undergoing primary total hip (40%) or knee (60%) arthroplasty for osteoarthritis during 2014. The AM-PAC Mobility score was prospectively collected by physical therapists within 24 hours of surgery. We constructed 2 multivariable logistic regression models for each study outcome: (1) a base model containing age, sex, Charlson Comorbidity Index, and procedure type and (2) the AM-PAC model including the aforementioned variables and this score. The predictive performance of these models was assessed and compared using the area under the receiver operating characteristic (AUC) curve. ResultsThe AM-PAC model provided better prediction of discharge disposition (AUC = 0.777) than the base model (AUC = 0.716; 22% relative improvement). Although the AM-PAC model performed 32% and 27% better than the base model in estimating prolonged hospital stay and ED visits, the model’s predictive performance was poor (prolonged stay: AUC = 0.639; ED visit: AUC = 0.658). The AM-PAC model also showed poor discrimination of readmissions (AUC = 0.657), and there was no relative improvement in predictive performance compared to that of the base model. ConclusionThe AM-PAC “6-Clicks” Mobility score is a valid, simple tool for predicting discharge disposition after TJA.

Unsupervised accent classification for deep data fusion of accent and language information

Automatic Dialect Identification (DID) has recently gained substantial interest in the speech processing community. Studies have shown that the variation in speech due to dialect is a factor which significantly impacts speech system performance. Dialects differ in various ways such as acoustic traits (phonetic realization of vowels and consonants, rhythmical characteristics, prosody) and content based word selection (grammar, vocabulary, phonetic distribution, lexical distribution, semantics). The traditional DID classifier is usually based on Gaussian Mixture Modeling (GMM), which is employed as baseline system. We investigate various methods of improving the DID based on acoustic and text language sub-systems to further boost the performance. For acoustic approach, we propose to use i-Vector system. For text language based dialect classification, a series of natural language processing (NLP) techniques are explored to address word selection and grammar factors, which cannot be modeled using an acoustic modeling system. These NLP techniques include: two traditional approaches, including N-Gram modeling and Latent Semantic Analysis (LSA), and a novel approach based on Term Frequency–Inverse Document Frequency (TF-IDF) and logistic regression classification. Due to the sparsity of training data, traditional text approaches do not offer superior performance. However, the proposed TF-IDF approach shows comparable performance to the i-Vector acoustic system, which when fused with the i-Vector system results in a final audio-text combined solution that is more discriminative. Compared with the GMM baseline system, the proposed audio-text DID system provides a relative improvement in dialect classification performance of +40.1% and +47.1% on the self-collected corpus (UT-Podcast) and NIST LRE-2009 data, respectively. The experiment results validate the feasibility of leveraging both acoustic and textual information in achieving improved DID performance.

RABOC: An approach to handle class imbalance in multimodal biometric authentication

Class imbalance poses serious difficulties to most standard two-class classifiers, when applied in performing classification in the context of multimodal biometric authentication. Most conventional classifiers assume equally balanced classes. They do not work well when impostor samples vastly outnumber the samples of the genuine user class. In this paper, we propose an algorithm, called RABOC, which inherits the natural capabilities of one-class classification and Real AdaBoost algorithm to handle the class imbalance problem in biometric systems. Particularly, we develop a weak classifier, which consists of one-class classifiers and is trained using data from both classes. We then exploit Real AdaBoost to combine the multiple weak classifiers in order to improve their performance without causing overfitting. Unlike the conventional Real AdaBoost, the weak classifiers in the proposed schema are learned on the same data set, but with different parameter choices. This not only generates the diversity necessary to make RABOC work, but also reduces the number of user-specified parameters. Extensive experiments were carried out on the BioSecure DS2 and XM2VTS benchmark databases, which involve data with extremely imbalanced class distribution. They demonstrate that the proposed RABOC algorithm can achieve a relative performance improvement of 28%, 24%, and 22% as compared to other state-of-the-art techniques, specifically the sum of scores, likelihood ratio based score fusion, and Support Vector Machines.

Fast Single- and Cross-Show Speaker Diarization Using Binary Key Speaker Modeling

Speaker diarization has become a key process within other speech processing systems which take advantage of single-speaker speech signals. Furthermore, finding recurrent speakers among a set of audio recordings, known as cross-show diarization, is gaining attention in the last years. Current state-of-the-art-systems provide good performance, but usually at the cost of long processing times. This limitation may make current systems not suitable for real-life applications. In this line, the speaker diarization approach based on binary key modeling provides a very fast yet accurate alternative. In this paper, we present the last improvements applied in binary key speaker diarization with the aim of further speeding up the process and improving performance. In addition, we propose a novel method for cross-show speaker diarization based on binary keys. Experimental results show the effectiveness of the proposed improvements for single-show speaker diarization, both in terms of speed and performance, obtaining a real-time factor of 0.0354xRT and a 16.8% relative improvement in performance. Furthermore, our proposed cross-show approach provides very competitive performance, just slightly worse than its single-show diarization counterpart, and exhibits a real time factor of 0.04xRT.

Performance and development plans for the Inner Detector trigger algorithms at ATLAS

A description of the design and performance of the newly re-implemented tracking algorithms for the ATLAS trigger for LHC Run 2, to commence in spring 2015, is presented. The ATLAS High Level Trigger (HLT) has been restructured to run as a more flexible single stage process, rather than the two separate Level 2 and Event Filter stages used during Run 1. To make optimal use of this new scenario, a new tracking strategy has been implemented for Run 2. This new strategy will use a FastTrackFinder algorithm to directly seed the subsequent Precision Tracking, and will result in improved track parameter resolution and significantly faster execution times than achieved during Run 1 and with better efficiency. The timings of the algorithms for electron and tau track triggers are presented. The profiling infrastructure, constructed to provide prompt feedback from the optimisation, is described, including the methods used to monitor the relative performance improvements as the code evolves. The online deployment and commissioning are also discussed.

Cross‐domain variation compensation for robust speaker verification

Recent studies have shown that when state-of-the-art probabilistic linear discriminant analysis (PLDA) speaker verification systems are developed with out-domain data, the mismatch between development data and evaluation data significantly degrades speaker verification performance. An unsupervised cross-domain variation compensation (CDVC) approach to compensate the domain mismatch is proposed. This approach is based on the assumption that the inter-domain variability is an additive factor with normal distribution in the i -vector space. The effect of the approach on the domain adaption challenge of the JHU 2013 speaker recognition workshop is tested. Applying the CDVC approach on evaluation i -vectors, the out-domain PLDA system achieves a relative performance improvement of 61.9% in equal error rate.

Searching for people using semantic soft biometric descriptions

It is not uncommon to hear a person of interest described by their height, build, and clothing (i.e. type and colour). These semantic descriptions are commonly used by people to describe others, as they are quick to communicate and easy to understand. However such queries are not easily utilised within intelligent video surveillance systems, as they are difficult to transform into a representation that can be utilised by computer vision algorithms. In this paper we propose a novel approach that transforms such a semantic query into an avatar in the form of a channel representation that is searchable within a video stream. We show how spatial, colour and prior information (person shape) can be incorporated into the channel representation to locate a target using a particle-filter like approach. We demonstrate state-of-the-art performance for locating a subject in video based on a description, achieving a relative performance improvement of 46.7% over the baseline. We also apply this approach to person re-detection, and show that the approach can be used to re-detect a person in a video steam without the use of person detection.

Effect of Particle Size and Operating Conditions on Pt3Co PEMFC Cathode Catalyst Durability

The initial performance and decay trends of polymer electrolyte membrane fuel cells (PEMFC) cathodes with Pt3Co catalysts of three mean particle sizes (4.9 nm, 8.1 nm, and 14.8 nm) with identical Pt loadings are compared. Even though the cathode based on 4.9 nm catalyst exhibited the highest initial electrochemical surface area (ECA) and mass activity, the cathode based on 8.1 nm catalyst showed better initial performance at high currents. Owing to the low mass activity of the large particles, the initial performance of the 14.8 nm Pt3Co-based electrode was the lowest. The performance decay rate of the electrodes with the smallest Pt3Co particle size was the highest and that of the largest Pt3Co particle size was lowest. Interestingly, with increasing number of decay cycles (0.6 to 1.0 V, 50 mV/s), the relative improvement in performance of the cathode based on 8.1 nm Pt3Co over the 4.9 nm Pt3Co increased, owing to better stability of the 8.1 nm catalyst. The electron microprobe analysis (EMPA) of the decayed membrane-electrode assembly (MEA) showed that the amount of Co in the membrane was lower for the larger particles, and the platinum loss into the membrane also decreased with increasing particle size. This suggests that the higher initial performance at high currents with 8.1 nm Pt3Co could be due to lower contamination of the ionomer in the electrode. Furthermore, lower loss of Co from the catalyst with increased particle size could be one of the factors contributing to the stability of ECA and mass activity of electrodes with larger cathode catalyst particles. To delineate the impact of particle size and alloy effects, these results are compared with prior work from our research group on size effects of pure platinum catalysts. The impact of PEMFC operating conditions, including upper potential, relative humidity, and temperature on the alloy catalyst decay trends, along with the EMPA analysis of the decayed MEAs, are reported.

The performance and development for the Inner Detector Trigger algorithms at ATLAS

A redesign of the tracking algorithms for the ATLAS trigger for LHC's Run 2 starting in 2015 is in progress. The ATLAS HLT software has been restructured to run as a more flexible single stage HLT, instead of two separate stages (Level 2 and Event Filter) as in Run 1. The new tracking strategy employed for Run 2 will use a Fast Track Finder (FTF) algorithm to seed subsequent Precision Tracking, and will result in improved track parameter resolution and faster execution times than achieved during Run 1. The performance of the new algorithms has been evaluated to identify those aspects where code optimisation would be most beneficial. The performance and timing of the algorithms for electron and muon reconstruction in the trigger are presented. The profiling infrastructure, constructed to provide prompt feedback from the optimisation, is described, including the methods used to monitor the relative performance improvements as the code evolves.

Thermal performance of thin loop-type vapor chamber

This experimental study examines the thermal performances of a newly devised thin loop-type vapor chamber (VC) of 0.5mm interior height at the controlled evaporator heat fluxes (q̇) and condenser thermal resistances (Rth,con) with vertical and horizontal orientations. The capillary layer over the entire heated VC endwall is constructed by the 0.05mm deep parallel/staggered square grooves with a copper woven wire mesh sintered above the staggered grooves on the evaporator endwall. The vapor–liquid flow images and their responsive overall thermal resistances (Rth) and effective spreading thermal conductivities (keff) are simultaneously detected with the objective to study the effects of q̇, Rth,con and VC orientation on the vapor–liquid two-phase flow phenomena and the associated heat transfer performances. At various test conditions, the responsive Rth and keff results to the interfacial flow structures, the comparative Rth and keff properties as well as the relative thermal performance improvements from the copper plate for the test VC are examined.

Squashing Alternatives for Software-Based Speculative Parallelization

Speculative parallelization is a runtime technique that optimistically executes sequential code in parallel, checking that no dependence violations arise. In the case of a dependence violation, all mechanisms proposed so far either switch to sequential execution, or conservatively stop and restart the offending thread and all its successors, potentially discarding work that does not depend on this particular violation. In this work we systematically explore the design space of solutions for this problem, proposing a new mechanism that reduces the number of threads that should be restarted when a data dependence violation is found. Our new solution, called exclusive squashing, keeps track of inter-thread dependencies at runtime, selectively stopping and restarting offending threads, together with all threads that have consumed data from them. We have compared this new approach with existent solutions on a real system, executing different applications with loops that are not analyzable at compile time and present as much as 10% of inter-thread dependence violations at runtime. Our experimental results show a relative performance improvement of up to 14%, together with a reduction of one-third of the numbers of squashed threads. The speculative parallelization scheme and benchmarks described in this paper are available under request.

Relative health performance in BRICS over the past 20 years: the winners and losers.

To determine whether the health performance of Brazil, the Russian Federation, India, China and South Africa--the countries known as BRICS--has kept in step with their economic development. Reductions in age- and sex-specific mortality seen in each BRICS country between 1990 and 2011 were measured. These results were compared with those of the best-performing countries in the world and the best-performing countries with similar income levels. We estimated each country's progress in reducing mortality and compared changes in that country's mortality rates against other countries with similar mean incomes to examine changes in avoidable mortality. The relative health performance of the five study countries differed markedly over the study period. Brazil demonstrated fairly even improvement in relative health performance across the different age and sex subgroups that we assessed. India's improvement was more modest and more varied across the subgroups. South Africa and the Russian Federation exhibited large declines in health performance as well as large sex-specific inequalities in health. Although China's levels of avoidable mortality decreased in absolute terms, the level of improvement appeared low in the context of China's economic growth. When evaluating a country's health performance in terms of avoidable mortality, it is useful to compare that performance against the performance of other countries. Such comparison allows any country-specific improvements to be distinguished from general global improvements.

The Elixhauser comorbidity method outperforms the Charlson index in predicting inpatient death after orthopaedic surgery.

Scores derived from comorbidities can help with risk adjustment of quality and safety data. The Charlson and Elixhauser comorbidity measures are well-known risk adjustment models, yet the optimal score for orthopaedic patients remains unclear. We determined whether there was a difference in the accuracy of the Charlson and Elixhauser comorbidity-based measures in predicting (1) in-hospital mortality after major orthopaedic surgery, (2) in-hospital adverse events, and (3) nonroutine discharge. Among an estimated 14,007,813 patients undergoing orthopaedic surgery identified in the National Hospital Discharge Survey (1990-2007), 0.80% died in the hospital. The association of each Charlson comorbidity measure and Elixhauser comorbidity measure with mortality was assessed in bivariate analysis. Two main multivariable logistic regression models were constructed, with in-hospital mortality as the dependent variable and one of the two comorbidity-based measures (and age, sex, and year of surgery) as independent variables. A base model that included only age, sex, and year of surgery also was evaluated. The discriminative ability of the models was quantified using the area under the receiver operating characteristic curve (AUC). The AUC quantifies the ability of our models to assign a high probability of mortality to patients who die. Values range from 0.50 to 1.0, with 0.50 indicating no ability to discriminate and 1.0 indicating perfect discrimination. Elixhauser comorbidity adjustment provided a better prediction of in-hospital case mortality (AUC, 0.86; 95% CI, 0.86-0.86) compared with the Charlson model (AUC, 0.83; 95% CI, 0.83-0.84) and to the base model with no comorbidities (AUC, 0.81; 95% CI, 0.81-0.81). In terms of relative improvement in predictive performance, the Elixhauser measure performed 60% better than the Charlson score in predicting mortality. The Elixhauser model discriminated inpatient morbidity better than the Charlson measure, but the discriminative ability of the model was poor and the difference in the absolute improvement in predictive power between the two models (AUC, 0.01) is of dubious clinical importance. Both comorbidity models exhibited the same degree of discrimination for estimating nonroutine discharge (AUC, 0.81; 95% CI, 0.81-0.82 for both models). Provider-specific outcomes, particularly inpatient mortality, may be evaluated differently depending on the comorbidity risk adjustment model selected. Future research assessing and comparing the performance of the Charlson and Elixhauser measures in predicting long-term outcomes would be of value. Level II, prognostic study. See the Instructions for Authors for a complete description of levels of evidence.