Abstract Small and highly imbalanced text datasets often produce brittle classifiers because scarce minority examples provide limited coverage of class semantics, while skewed class distributions bias decision boundaries toward the majority class. We propose Selective Few-Random-Shot Augmentation (SFRSA), a generate-then-select framework for low-resource imbalanced text classification. SFRSA first over-generates minority-class candidates through few-random-shot LLM prompting, then selects a fixed-budget subset by optimizing the trade-off between relevance to minority examples and diversity among selected samples in embedding space. To operationalize this idea, we develop and compare three selection modules: centroid-guided clustering (CGC), determinantal point process (DPP) selection, and a training-aware influence-utility DPP (IU-DPP) variant. Across three benchmark datasets under training sizes of 500, 1000, and 2000 and imbalance ratios of 4:1 and 9:1, SFRSA consistently improves minority-sensitive performance over classic text augmentation methods and standard LLM-based generation baselines, with the largest gains appearing in the most data-scarce settings. A similarity–diversity analysis links the selection objective to downstream gains, showing that SFRSA preserves minority semantics while avoiding near-duplicate generations. The method is model-agnostic and can be integrated into existing NLP pipelines to improve classification in low-resource, imbalanced settings.

In recent years, we have seen the rise of wearable sensors to obtain physiological data used to derive health metrics, such as heart rate (HR) and heart rate variability (HRV). One such recently made wearable sensor is the OmniRing. OmniRing is a PPG-based wearable device designed for continuous physiological data monitoring. In this study, we collected data from participants while they were simultaneously connected to an OmniRing and Hexoskin device. We then evaluated the accuracy of the OmniRing against Hexoskin. Additionally, we used the publicly available G-Rex dataset, which contains over 380 hours of data collected from more than 190 participants in 31 movie sessions. The dataset includes photoplethysmography (PPG) and electrodermal activity (EDA) signals, along with self-reported arousal and valence ratings. We define an algorithm that maps these arousal and valence ratings to five emotion classes based on their positions within the Russel Circumplex Model of Affect. We then evaluate the ability of different machine learning models to predict emotions given contextual information and physiological data. Our analysis shows that Gradient Boosted Trees achieved an accuracy of 37.2%, outperforming the other models used. This modest accuracy may be due to the inclusion of baseline data in the post-onset event physiological signals, potential recall bias in self-reports of arousal and valence and the limited size of the dataset.

Cyclic protein oligomers play vital roles in many biological processes and have captured the attention of protein designers. An important step of protein design is to predict the quaternary structure for a given primary sequence of a protein complex. In recent years, results from deep-learning-based methods for protein structure prediction are becoming increasingly accurate. Unfortunately, these methods are usually slow and require a large amount of memory, which can become extremely problematic for researchers, especially when facing the reverse problem: finding a sequence that folds into a required structure. In this paper, we present a new method, AF2Ring, that achieves accurate prediction with a moderate memory requirement for predicting quaternary structures of cyclic protein assemblies with repeating subunits. AF2Ring utilizes AlphaFold2 with a reduced number of chains as input and then uses the partial results to construct the complete quaternary structure. Our experimental results show that AF2Ring significantly reduced memory consumption and computation time, while achieving prediction accuracy comparable to running AlphaFold2 directly, when possible, on full-length protein sequences. This method facilitates protein design methods by providing large-scale validation of different designs and a longer sequence length allowance.

Digital phenotyping is the "moment-by-moment quantification of the individual-level human phenotype in situ using data from personal digital devices," according to Onnela and Rauch. Digital phenotyping research has historically contained many entry barriers due to high costs and complexity. However, the growing popularity of personal devices such as mobile phones has enabled researchers to collect participant data with more convenience and lower costs than ever before. This paper presents the Intelligent Phenotype Analysis Suite (IPAS), a new AI-powered tool for streamlined phenotype investigation. This tool provides researchers the ability to generate descriptions, predictions, and visualizations of their digital phenotyping data through a simple, novel chat-bot interface. IPAS combines an array of data science techniques with natural language processing capabilities of large language models to accelerate the data analysis process for researchers. IPAS extracts raw Beiwe data, an intuitive data collection platform which only requires participants to install an application on their mobile phone and permit data collection. Furthermore, we evaluate the accuracy of IPAS by using key LLM performance metrics: Precision, Recall, and F1 Score. While testing, IPAS sometimes struggled to generate code. However, in all test prompts, IPAS correctly identified the pre-written function needed to perform the requested task. Altogether, IPAS improves prior methods by enabling researchers of all levels of experience to analyze digital phenotyping data using natural language queries.

Detecting small objects in aerial images is challenging due to limited information of the object and complex backgrounds surrounding them. Most of the existing object detection methods focus on detecting individual objects and have different feature extraction methods based on object appearance. In this paper, we propose a new Graph Neural Networks (GNN) based method to refine detection results generated by object detectors. In this method, we construct a detection graph by using the predicted detection bounding boxes as nodes, while the features of a bounding box become its node features. Edges are added using distance and topological information. Then, based on the detection confidences of the bounding boxes, some nodes are labeled as bird or non-bird, where nodes with high detection confidences are labelled bird and nodes with very low detection confidences are labeled as non-bird. GNN algorithms are trained using the labeled portion of the graph to infer the classes of the unlabeled nodes. Our experimental results on detecting waterfowl in aerial images show that the new method significantly improved detection accuracy and robustness by significantly reducing false positives.

The functions of proteins are largely determined by their three-dimensional (3D) structures. Loop modeling tries to predict the conformation of a relatively short stretch of protein backbone and sidechain. It is a difficult problem due to conformational variability. Recently, AlphaFold2 has achieved outstanding results in 3-D protein structure prediction and is expected to perform well on loop modeling. In this paper, we investigate the performances of AlphaFold2 variants on popular loop modeling benchmark datasets and propose an efficient protocol of using AlphaFold2 for loop modeling, called IAFLoop. To predict the structure of a loop region, IAFLoop gives a moderately extended segment of the target loop region as input to AlphaFold2, runs a fast version of AlphaFold2 using a reduced database without ensembling, and uses RMSD based consensus scores to select the final output models. Our experimental results on benchmark datasets show that IAFLoop generated highly accurate loop models. It achieves comparable performance to the original application of AlphaFold2 in terms of RMSD error, and achieving much better results on some targets, while only using half of the time. Compared to the best previous methods, IAFLoop reduces the RMSD error by almost half on the 8-residual loop dataset, and more than 70% on the 12-residual loop dataset.

Camera traps are widely used to monitor the biodiversity and population density of animal species. Camera trap images are usually taken in bursts, and the animal counting problem for a sequence of camera trap images is also an important part of evaluating animal population density. In this paper, two new animal counting methods based on Microsoft MegaDetector V 4 have been proposed. FilterDetector uses different filters with bounding box ensemble algorithms to achieve more accurate bounding box detection. DLEDetector is an ensemble method that uses two base deep learning models to correct and enhance the detection result of MegaDetector. Our experimental results in iWildCam 2022 competition test dataset show that both methods outperformed the best method in iWildCam 2021 and the baseline method based on MegaDetector V 4 in iWildCam 2022 competition by 9.09% and 6.44%, respectively, and ranked first and third in the competition.

Domain-specific knowledge is necessary for critical analysis and decision-making in any scientific field. As a result, it is important that we have mechanisms for collecting and applying knowledge contributed by the larger scientific community. The current paradigm involves collecting knowledge in human-readable scientific papers across various scientific journals. Extracting useful information from these papers is a labor-intensive task and the growing population of papers makes it difficult to consider older works. The implementation of a knowledge network would allow for the automation of this process, but there is no existing pipeline for the reorganization of data collected through Ecological Momentary Assessment (EMA) into knowledge graphs. In this paper, we present EMA-KN, an automatically generated knowledge graph built using the AI-KG architecture. This architecture features state-of-the-art extraction by employing the DyGIE++ and StanfordCoreNLP tools. We test our pipeline using a dataset of 74 EMA-related papers and compare the output to that of AI-KG using a dataset of 74 CS-related papers to capture the success of knowledge graph construction. Further, we evaluate knowledge graph embedding using different metrics. Results show that our pipeline has a slightly lower performance rate than AI-KG, sacrificing triple quality for domain plug-ability. In the future, we seek to improve the system to match the performance of dedicated domain-specific solutions.

The rapid technological development of small Unmanned Aircraft Systems (sUAS) has led to an increase in capabilities of aerial image collection and analysis for monitoring a variety of wildlife species including waterfowl. Biologists mainly rely on conducting ocular surveys from fixed-wing aircraft or helicopters to estimate waterfowl abundance. sUAS provide an alternative that is safer, less expensive, and more flexible. Researchers have attempted to estimate waterfowl abundance from aerial imagery, but this method has proven to be too time consuming. Machine learning provides the opportunity to more efficiently estimate waterfowl abundance from aerial imagery. In this paper, we present a new integrated system of sUAS and machine learning for waterfowl population surveys. This system provides a user-friendly process for sUAS survey design, deployment, and data post-processing using deep learning methods to automatically detect and count waterfowl. To develop this system, we conducted many sUAS flights to capture a diversity of imagery and assembled six datasets of imagery taken from both fix-winged aircraft and sUAS flights. We used these datasets to develop and evaluate state-of-the-art deep learning models for waterfowl detection. Our system of using a combination of sUAS and machine learning has proved to be an efficient and accurate approach for collecting, analyzing, and estimating waterfowl abundance.

Analysis of communication effectiveness is an important task for understanding business outcomes. Prior research has shown that voice data can be used to predict communication effectiveness. However, to our knowledge, no existing studies have used both vocal and verbal cues to predict conversation outcomes in naturally occurring, dyadic business interactions. We use recorded audio calls collected from a partnering Fortune 500 firm that captures conversations between inside salespeople and business customers. Analysis of communication effectiveness is accomplished by transcribing these audio files and subsequently segmenting each conversation into customer and salesperson speaker turns to enable extraction of audio features and text embeddings for each speaker turn. All the speaker turns from the same conversation can be treated as a time series data, which can be modeled by temporal models, like LSTM or transformers. In this paper we propose that a multimodal transformer network (MTN) can capture the importance of different speaker turns and can be used to effectively predict the outcome of the call using both audio and text features. Results from the proposed model outperform current state-of-the-art results and reveal that text features offer superior outcome prediction compared to audio features.

MOTIVATION: Protein secondary structure and backbone torsion angle prediction can provide important information for predicting protein 3D structures and protein functions. Our new methods MUFold-SS, MUFold-Angle, MUFold-BetaTurn and MUFold-GammaTurn, developed based on advanced deep neural networks, achieved state-of-the-art performance for predicting secondary structures, backbone torsion angles, beta-turns and gamma-turns, respectively. An easy-to-use web service will provide the community a convenient way to use these methods for research and development. RESULTS: MUFold-SSW, a new web server, is presented. It provides predictions of protein secondary structures, torsion angles, beta-turns and gamma-turns for a given protein sequence. This server implements MUFold-SS, MUFold-Angle, MUFold-BetaTurn and MUFold-GammaTurn, which performed well for both easy targets (proteins with weak sequence similarity in PDB) and hard targets (proteins without detectable similarity in PDB) in various experimental tests, achieving results better than or comparable with those of existing methods. AVAILABILITY AND IMPLEMENTATION: MUFold-SSW is accessible at http://mufold.org/mufold-ss-angle. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Protein tertiary structure prediction is an important open challenge in bioinformatics and requires effective methods to accurately evaluate the quality of protein 3-D models generated computationally. Many quality assessment (QA) methods have been proposed over the past three decades. However, the accuracy or robustness is unsatisfactory for practical applications. In this paper, two new heuristic QA methods are proposed: MUfoldQA_S and MUfoldQA_C. The MUfoldQA_S is a quasi-single-model QA method that assesses the model quality based on the known protein structures with similar sequences. This algorithm can be directly applied to protein fragments without the necessity of building a full structural model. A BLOSUM-based heuristic is also introduced to help differentiate accurate templates from poor ones. In MUfoldQA_C, the ideas from MUfoldQA_S were combined with the consensus approach to create a multi-model QA method that could also utilize information from existing reference models and have demonstrated improved performance. Extensive experimental results of these two methods have shown significant improvement over existing methods. In addition, both methods have been blindly tested in the CASP12 world-wide competition in the protein structure prediction field and ranked as top performers in their respective categories.

Object detection in aerial images remains a challenging problem due to low image resolutions, complex backgrounds, and variations of sizes and orientations of objects in images. In recent years, several multiscale and rotated box-based deep neural networks (DNNs) have been proposed and have achieved promising results. In this article, a new method for designing loss function, called adaptive saliency biased loss (ASBL), is proposed for training DNN object detectors to achieve improved performance. ASBL can be implemented at the image level, which is called image-based ASBL, or at the anchor level, which is called anchor-based ASBL. The method computes saliency information of input images and anchors generated by DNN object detectors, weights of different training examples, and anchors differently based on their corresponding saliency measurements. It gives complex images and difficult targets more weights during training. In our experiments using two of the largest public benchmark data sets of aerial images, DOTA and NWPU VHR-10, the existing RetinaNet was trained using ASBL to generate an one-stage detector, ASBL-RetinaNet. ASBL-RetinaNet significantly outperformed the original RetinaNet by 3.61 and 12.5 mean average precision (mAP), respectively, on the two data sets. In addition, ASBL-RetinaNet outperformed ten other state-of-the-art object detection methods.

When predicting proteins' 3-D structures from their primary sequences, many existing tools use predicted residue contact information, i.e. which residues are in contact with each other. In this paper, we propose two new methods: MUFold-Contact, a new two-stage multi-branch deep neural network for predicting structure contact from protein sequences, and TPCref for refining the result of a contact prediction tool using template information. MUFold-Contact uses four independently-trained deep neural networks to predict residue-residue distances in various ranges, followed by one deep neural network to predict residue contact. TPCref is a novel approach to use protein templates to refine contact prediction generated by a particular contact prediction method. It first finds multiple template sequences based on the target sequence, and use the templates' structures and the templates' predicted contact map generated by the contact-prediction method to form a target contact-map filter, which is then used to refine the predicted contact map of the target sequence. Experimental results using recently released PDB proteins show that the performance of MUFold-Contact was comparable with those of the state-of-the-art methods, while TPCref significantly improved the contact prediction results of existing methods.

Computational protein structure prediction is an important problem in bioinformatics. In recent years, several deep learning methods have been developed for protein 3D model quality assessment (QA) and achieved promising results. In this paper, we propose a new approach of applying deep learning to the QA problem, called DeepPTQA: training one customized deep learning network for each target protein whose models are to be evaluated, i.e. the per-target training approach. For each target protein, a deep neural network is first trained using a labeled training set created based on templates of the target protein, and then applied to evaluate the quality of models of the target protein. We implemented this approach using new deep inception networks based on templates of the target protein. To predict the quality of a protein model, the deep networks use an extensive set of 1-D and 2-D features of the target protein and the target model, including predicted and real secondary structures, solvent accessibility, distance matrix, co-evolutionary information, to predict the model's true GDT-TS score, a widely used quality score in the range of [0, 1]. Experimental results using CASP targets show that DeepPTQA outperformed the best existing QA methods on the CASP Stage 2 QA task.

From SurveyMonkey to Google Forms, online surveys have become a cornerstone of modern research. However, these survey platforms lack the ability to provide advanced analysis to researchers, often requiring expensive third-party analytics software to rectify their shortcomings. We propose to solve this problem by adding data analysis capabilities onto TigerAware, an existing data collection platform. TigerAware offers a generic and customizable tool which allows researchers without technical expertise to create, manage, and deploy custom mobile surveys to participants in real-time. We seek to add data analysis functionalities to the TigerAware platform ranging from basic statistics functions to emotion recognition via deep learning. Our analysis platform uses data collected by TigerAware to give researchers real-time analytics throughout the duration of their study. Through our additions to the TigerAware platform, we present a novel all-in-one tool for researchers to facilitate effective survey creation, survey administration, data collection, and data analysis.

Prediction of protein backbone torsion angles (Psi and Phi) can provide important information for protein structure prediction and sequence alignment. Existing methods for Psi-Phi angle prediction have significant room for improvement. In this paper, a new deep residual inception network architecture, called DeepRIN, is proposed for the prediction of Psi-Phi angles. The input to DeepRIN is a feature matrix representing a composition of physico-chemical properties of amino acids, a 20-dimensional position-specific substitution matrix (PSSM) generated by PSI-BLAST, a 30-dimensional hidden Markov Model sequence profile generated by HHBlits, and predicted eight-state secondary structure features. DeepRIN is designed based on inception networks and residual networks that have performed well on image classification and text recognition. The architecture of DeepRIN enables effective encoding of local and global interatcions between amino acids in a protein sequence to achieve accruacte prediction. Extensive experimental results show that DeepRIN outperformed the best existing tools significantly. Compared to the recently released state-of-the-art tool, SPIDER3, DeepRIN reduced the Psi angle prediction error by more than 5 degrees and the Phi angle prediction error by more than 2 degrees on average. The executable tool of DeepRIN is available for download at http://dslsrv8.cs.missouri.edu/~cf797/MUFoldAngle/.

Computational protein structure prediction is a long-standing challenge in bioinformatics. In the process of predicting protein 3D structures, it is common that parts of an experimental structure are missing or parts of a predicted structure need to be remodeled. The process of predicting local protein structures of particular regions is called loop modeling. In this paper, five new loop modeling methods based on machine learning techniques, called NearLooper, ConLooper, ResLooper, HyLooper1, and HyLooper2 are proposed. NearLooper is based on the nearest neighbor technique. ConLooper applies deep convolutional neural networks to predict ${\mathrm{C}}_{{{\alpha }}}$Cα atoms distance matrix as an orientation-independent representation of protein structure. ResLooper uses residual neural networks instead of deep convolutional neural networks. HyLooper1 combines the results of NearLooper and ConLooper while HyLooper2 combines NearLooper and ResLooper. Three commonly used benchmarks for loop modeling are used to compare the performance between these methods and existing state-of-the-art methods. The experiment results show promising performance in which our best method improves existing state-of-the-art methods by 28 and 54 percent of average RMSD on two datasets while being comparable on the other one.

Patient self-management of diabetes is crucial to reducing its chronic progression and serious future health complications. Due to the physical and emotional complexities of diabetes, adherence to self-care activities can quickly decline because of disheartening progress. While there exists technology that caters to diabetic self-management, the current state of the art mostly assists patients through mobile applications. For the elderly, these mobile apps are marginally effective and even frustrating to use. Therefore, this project proposes Healthy Coping with Diabetes, a Google Home assistant application that acts as an innovative intervention strategy to assist elderly patients with self-management of type 2 diabetes. This application framework combines the voice interface of Google Home for hosting the conversational agent and a web interface for data visualization in order to reduce the burden of monitoring diabetic consequences for the user. We perform a feature-based comparison of our application against currently available mobile apps to determine the relative fulfillment of official American Association of Diabetes Educators self-management components. Finally, we calculate standard usability metrics to receive feedback and further improve the application capabilities.

As Convolutional Neural Networks continue to produce state of the art results, more types of data are being used to see the results that would be produced. Using the heart rate data that was collected using sensors from various subjects who consumed alcohol, we converted it from the 1D waveform into a set of spectrograms. The spectrograms were fed into two pretrained CNNs, CaffeNet and AlexNet, to determine whether or not the given spectrogram was an instance of alcohol consumption. Using 80 training images (40 positive, 40 negative) and 20 test images (10 positive, 10 negative), we achieved a test accuracy, after adjusting learning rate, number of iterations, and gradient descent algorithm, as well as the time window and coloration of the spectrograms, of 72% (n=20, 5 trials), which are promising results for non-audio waveforms.

Fish classification and detection in the wild remains an unsolved problem due to the variation of pose, uncontrolled environmental conditions and the limited amount of training data. We propose a deep neural fish classification system to automatically label fish using a camera without interference from human. In the system, classification is done in two independent branches. One branch, where the fish are detected, aligned, and fed to classifiers, aims to handle the variation of pose and scale of fish and extract discriminative features to distinguish fish. The other branch makes use of context information in the scenario to infer the type of fish, under the assumption that the types of captured fish are correlated with the environmental conditions around such as boat, fish gears, daylight and weather. The preliminary classification prediction made in the two branches is then reweighted and averaged as the final prediction. In the world-wide competition of The Nature Conservancy Fisheries Monitoring" hosted by Kaggle our model achieved top 0.7% rank on the final leaderboard among solutions submitted by 2 293 different competitive teams.

A protein secondary structure defines the local conformation of the protein's polypeptide backbone, which provides important information for protein 3D structure prediction and protein functions. In this study, a new deep neural network, the deep neighbor residual network (DeepNRN), is proposed for protein secondary structure predictions. The network takes three types of inputs, namely protein sequence features, profile features generated by PSI-BLAST, and profile features generated by HHBlits, and predicts the protein secondary structure in either one of eight states (Q8) or one of three states (Q3). The basic building block of the network, the neighbor residual unit, is designed with two types of short-cut connections that are more general and expressive than residual units in existing residual deep neural networks, yet can still be computed efficiently. In addition, the prediction result of DeepNRN can be refined by a Struct2Struct network to make the result more protein-like. Extensive experimental results on multiple widely used benchmark data sets show that the new DeepNRN-based method outperformed existing methods and obtained the best results across multiple data sets.

Biology and medicine have a long-standing interest in computational structure prediction and modeling of proteins. There are often missing regions or regions that need to be remodeled in protein structures. The process of predicting particular missing regions in a protein structure is called loop modeling. In this paper, we propose a generative adversarial network (GAN) in deep learning for loop modeling using the idea of image inpainting. The generative network is to capture the context of the loop region and predict the missing area. The adversarial network is to make the prediction look real and provide gradients to the generative network. The proposed network was evaluated on a common benchmark for loop modeling. Experiments show that our method can successfully predict the loop region and has achieved better performance than the state-of-the-art tools. To our knowledge, this work represents the first attempt of using GAN for any bioinformatics studies.

Computational protein structure prediction is very important for many applications in bioinformatics. In the process of predicting protein structures, it is essential to accurately assess the quality of generated models. Although many single-model quality assessment (QA) methods have been developed, their accuracy is not good enough for most real applications. In this paper, a new approach, called DeepCon-QA, based on deep convolutional neural networks and continuous distributed representation of protein sequences is proposed for single-model QA problem. By combining novel features of protein vector representation with existing features of profile information, together with the power of deep networks, our method shows promising results in solving the problem of protein quality assessment. In experiments on selected targets from the Critical Assessment of Structure Prediction (CASP) competition, DeepCon-QA obtained similar performance with the best state-of-the-art QA method MQAPmulti while outperforming other QA method.

Computational protein structure prediction is one of the most important problems in bioinformatics. In the process of protein three-dimensional structure prediction, assessing the quality of generated models accurately is crucial. In recent years, deep neural networks have obtained good results for various structure prediction related problems. In this paper, two new deep neural network structures are developed, trained and tested for single-model quality assessment. These two networks are based on a comprehensive set of model structure features, such as predicted and real secondary structure and solvent accessibility, distance matrix and φ, ψ angle. For each model, using these features as inputs, trained deep neural networks are able to predict a quality score in the range of [0, 1]. Experimental results show that the deep neural networks obtained excellent results, comparable with existing state of art single-model QA methods on CASP 12 dataset.

Mobile ambulatory assessment systems are being actively developed for various psychological studies. However, to the best of our knowledge, there are very few being used for the detection of alcohol usage. This paper presents a new automatic data analysis pipeline, called ADA (Automatic Detection of Alcohol usage), to assess alcohol use or cravings and emotional dysregulation associated with underlying psychological disorders. ADA works closely with mAAS, a smartphone-based mobile ambulatory assessment system for psychology research, currently being used by approximately 30 subjects in the natural environment. mAAS provides real-time sensor and survey data monitoring and collection tools for real-life subject behavioral and psychological data, as well as physiological data. ADA builds models with the collected data to classify alcohol use or cravings and predict emotional dysregulation. Experiments on real patients and data analysis results are reported. Results show a significant difference between various factors related to drinking, such as emotional data and physiological data.

Mobile computing and big data analytics have great potential for improving efficiency, productivity, and knowledge discovery in conservation tasks. This paper presents some recent development of mobile computing and data analysis systems for the Missouri Department of Conservation (MDC), including a mobile data collection system, an improved bear tracking website, and new analytics results obtained from real Missouri bear and deer GPS trajectories. The mobile data collection system using Android tablets is developed for surveying the usage and status of conservation areas. The new bear tracking website is developed to be mobile friendly and can dynamically present information of tracked bears. Lastly, for analyzing real GPS trajectories obtained from Missouri bears and deer, a density-based spatial and temporal clustering method is developed for identifying stay regions in trajectories of low sampling rate GPS points. Using real world data, interesting movement patterns of a large number of bears and white-tailed deer have been obtained, therefore advancing a step in achieving a better understanding of the behaviors of various types of animals in their natural environments.

Although deep learning has achieved outstanding performances on several difficult machine learning applications, there are multiple issues that make its application on new problems difficult: speed of training, local minima, and manual selection of hyper-parameters. To overcome these problems, this paper proposes a new evolutionary method, EvoAE, to train auto encoders for deep learning networks. By evolving a population of auto encoders, EvoAE learns multiple features in each auto encoder in the form of hidden nodes, evaluates the auto encoders based on their reconstruction quality, and generates new auto encoders using crossover and mutation with chromosomes made up of hidden nodes and associated connections and weights. EvoAE optimizes network weights and structures of auto encoders simultaneously and employs a mini-batch variant, called Evo-batch, to speed up auto encoder search on large datasets. Furthermore, EvoAE supports different training methods in data partitioning and selection, requires little manual intervention, and reduces overall training time drastically over traditional methods on large datasets.

BACKGROUND: As the next-generation sequencing (NGS) technologies producing hundreds of millions of reads every day, a tremendous computational challenge is to map NGS reads to a given reference genome efficiently. However, existing methods of all-mappers, which aim at finding all mapping locations of each read, are very time consuming. The majority of existing all-mappers consist of 2 main parts, filtration and verification. This work significantly reduces verification time, which is the dominant part of the running time. RESULTS: An efficient all-mapper, BitMapper, is developed based on a new vectorized bit-vector algorithm, which simultaneously calculates the edit distance of one read to multiple locations in a given reference genome. Experimental results on both simulated and real data sets show that BitMapper is from several times to an order of magnitude faster than the current state-of-the-art all-mappers, while achieving higher sensitivity, i.e., better quality solutions. CONCLUSIONS: We present BitMapper, which is designed to return all mapping locations of raw reads containing indels as well as mismatches. BitMapper is implemented in C under a GPL license. Binaries are freely available at http://home.ustc.edu.cn/%7Echhy.

Protein structure prediction, i.e., computationally predicting the three-dimensional structure of a protein from its primary sequence, is one of the most important and challenging problems in bioinformatics. Model refinement is a key step in the prediction process, where improved structures are constructed based on a pool of initially generated models. Since the refinement category was added to the biennial Critical Assessment of Structure Prediction (CASP) in 2008, CASP results show that it is a challenge for existing model refinement methods to improve model quality consistently. This paper presents three evolutionary algorithms for protein model refinement, in which multidimensional scaling(MDS), the MODELLER software, and a hybrid of both are used as crossover operators, respectively. The MDS-based method takes a purely geometrical approach and generates a child model by combining the contact maps of multiple parents. The MODELLER-based method takes a statistical and energy minimization approach, and uses the remodeling module in MODELLER program to generate new models from multiple parents. The hybrid method first generates models using the MDS-based method and then run them through the MODELLER-based method, aiming at combining the strength of both. Promising results have been obtained in experiments using CASP datasets. The MDS-based method improved the best of a pool of predicted models in terms of the global distance test score (GDT-TS) in 9 out of 16test targets.

Computational protein structure prediction is very important for many applications in bioinformatics. In the process of predicting protein structures, it is essential to accurately assess the quality of generated models. Although many single-model quality assessment (QA) methods have been developed, their accuracy is not high enough for most real applications. In this paper, a new approach based on C-α atoms distance matrix and machine learning methods is proposed for single-model QA and the identification of native-like models. Different from existing energy/scoring functions and consensus approaches, this new approach is purely geometry based. Furthermore, a novel algorithm based on deep learning techniques, called DL-Pro, is proposed. For a protein model, DL-Pro uses its distance matrix that contains pairwise distances between two residues' C-α atoms in the model, which sometimes is also called contact map, as an orientation-independent representation. From training examples of distance matrices corresponding to good and bad models, DL-Pro learns a stacked autoencoder network as a classifier. In experiments on selected targets from the Critical Assessment of Structure Prediction (CASP) competition, DL-Pro obtained promising results, outperforming state-of-the-art energy/scoring functions, including OPUS-CA, DOPE, DFIRE, and RW.

Deep learning has been shown to achieve outstanding performance in a number of challenging real-world applications. However, most of the existing works assume a fixed set of labeled data, which is not necessarily true in real-world applications. Getting labeled data is usually expensive and time consuming. Active labelling in deep learning aims at achieving the best learning result with a limited labeled data set, i.e., choosing the most appropriate unlabeled data to get labeled. This paper presents a new active labeling method, AL-DL, for cost-effective selection of data to be labeled. AL-DL uses one of three metrics for data selection: least confidence, margin sampling, and entropy. The method is applied to deep learning networks based on stacked restricted Boltzmann machines, as well as stacked autoencoders. In experiments on the MNIST benchmark dataset, the method outperforms random labeling consistently by a significant margin.

The multiple longest common subsequence (MLCS) problem, related to the identification of sequence similarity, is an important problem in many fields. As an NP-hard problem, its exact algorithms have difficulty in handling large-scale data and time- and space-efficient algorithms are required in real-world applications. To deal with time constraints, anytime algorithms have been proposed to generate good solutions with a reasonable time. However, there exists little work on space-efficient MLCS algorithms. In this paper, we formulate the MLCS problem into a graph search problem and present two space-efficient anytime MLCS algorithms, SA-MLCS and SLA-MLCS. SA-MLCS uses an iterative beam widening search strategy to reduce space usage during the iterative process of finding better solutions. Based on SA-MLCS, SLA-MLCS, a space-bounded algorithm, is developed to avoid space usage from exceeding available memory. SLA-MLCS uses a replacing strategy when SA-MLCS reaches a given space bound. Experimental results show SA-MLCS and SLA-MLCS use an order of magnitude less space and time than the state-of-the-art approximate algorithm MLCS-APP while finding better solutions. Compared to the state-of-the-art anytime algorithm Pro-MLCS, SA-MLCS and SLA-MLCS can solve an order of magnitude larger size instances. Furthermore, SLA-MLCS can find much better solutions than SA-MLCS on large size instances.

Current methods in clinical psychology primarily rely on questionnaires and interviews with examiners. This paper presents preliminary work towards a smartphone-based wireless body area sensing system that will be used to improve current methods and provide real-time interventions if necessary. This system consists of several wearable sensors for measuring physiological data, a smartphone, and a web server. The smartphone is the centerpiece, responsible for collecting sensor data, interacting with the user, performing real-time computation, and communicating with the web server. The system collects physiological data, self-reported emotional and behavioral state, and other user-context data such as GPS location or ambient audio recording.

In this paper, SDNAN, a first attempt to implement software-defined networking (SDN) over a wireless ad hoc network of smartphones, is presented. Its modular ad hoc network management structure can be easily modified and extended. Its abstractions and interfaces allow components to communicate without knowing how other components work. Third-party applications can use the interfaces to access the ad hoc network, significantly reducing development time and program complexity. A prototype system has been implemented on Android smartphones over Wi-Fi and achieved good preliminary results.

This paper presents Wake IQ, a smartphone program to reduce sleep inertia. Wake IQ aims at waking its user with proper timing so as to reduce or eliminate sleep inertia and increase overall alertness, while maintaining a convenience conducive to everyday use. It is constructed based on machine learning techniques, phone accelerometer data sampled during slumber, and user alertness test scores. Tests of the system have demonstrated a gradually increasing average response speed to reaction time tests over a period of a number of weeks, indicative of a trend towards reduced sleep inertia arising from use of the smartphone application.

The problem of determining the position of a sound source using mobile devices such as smart phones is considered in this paper. A novel calibration stage is developed to cancel out the time offsets of the time differences of arrival measurements. Two localization algorithms, one closed-form solution and the other iterative approach, are implemented to examine the performance of the proposed sound source localization system. Real-world experiments were conducted for 2-D and 3-D in-door localization. The experimental results validate the proposed localization technique which has acceptable localization bias and mean-square error.

The multiple longest common subsequence (MLCS) problem, which is related to the measurement of sequence similarity, is one of the fundamental problems in many fields. As an NP-hard problem, finding a good approximate solution within a reasonable time is important for solving large-size problems in practice. In this paper, we present a new progressive algorithm, Pro-MLCS, based on the dominant point approach. Pro-MLCS can find an approximate solution quickly and then progressively generate better solutions until obtaining the optimal one. Pro-MLCS employs three new techniques: 1) a new heuristic function for prioritizing candidate points; 2) a novel d-index-tree data structure for efficient computation of dominant points; and 3) a new pruning method using an upper bound function and approximate solutions. Experimental results show that Pro-MLCS can obtain the first approximate solution almost instantly and needs only a very small fraction, e.g., 3 percent, of the entire running time to get the optimal solution. Compared to existing state-of-the-art algorithms, Pro-MLCS can find better solutions in much shorter time, one to two orders of magnitude faster. In addition, two parallel versions of Pro-MLCS are developed: DPro-MLCS for distributed memory architecture and DSDPro-MLCS for hierarchical distributed shared memory architecture. Both parallel algorithms can efficiently utilize parallel computing resources and achieve nearly linear speedups. They also have a desirable progressiveness property-finding better solutions in shorter time when given more hardware resources.

This paper presents Nest, a novel system using wirelessly connected smartphones to localize remote targets based on sound and image inputs. The system has four major components: image-based localization, acoustics-based localization, wireless ad-hoc networking, and middle-ware services for time synchronization and secure communication. Single-image, two-image, and TDOA-acoustics based methods have been developed and a prototype system has been implemented on Google Nexus One smartphones running Android. Experimental results show that the localization accuracies of the single-image-based and two-image-based method are around 94.6% and up to 92.1%, respectively. The localization errors of the acoustics-based method are within 80 centimeters.

With the emergence of cloud computing, users are increasingly using storage services of cloud providers to satisfy their storage needs. However, these cloud providers are not 100% reliable, resulting in periodic service outages. End users require a solution that minimizes the impact of cloud service outages and optimizes the chances of successfully retrieving stored objects in cases of provider failures. In this paper, we propose to build a federated cloud agent, enabling end users to simultaneously deploy storage objects to multiple cloud service providers, based on their unit storage costs and failure probabilities. We study two service provider failure models: the first case assumes that at most one service provider fails at any given time, and the second case assumes that cloud service providers may fail independently with certain probabilities. We cast both scenarios as constrained optimization problems to minimize the overall storage cost while enhancing the chances of recovering all the storage objects in case of service outages. In the experiments, we evaluated the proposed optimal storage schemes against a fixed server deployment strategy and find our schemes can achieve good recovery ratios, while incurring minimal storage costs.

Smartphone's increasing computing power and built-in sensors such as digital camera and GPS have created a new and open platform for developing compelling mobile multimedia tools and systems. In this paper, a novel, accurate and computationally efficient video-based remote target positioning and tracking system on Android smart phones is proposed and presented, with the assumption that the physical size of the remote target is known. Smartphone's user-friendly interface is exploited in our system to facilitate accurate and efficient video object tracking. With the help of some initial user input, our algorithm generates tight bounding boxes around the moving targets across video frames with high accuracy and low complexity. Unlike most existing video tracking systems that are limited to tracking on the image plane, we re-project the accurately detected object locations on the image plane to real world positions to estimate the remote target's moving trajectory. In addition, the Extended Kalman Filtering model is applied to provide a smooth remote target trajectory and velocity estimation. Our experimental results show that the proposed mobile multimedia system works accurately and efficiently for tracking remote moving targets.

Melanoma skin cancer accounts for less than 5% of skin cancer cases but causes the most deaths due to skin cancer. Convenient automated diagnosis of skin lesions and melanoma recognition can greatly improve early detection of melanomas. This paper presents a prototype of an image-based automated melanoma recognition system on Android smart phones. The system consists of three major components: image segmentation, feature calculation, and classification. It is designed to run on a mobile device with a camera, such as a smart phone or a tablet PC. A skin lesion image is converted to a monochrome image for outline contour detection. Color and shape features of the lesion are extracted and used as input to a kNN classifier. Initial experimental result shows that the system is efficient and works well on well-lighted test images, achieving an average accuracy of 66.7%, with average malignant class recall/sensitivity of 60.7% and specificity of 80.5%.

In protein tertiary structure prediction, a crucial step is to select near-native structures from a large number of predicted structural models. Over the years, many methods have been proposed for the protein structure selection problem. Despite significant advances, the discerning power of current approaches is still unsatisfactory. In this paper, we propose a new algorithm, CC-Select, that combines consensus with clustering techniques. Given a set of predicted models, CC-Select first calculates a consensus score for each structure based on its average pair wise structural similarity to other models. Then, similar structures are grouped into clusters using multidimensional scaling and clustering algorithms. In each cluster, the one with the highest consensus score is selected as a candidate model. Using extensive benchmark sets of a large collection of predicted models, we compare CC-Select with existing state-of-the-art quality assessment methods and show significant improvement.

We present PositionIt, a novel, low-cost, image-based remote target localization system using commodity smartphones. By leveraging smartphones' built-in sensors such as camera, digital compass, GPS, etc., the system can estimate the distance/position of remote targets. The system also takes advantage of the user-friendly interface of the smartphones to facilitate low complexity implementation (e.g., to locate target of interest and restrict the search area for matching).The feasibility of both single-image and two-image based systems are demonstrated.

In protein tertiary structure prediction, a crucial step is to select near-native structures from a large number of predicted structural models. Over the years, extensive research has been conducted for the protein structure selection problem with most approaches focusing on developing more accurate energy or scoring functions. Despite significant advances in this area, the discerning power of current approaches is still unsatisfactory. In this paper, we propose a novel consensus-based algorithm for the selection of predicted protein structures. Given a set of predicted models, our method first removes redundant structures to derive a subset of reference models. Then, a structure is ranked based on its average pairwise similarity to the reference models. Using the CASP8 data set containing a large collection of predicted models for 122 targets, we compared our method with the best CASP8 quality assessment (QA) servers, which are all consensus based, and showed that our QA scores correlate better with the GDT-TSs than those of the CASP8 QA servers. We also compared our method with the state-of-the-art scoring functions and showed its improved performance for near-native model selection. The GDT-TSs of the top models picked by our method are on average more than 8 percent better than the ones selected by the best performing scoring function.

Finding the longest common subsequence (LCS) of multiple strings is an NP-hard problem, with many applications in the areas of bioinformatics and computational genomics. Although significant efforts have been made to address the problem and its special cases, the increasing complexity and size of biological data require more efficient methods applicable to an arbitrary number of strings. In this paper, we present a new algorithm for the general case of multiple LCS (or MLCS) problem, i.e., finding an LCS of any number of strings, and its parallel realization. The algorithm is based on the dominant point approach and employs a fast divide-and-conquer technique to compute the dominant points. When applied to a case of three strings, our algorithm demonstrates the same performance as the fastest existing MLCS algorithm designed for that specific case. When applied to more than three strings, our algorithm is significantly faster than the best existing sequential methods, reaching up to 2-3 orders of magnitude faster speed on large-size problems. Finally, we present an efficient parallel implementation of the algorithm. Evaluating the parallel algorithm on a benchmark set of both random and biological sequences reveals a near-linear speedup with respect to the sequential algorithm.

Finding the longest common subsequence (LCS) of multiple strings is an NP-hard problem, with many applications in the areas of bioinformatics and computational genomics. Although significant efforts have been made to address the problem and its special cases, the increasing complexity and size of biological data require more efficient methods applicable to an arbitrary number of strings. In this paper, a novel search algorithm, MLCS-A*, is presented for the general case of multiple LCS (or MLCS) problems. MLCS-A* is a variant of the A* algorithm. It maximizes a new heuristic estimate of the LCS in each search step so that the longest common subsequence can be found. As a natural extension of MLCS-A*, a fast algorithm, MLCS-APP, is also proposed to deal with large volume of biological data for which finding a LCS within reasonable time is impossible. The benchmark test shows that MLCS-APP is able to extract common subsequences close to the optimal ones and that MLCS-APP significantly outperforms existing heuristic approaches. When applied to 8 protein domain families, MLCS-APP produced more accurate results than existing multiple sequence alignment methods.

The VoIDP algorithm is the first optimal algorithm for efficiently selecting the subset of observations in chain graphical models. The original VoIDP algorithm has a mistake in the process of recovering the optimal selections, and fails to produce correct outputs. In this paper, we present an improved version of the algorithm; which fixes the mistakes and verifies the solutions in experiments. Further more, we discuss some recent works in the area of subset selection problems, and present a simplified solution for computing the maximum expected total reward for a sub chain under certain circumstances.

In protein tertiary structure prediction, a crucial step is to select near-native structures from a large number of predicted structural models. Over the years, extensive research has been conducted for the protein structure selection problem with most approaches focusing on developing more accurate energy or scoring functions. Despite significant advances in this area, the discerning power of current approaches is still unsatisfactory. In this paper, we propose a novel consensus-based method that outperforms state-of-the-art scoring functions. Given a set of predicted models, the method first removes redundant models to derive a subset of reference models. Then a model is ranked based on its average pairwise similarity score to the set of reference models. Using the CASP8 data set containing a large collections of predicted models for 122 targets, we compared our method with an extensive collection of state-of-the-art scoring functions and showed its superior performance. The top models selected by our method are on average more than 8% better than the ones selected by the best competing method.

In recent years, we have witnessed a gold rush of location based services enabled by location awareness and persistent Internet connections of modern smartphones. However, the poor performance of indoor localization remains an obstacle to providing these services effectively inside buildings. Obtaining a comprehensive database of points of interest and high quality indoor maps is another obstacle because extensive human involvement is usually required for such tasks. In this paper, we propose a novel system SMART (Simultaneous Map Acquisition and Repeated Tracking) to address both problems. By tracking a subject based on sensor inputs and radio signals and detecting environmental fingerprints, SMART achieves simultaneous indoor localization and environment map construction using smartphones based on radio signals from surrounding WiFi access points and their own measured motion dynamics. Simulation results show that SMART outperforms a dead reckoning method by approximately 9 times in localization accuracy and constructs environment maps of 89% accuracy on average. SMART is robust against sensing errors and can automatically adapt to environment changes.

Frequently appeared sensor faults greatly reduce the usability and reliability of sensor networks. Distributed automatous faulty sensor detection is critical for self-managed and sustainable sensor networks. A number of detection methods have been proposed for specific fault types. In this paper, we propose a general approach for detecting arbitrary types of faults. The approach includes a new general measurement mutual divergence for evaluating detecting results in the absence of the ground truth, and a distributed collective detection method that produces probabilistic decision results. For comparison purposes, we also introduce two detection methods in both distributed and centralized manners. We show that mutual divergence correctly measures the average fault amount and the distributed collective method consistently outperforms the other methods with up to 50% higher detection accuracy.

The world we live in is becoming smarter. Miniature sensors and actuators integrated with processors are being embedded in our environments to monitor and control the physical processes. These sensor and actuator systems are evolving towards a physically heterogeneous and distributed but functionally coherent environment consisting of both wired and wireless communication components. Battery-powered networked devices are increasingly more common. Mobile ad hoc and wireless sensor and actuator networks are receiving considerable research attention. Applications of intelligent networked sensor and actuator systems are tremendous and have the potential to dwarf the 20th century computing revolution. They include intelligent traffic control and safety, advanced automotive systems, process control, energy conservation, environmental control, advanced medical and health care devices and systems, critical infrastructure control, communications systems, distributed robotics, and security surveillance and protection. For example, transportation systems could benefit considerably from better embedded intelligence in automobiles and in transportation infrastructures, which could improve safety and efficiency. Embedded sensor and actuator systems are not new. Yet most existing systems are special purpose, closed systems with tightly coupled components. Networking and managing a large number of embedded sensors and actuators poses significant technical challenges. For example, in a network environment, the behaviors of sensor and actuator systems are much more unpredictable, and it is impossible to test software under all possible conditions. The technical challenges are not only efficiency, but also robustness, scalability, and predictability. These unique characteristics in the design of distributed systems of sensors and actuators have motivated a growing interest and a considerable amount of research activities in this emerging area. This special issue includes a collection of 13 outstanding research papers which cover diverse topics on the design, analysis, experimental evaluation, and applications of distributed systems of sensors and actuators, particularly the ones connected through wireless networks. This special issue begins with three survey papers on wireless sensor networks (WSN): applications in intelligent transportation systems, radio power management, and mobility-aware MAC protocols. In “Wireless sensor networks in intelligent transportation systems,” Tubaishat et al. reviewed existing sensor network technologies and applications for intelligent transportation systems and proposed new methods for using WSN in traffic modeling and estimation and traffic control that outperform existing methods. In “Mobility handling in MAC for wireless ad hoc networks,” Raja and Su reviewed the current state of the arts in handling mobility in sensor networks, presented simulation results to analyze existing mobility-aware MAC protocols, and discussed open issues in dealing with mobility in sensor networks. In “Towards unified radio power management for wireless sensor networks,” Xing et al. reviewed existing transmission power control and sleep scheduling approaches to reducing radio power consumption in the transmission state and the idle state and proposed a Unified Radio Power Management framework for the design and implementation of holistic radio power management solutions in WSN. The next three papers address architecture, middleware, and data storage issues in WSN. In “Lifetime-maximized cluster association in two-tiered wireless sensor networks,” Song et al. studied the two-tiered WSN architecture and proposed an optimal cluster association algorithm to maximize the overall network lifetime. In “Embedding standard distributed object-oriented middlewares in wireless sensor networks,” Moya et al. analyzed the design space available to middleware developers in the context of WSN and presented a development framework based on standard distributed objects and a set of specific services designed to support highly dynamic and scalable WSN applications. In “Improving data centric storage with diffuse caching in wireless sensor networks,” Ma et al. studied data-centric storage problems in a sensor network and proposed new diffuse caching techniques to alleviate the hot-spot problem and reduce the message overheads. The following three papers present new methods for data aggregation and image delivery in sensor networks. In “Isolines: Efficient spatio-temporal data aggregation in sensor networks,” Solis and Obraczka introduced an energy-efficient data collection technique that takes advantage of spatial and temporal data correlation for monitoring continuously changing conditions. In “Reward oriented packet filtering algorithm for wireless sensor networks,” Shu et al. studied packet filtering for reducing energy consumption in sensor networks and proposed a two-tier buffer model and two packet filtering algorithms for real time and non-real time transmissions, respectively. In “An adaptive approach for image encryption and secure transmission over multirate wireless sensor networks,” Wang et al. proposed a new adaptive secure image delivery approach to achieve effective and efficient image encryption based on an optimal transmission rate in terms of energy efficiency in WSN. The final four papers propose new methods for routing and topology control in wireless networks. In “Energy-efficient geographic multicast routing for error-prone wireless sensor networks,” Sanchez and Ruiz proposed a new protocol, Localized Energy-Efficient Multicast Algorithm, to address the error-prone nature of wireless links when designing routing protocols for WSN. In “Receiver-oriented load-balancing and reliable routing in wireless sensor networks,” Chen et al. proposed a novel receiver-oriented load-balancing and reliable routing protocol that achieves relatively longer network lifetime and higher reliability than other existing schemes. In “QuESt: A QoS-based energy efficient sensor routing protocol,” Saxena et al. used multi-objective genetic algorithm to design a QoS-based energy-efficient sensor routing protocol that determines application-specific, near-optimal sensory-routes demand by optimizing multiple QoS parameters and energy consumption. In “A novel distributed MCDS approximation algorithm for wireless sensor networks,” Xie et al. proposed a novel distributed approximation algorithm to construct the minimal connected dominating set for WSN. This algorithm generates smaller connected dominating sets while requiring less communication overhead compared with the existing distributed approximation algorithms. We thank all the authors who submitted their papers to this special issue. Owing to the limitation of space, we can only include limited number of papers in the special issue. We are grateful to all the reviewers for their time and efforts in carefully reviewing all the papers and providing valuable review comments. We also thank the Editor-in-Chief, Mohsen Guizani, for his continuous support for this special issue, and all the publication staff for their support during the publishing process.

Abstract Wireless sensor networks (WSNs) offer the potential to significantly improve the efficiency of existing transportation systems. Currently, collecting traffic data for traffic planning and management is achieved mostly through wired sensors. The equipment and maintenance cost and time‐consuming installations of existing sensing systems prevent large‐scale deployment of real‐time traffic monitoring and control. Small wireless sensors with integrated sensing, computing, and wireless communication capabilities offer tremendous advantages in low cost and easy installation. In this paper, we first survey existing WSN technologies for intelligent transportation systems (ITSs), including sensor technologies, energy‐efficient networking protocols, and applications of sensor networks for parking lot monitoring, traffic monitoring, and traffic control. Then, we present new methods on applying WSNs in traffic modeling and estimation and traffic control, and show their improved performance over existing solutions. Copyright © 2008 John Wiley & Sons, Ltd.

Real-time traffic light controllers (TLCs) play a major role in optimizing traffic flow. We uses wireless sensor network (WSN) to decrease vehicles' average trip waiting time (ATWT) on the road. We studied the performance of using one sensor and two sensors and designed corresponding controllers. In the case of one sensor we developed two models; a non-occupancy detection (NOD) and an occupancy detection (OD). NOD detects passing vehicles only whereas, OD detects vehicles that pass the sensor or stop at it. In both methods, we changed the sensor location relative to the traffic light's location. We then used two sensors to calculate number of vehicles waiting or approaching a traffic light. We tested different distances between these two sensors. Two sensors-based controller outperform the one sensor controller and produced results comparable to the ideal control that knows exact number of waiting vehicles. In analyzing our simulation, we found that the distance between the two sensors does not affect the performance of our design. Hence, placing both the sensors close to each others produce the best performance in terms of quality of the data and reduce energy consumption which leads to extending the life time of the WSN.

Measuring traffic flow plays an important role in intelligent transportation systems. In recent years, the technology of sensor network has been brought into the field due to their reliability and non-intrusiveness. In this paper, we propose a framework to reduce the installation and maintenance cost of traffic measuring sensor networks. The key to the solutions lies on predicting the complete measurements using the readings at a limited number of observing locations. We describe two correlation-based prediction methods and show that the Gaussian method is more informative and achieves better accuracy. We propose an analytical approach that eases the procedure of acquiring the Gaussian parameters. We demonstrate through experimental results that the model is correct and achieves prediction very close to the model learned over a large set of training data.

Collaborative filtering is one of the most successful techniques for recommendation systems and has been used in many commercial services provided by major companies including Amazon, TiVo and Netflix. In this paper we focus on memory-based collaborative filtering (CF). Existing CF techniques work well on dense data but poorly on sparse data. To address this weakness, we propose to use z-scores instead of explicit ratings and introduce a mechanism that adaptively combines global statistics with item-based values based on data density level. We present a new adaptive framework that encapsulates various CF algorithms and the relationships among them. An adaptive CF predictor is developed that can self adapt from user-based to item-based to hybrid methods based on the amount of available ratings. Our experimental results show that the new predictor consistently obtained more accurate predictions than existing CF methods, with the most significant improvement on sparse data sets. When applied to the Netflix Challenge data set, our method performed better than existing CF and singular value decomposition (SVD) methods and achieved 4.67% improvement over Netflix's system.

The energy-constrained wireless sensor nodes require the designs of energy-efficient content authentication protocols. In this paper, we combine the approaches of cryptography and content analysis to propose a new light-weight content authentication framework Cobra. In Cobra, only a small fraction of transmitted packets are protected with signatures. The unprotected packets are authenticated using the contents of the protected packets and a pre-learned correlation model. Cobra is resilient to four common types of data integrity attacks. Both our theoretical analysis and the experimental results show that Cobra offers high detection rate against aggressive attacks. Compared to three competing methods, it reduces the average data error by up to 60% and the security overhead by an order of one magnitude. As the attacks become more aggressive, Cobra's average data error is almost unaffected whereas the average data errors of the competing methods all grow linearly.

This paper proposes a new anchor placement algorithm for improving localization performance. Motivated from the result obtained on the anchor placement error analysis (Shi et al., 2005), an near-exhaustive search is done on anchor placement for localization variance. Based on the best anchor area found in the search, we design a selective mobile anchor placement scheme (SAP). Simulation result shows that the SAP anchor placement scheme is effective in a sensor network with a small anchor ratio. Localization simulation result shows both the position variance and accuracy are improved by more than 10% of the radio range.

Finding the multiple longest common subsequence (MLCS) is an important problem in the areas of bioinformatics and computational genomics. Approaches that are more efficient than the standard dynamic programming method have been introduced and successfully parallelized for the special cases of 2 sequences. However, the increasing complexity and size of biological data require an efficient method applicable to an arbitrary number of sequences as well as its efficient parallelization. A recently developed dominant points method for a general MLCS problem has been shown a significant performance improvement over the dynamic programming method, when number of sequences is larger than two. At the same time, the approach has revealed strong demand for its parallelization, in order to be applied to the larger families of sequences or sequences of the greater lengths. In this paper, we introduce an efficient parallel algorithm to find a MLCS for an arbitrary number of sequences, which is based on the dominant points method. When the number of processors is not greater than the size of alphabet multiplied by the number of sequences, the parallel algorithm is estimated to have the asymptotically linear speed up. We experimentally tested the algorithm using sets of randomly generated sequences over different alphabets as well as the protein sequences from a family of homologous proteins. We found that the performance of the algorithm increases with the number of input sequences and reaches a near-linear speedup for eight sequences.

In protein tertiary structure prediction, it is a crucial step to select near-native structures from a large number of candidate structural models. Despite much effort to tackle the problem of protein structure selection, the discerning power of current scoring functions is still unsatisfactory. In this paper, we developed a new clustering-based method for selecting near-native protein structures. Our method consists of three phases: filtering, clustering and cluster reduction, and centroid construction. Given a set of Cαprotein structures, we apply one or multiple existing scoring functions to filter out bad structures. Then, we group the remaining structures into clusters based on pair-wise similarity measured by RMSD. Each cluster is reduced iteratively to remove outliers and bad structures. Finally, we construct a centroid for each cluster by applying multi-dimensional scale techniques. The centroids are the final models. In experiments, we applied our method to a test set of representative proteins and obtained significant improvement over existing methods.

Wireless sensor networks have been applied to data gathering and target localization and tracking in large geograph- ical areas. In sparsely deployed sensor networks, optimal access point placement is a fundamental issue because of its marked influence on the localization accuracy and data communication rate and delay. In this paper, based on a representative sensor network system for lost hiker localization in wilderness areas, we study the problem of access point deployment in constrained environments with the goal of minimizing location uncertainty. We first identify several key problem features including the trail topology, the number of hikers, and hiker movement mode, and present theoretical analysis of the access-point placement problem for a given set of problem attributes. Then we propose algorithms to compute the location uncertainty, the optimal locations of access points and the necessary number of access points for a certain location uncertainty. The simulation results support our models.

In this paper, we propose a novel decentralized traffic light control using wireless sensor network. The system architecture is classified into three layers; the wireless sensor network, the localized traffic flow model policy, and the higher level coordination of the traffic lights agents. The wireless sensors are deployed on the lanes going in and out the intersection. These sensors detect vehicles' number, speed, etc. and send their data to the nearest Intersection Control Agent (ICA) which, determines the flow model of the inter- section depending on sensors' data (e.g., number of vehicles approaching a specific intersection). Coping with dynamic changes in the traffic volume is one of the biggest challenges in intelligent transportation system (ITS). Our main contribution is the real-time adaptive control of the traffic lights. Our aim is to maximize the flow of vehicles and reduce the waiting time while maintaining fairness among the other traffic lights. Each traffic light controlled intersection has an intersection control agent that collects information from the sensor nodes. An intersection control agent manages its intersection by controlling its traffic lights. Multiple intersection agents can exchange information among themselves to control a wider area.

There are many difficult challenges ahead in the design of an energy-efficient communication stack for wireless sensor networks. Due to the severe nodes' constraints, protocols have to be simple yet scalable. To this end, collective social insects behavior could be adopted to guide the design of these protocols. We exploit the simple behaviors of ant swarms in foraging and brood sorting to design a hierarchical and scalable data dissemination strategy. In order to allow a realistic evaluation, a comprehensive simulator involving critical components of the communication stack is incorporated. The proposed scheme promotes a uniform distribution of clusterheads, which subsequently enables substantial energy savings over other clustering algorithms.

Out of its importance to various applications and services, the geographical location of the sensed event is to be associated with the event itself being reported. Despite the numerous number of localization algorithms proposed, very few of them are really ad-hoc methods that are appropriate for sensor networks. In this paper, our contribution is double-folded. First, we design an experimental framework to evaluate localization methods for sensor networks. We use this framework to evaluate three localization methods: ad-hoc positioning system (APS), multi-dimensional scaling (MDS), and semi-definite programming (SDP). Using this evaluation, we identify five network properties that affect the localization accuracy. Second, we propose an adaptive localization method that we refer to as: network-aware positioning (NAP). NAP starts by assuming known network properties. Given these properties, NAP determines the best localization algorithm to use. Simulation results show that NAP performs the best among the three algorithms under all network conditions.

We propose an algorithm to optimize the delivery quality of fine-scalable coded images on peer-to-peer networks when a requesting peer is delay-sensitive and has to display the images within a certain delay bound. First, we introduce the peer assignment problem using an example. Then, we formally define and formulate this problem as two integer programming problems, from which we derive an optimal peer assignment solution. Finally, we carry out experiments to verify the excellent performance of the proposed peer assignment algorithm by comparing it with two heuristic schemes.

It is often useful to know the geographic positions of nodes in a communications network, but adding GPS receivers or other sophisticated sensors to every node can be expensive. MDS-MAP is a recent localization method based on multidimensional scaling (MDS). It uses connectivity information - who is within communications range of whom - to derive the locations of the nodes in the network, and can take advantage of additional data, such as estimated distances between neighbors or known positions for certain anchor nodes, if they are available. However, MDS-MAP is an inherently centralized algorithm and is therefore of limited utility in many applications. In this paper, we present a new variant of the MDS-MAP method, which we call MDS-MAP(P) standing for MDS-MAP using patches of relative maps, that can be executed in a distributed fashion. Using extensive simulations, we show that the new algorithm not only preserves the good performance of the original method on relatively uniform layouts, but also performs much better than the original on irregularly-shaped networks. The main idea is to build a local map at each node of the immediate vicinity and then merge these maps together to form a global map. This approach works much better for topologies in which the shortest path distance between two nodes does not correspond well to their Euclidean distance. We also discuss an optional refinement step that improves solution quality even further at the expense of additional computation.

Summary form only given. Knowing the positions of nodes in sensor networks are essential for many applications. We propose a new method, MDS-MAP(R), that determines the relative positions of nodes based on local distance information. Given a communication path between a starting node and a remote node, the nodes on the path first compute their local relative maps based on local distance estimates in a distributed fashion, using the MDS-MAP method. Then, the relative maps of adjacent nodes along the path are aligned based on their common nodes and their optimal linear transformation parameters are computed. Finally, the relative position of the remote node in the coordinate system of the starting node is determined by applying the sequence of linear transformations. The method can easily be extended from the single-path peer-to-peer communication situation to other types of communications such as multicasting or broadcasting. The accuracy of relative position estimation is affected by several factors, including the network connectivity, the errors of local distance measures, the length of the path, and the number of common nodes of two adjacent relative maps. Through simulation, we analyze the effect of these factors on the position estimation errors. We show that the new method performs well on both regular and irregular networks when there are enough connectivity and the distance measurement errors are small.

In dense sensor networks, density control is an important technique for prolonging network lifetime while providing sufficient sensing coverage. We present a new density control algorithm called NODC (non-overlapping density control). The goal of NODC is maximizing coverage while avoiding the overlap of sensing areas of active sensors. We derive the optimality conditions and develop a distributed algorithm. Experimental results show that NODC can use up to 20% fewer active nodes than OGDC with less than 5% reduction in coverage.

In ad-hoc wireless sensor networks, one approach for estimating the positions of sensor nodes is to use connectivity information of the network or local distance measurement. We compare two representative methods, multilateration and MDS-MAP, and study some of their key issues that affect the performance. First, we study the various optimization techniques and the effect of anchor selection for multilateration and the effects of the sizes of local maps and the refinement techniques for MDS-MAP. Secondly, we investigate the error property of localization by deriving the Cramer-Rao bound for different topologies and compare the bounds with the results of the two methods on different anchor placements. Previous results suggest that anchors should be placed on the perimeter of the network to get more accurate solutions. Our results show that MDS-MAP tolerates "ill placement" of anchors much better than multilateration. Finally, we compare the performance of the two methods through extensive simulation and identify the situations when one method is better than the other.

We propose an approach that uses connectivity information - who is within communications range of whom - to derive the locations of nodes in a network. The approach can take advantage of additional information, such as estimated distances between neighbors or known positions for certain anchor nodes, if it is available. It is based on multidimensional scaling (MDS), an efficient data analysis technique that takes O(n/sup 3/) time for a network of n nodes. Unlike previous approaches, MDS takes full advantage of connectivity or distance information between nodes that have yet to be localized. Two methods are presented: a simple method that builds a global map using MDS and a more complicated one that builds small local maps and then patches them together to form a global map. Furthermore, least-squares optimization can be incorporated into the methods to further improve the solutions at the expense of additional computation. Through simulation studies on uniform as well as irregular networks, we show that the methods achieve more accurate solutions than previous methods, especially when there are few anchor nodes. They can even yield good relative maps when no anchor nodes are available.

It is often useful to know the geographic positions of nodes in a communications network, but adding GPS receivers or other sophisticated sensors to every node can be expensive. We present an algorithm that uses connectivity information who is within communications range of whom to derive the locations of the nodes in the network. The method can take advantage of additional information, such as estimated distances between neighbors or known positions for certain anchor nodes, if it is available. The algorithm is based on multidimensional scaling, a data analysis technique that takes O(n3) time for a network of n nodes. Through simulation studies, we demonstrate that the algorithm is more robust to measurement error than previous proposals, especially when nodes are positioned relatively uniformly throughout the plane. Furthermore, it can achieve comparable results using many fewer anchor nodes than previous methods, and even yields relative coordinates when no anchor nodes are available.

In this paper, we advocate a general constraint-based control framework that is highly promising for building control systems with complex dynamic structures, such as modular self-reconfigurable robots. In this framework, a controller consists of constraint solving components distributed in a network of embedded processors. Constraint solvers are goal oriented deliberative agents that can be used as control regulators or as information retrievers. The framework is built on the attribute/service model (ASM), a middleware for coordinating actuators, sensors and tasks in distributed real-time embedded systems. The communications and coordination among the services are realized via shared attributes. Examples of controlling a modular self-reconfigurable robot are illustrated in the paper.

In this paper, we present two ways to improve the precision of HITS-based algorithms on Web documents. First, by analyzing the limitations of current HITS-based algorithms, we propose a new weighted HITS-based method that assigns appropriate weights to in-links of root documents. Then, we combine content analysis with HITS-based algorithms and study the effects of four representative relevance scoring methods, VSM, Okapi, TLS, and CDR, using a set of broad topic queries. Our experimental results show that our weighted HITS-based method performs significantly better than Bharat's improved HITS algorithm. When we combine our weighted HITS-based method or Bharat's HITS algorithm with any of the four relevance scoring methods, the combined methods are only marginally better than our weighted HITS-based method. Between the four relevance-scoring methods, there is no significant quality difference when they are combined with a HITS-based algorithm.

No abstract available.

As the traditional relational databases and the new XML document repositories are being widely used on the Web as information storage, there is a great need for easy access to the information sources, particularly through natural language interactions. In this paper, we present the design and implementation of an intelligent system that interacts with users using a natural language, English, and retrieves information from sources, for the users. The system consists of four major parts: part of speech tagging, query knowledge base, query formation, and answer synthesis. In implementation, the system first uses QTAG, a Hidden Markov Model based speech part tagger, to tag each word in the input sentence. Then, important words in the main phrase axe identified. A thesaurus is applied to reduce the important words to basic keywords, which are used to query the database. The query is formed based on the query knowledge stored in the query knowledge base. Finally, the query result is synthesized into an English sentence, which is presented to the user as the answer. With an efficient part of speech tagger, intelligent subsystems for query formation and synthesis of query result, and user-friendly interface, the intelligent system can answer questions effectively.

Active learning is an effective learning approach. In this article we present an intelligent agent-assisted environment for active learning to better support the student-centered, self-paced, and highly interactive learning approach. The environment uses the student’s learning-related profile such as learning style and background knowledge in selecting, organizing, and presenting learning material, and it adopts a new approach to course content organization and delivery based on smart instructional components that can be integrated into a wide range of courses. The environment is being implemented using the prevalent Internet, Web, digital library, and multiagent technologies.

Active learning is an effective learning approach. In this article we present an intelligent agent-assisted environment for active learning to better support the student-centered, selfpaced, and highly interactive learning approach. The environment uses the students learningrelated profile such as learning style and background knowledge in selecting, organizing, and presenting learning material, and it adopts a new approach to course content organization and delivery based on smart instructional components that can be integrated into a wide range of courses. The environment is being implemented using the prevalent Internet, Web, digital library, and multiagent technologies.

We present a new statistical method for evaluating search engines' precision performance based on sample queries. The method consists of relevance evaluation and statistical comparison. In relevance evaluation, we present two scoring algorithms: one is a term-based algorithm based on the vector space model, and the other is a new three-level algorithm modeled after manual methods commonly used in information retrieval studies. In statistical comparison, we apply a statistical metric probability of win, in ranking the search engines. Based on a set of sample queries, our method evaluates the relevance of the pages returned by the search engines and compares them statistically In the experiment, our method was applied to three search engines, AltaVista, Google, and InfoSeek, using two query sets derived from the domain of parallel and distributed processing. Our results show that the three-level scoring algorithm with a typical set of parameters obtained results consistent with those obtained using the manual method, whereas the term-based algorithm did not.

In recent years, agent technology has been used increasingly in information management and distributed computing. A CSE curriculum that cultivates the knowledge of agent technology will increase the likelihood that the next generation of IT professionals have the background needed to design and develop software systems that are scalable, reliable, adaptable, and secure. In this paper, we present the rationale and our practice in incorporating agent technology into the CSE curriculum. We develop agent-based teaching materials and software modules and apply them to existing CSE courses including artificial intelligence, parallel and distributed processing, networking, and software engineering. Promising results have been obtained in teaching two graduate level courses using agent components.

In this paper, we present a multi-agent system for supporting student-centered, self-paced, and highly interactive learning in undergraduate computer science education. The system is based on a hybrid problem-based and case-based learning model, for both creative problem solving and mechanical experience simulation. It aims at enhancing the effectiveness of the undergraduate learning experience in computer science. Implemented using the prevalent Internet, Web, and digital library technologies, the system adopts an open architecture design and targets at large-scale, distributed operations. In the initial implementation of the system, a number of prototypes using different Java-based software environments have been developed. They offer tradeoffs in system performance and design complexity.

We present an agent-based approach for medical image interpretation. The system is based on the concept of active fusion for image recognition and is composed of two major types of intelligent agents: radiologist agents and patient representative agents. A patient representative agent takes images from the patient through a Web-based interface, asks for multiple opinions from radiologist agents in interpreting them, and then integrates the opinions for the user. A radiologist agent decomposes the image interpretation task into smaller subtasks, uses multiple agents to solve the subtasks, and combines the solutions to the subtasks intelligently to solve the image interpretation problem.

We present the design and implementation of an intelligent Web representative based on prevalent Internet and Web technologies. The design includes three parts: natural language parser, knowledge representation, and communication between the user and the computer. With an efficient pseudo-natural language processing, a database of knowledge stored in XML, and the user communication through Microsoft Agent, the intelligent agent can answer questions effectively.

We have proposed the discrete Lagrange-multiplier method (DLM) to solve satisfiability problems. Instead of restarting from a new starting point when the search reaches a local minimum in the objective space, the Lagrange multipliers of violated constraints in DLM provide a force to lead the search out of the local minimum and move it in a direction provided by the multipliers. We present the theoretical foundation of DLM for solving SAT problems and discuss some implementation issues. We study the performance of DLM on a set of hard satisfiability benchmark instances, and show the importance of dynamic scaling of Lagrange multipliers and the flat-move strategy. We show that DLM can perform better than competing local-search methods when its parameters are selected properly.

In this paper, we present a new discrete Lagrangian optimization method for designing multiplierless QMF (quadrature mirror filter) filter banks. In multiplierless QMF filter banks, filter coefficients are powers-of-two (PO2) where numbers are represented as sums or differences of powers of two (also cabled Canonical Signed Digit-CSD-representation), and multiplications can be carried out as additions, subtractions and shifting. We formulate the design problem as a nonlinear discrete constrained optimization problem, using the reconstruction error as the objective, and other performance metrics as constraints. One of the major advantages of this formulation is that it allows us to search for designs that improve over the best existing designs with respect to all performance metrics, rather than finding designs that trade one performance metric for another. We show that our design method can find designs that improve over Johnston's benchmark designs using a maximum of three to six ONE bits in each filter coefficient.

We study various global optimization methods for designing QMF (quadrature mirror filter) filter banks. We formulate the design problem as a nonlinear constrained optimization problem, using the reconstruction error as the objective, and other performance metrics as constraints. This formulation allows us to search for designs that are improvements over the best existing designs. We present NOVEL, a global optimization method for solving nonlinear continuous constrained optimization problems. We show that NOVEL finds better designs with respect to simulated annealing and genetic algorithms in solving QMF benchmark design problems. We also show that relaxing the constraints on transition bandwidth and stopband energy leads to significant improvements in the other performance measures.

Nonlinear constrained optimization problems can be solved by a Lagrange-multiplier method in a continuous space or by its extended discrete version in a discrete space. These methods rely on gradient descents in the objective space to find high-quality solutions, and gradient ascents in the Lagrangian space to satisfy the constraints. The balance between descents and ascents depends on the relative weights between the objective and the constraints that indirectly control the convergence speed and solution quality of the method. To improve convergence speed without degrading solution quality, the authors propose an algorithm to dynamically control these relative weights. Starting from an initial weight, the algorithm automatically adjusts the weights based on the behaviour of the search progress. With this strategy, one is able to eliminate divergence, reduce oscillations, and speed up convergence. They show improved convergence behaviour of the proposed algorithm on both nonlinear continuous and discrete problems.

In this paper, we study the performance of various IDA*-style searches and investigate methods to improve their performance by predicting in each stage the threshold to use for pruning. We first present three models to approximate the distribution of number of search nodes by lower bounds: exponential, geometric, and linear, and illustrate these distributions based on some well-known combinatorial search problems. We then show the performance of an ideal IDA* algorithm and identify reasons why existing IDA*-style algorithms perform well. In practice, we will be able to know from previous experience the distribution for a given problem instance but will not be able to determine the parameters of the distribution. Hence, we develop RIDA*, a method that estimates dynamically the parameters of the distribution, and predicts the best threshold to use, Finally, we compare the performance of several IDA*-style algorithms-Korf's IDA*, RIDA*, IDA* CR and DFS*-on several application problems, and identify conditions under which each of these algorithms will perform well.>

Two new crossover operators in genetic algorithms for solving some combinatorial problems with ordering are presented. One is enhanced order crossover (EOX). The other, GREE, is a heuristic crossover for a class of combinatorial optimization problems, such as traveling salesman problems (TSPs). Genetic algorithms using GREE as unique crossover run very fast and get good solutions. Combining GREE with EOX, genetic algorithms can find optimal or very near optimal solutions in a rather short time.>