Search

Educational Data management is a critical task for the researchers due to mammoth data generated by sensors and IoT (Internet of Things) devices. Managing this huge volume of data, cleaning this data from impurities is an inherent need. DF (Data Fusion) processes combine data from multiple sources based on their similarity for an easy management. DF processes focus on many factors like nature of data and application that uses that data. Many DFAs (Data Fusion approaches) have been proposed without detailing on the context for integrating data in fusion tasks. This work attempts to cover this gap of context’s relevance by proposing a technique CDFT (Context aware Data Fusion technique). In this research work, initially data from IoT devices will be gathered and pre-processed to make it clear for the fusion processing. In this work, boundary based noise reduction algorithm is introduced for data pre-processing which attempts to label the unlabelled attributes in the data’s that are gathered, so that data fusion can be done accurately. After pre-processing Context aware data fusion is performed which will combine the data’s from multiple IoT devices together with the concern of context. Finally this combined data will be learnt using the convolution neural network for data fusion performance checking. The proposed CDFT is simulated on Matlab whose results prove that the proposed technique obtains optimal outcomes.

Objective: Laryngopharyngeal Reflux Disease (LPRD) is prevalent and has a range of symptoms. However, diagnosing LPRD is difficult because of the lack of specific symptoms or clinical gold standard. An objective and reliable test, which does not rely on a perfect clinical gold standard, is required in clinical practice. Methods: 60 normal volunteers and 74 confirmed Laryngopharyngeal Reflux (LPR) patients were labelled based on the combined consideration of the reflux symptom index, reflux finding score, 24h oropharyngeal pH monitoring and results of anti-reflux treatment. 72 candidate features were extracted from pH recordings and the most efficient feature combination was detected using a stepwise wrapper method. The labelled data were combined with 1552 unlabelled data for feature selection and model training using semi-supervised learning. A latent class model method was used to assess the proposed model based on 64 additional validation data and an imperfect clinical reference test. Results: A new score (named W score), which significantly improved the sensitivity of the LPRD test (82.67% vs. 24.09%) and has a relatively high specificity (80.19%), was proposed. W score concurs with the complicated clinical test. Conclusion: W score which significantly improves LPRD diagnostic efficiency, can aid the clinical diagnosis of LPRD. W score provides an objective, efficient, and reliable indicator for the application of anti-reflux treatment in clinical practice. Significance: More LPRD patients can benefit from anti-reflux treatment in clinical practice and fewer patients may suffer from, for example, the side effects of unnecessary long-term Proton Pump Inhibitors (PPI) treatments.

In this work, we study music/video cross-modal recommendation, i.e. recommending a music track for a video or vice versa. We rely on a self-supervised learning paradigm to learn from a large amount of unlabelled data. We rely on a self-supervised learning paradigm to learn from a large amount of unlabelled data. More precisely, we jointly learn audio and video embeddings by using their co-occurrence in music-video clips. In this work, we build upon a recent video-music retrieval system (the VM-NET), which originally relies on an audio representation obtained by a set of statistics computed over handcrafted features. We demonstrate here that using audio representation learning such as the audio embeddings provided by the pre-trained MuSimNet, OpenL3, MusicCNN or by AudioSet, largely improves recommendations. We also validate the use of the cross-modal triplet loss originally proposed in the VM-NET compared to the binary cross-entropy loss commonly used in self-supervised learning. We perform all our experiments using the Music Video Dataset (MVD).

We tackle the problem of discovering novel classes in an image collection given labelled examples of other classes. We present a new approach called AutoNovel to address this problem by combining three ideas: (1) we suggest that the common approach of bootstrapping an image representation using the labelled data only introduces an unwanted bias, and that this can be avoided by using self-supervised learning to train the representation from scratch on the union of labelled and unlabelled data; (2) we use ranking statistics to transfer the model's knowledge of the labelled classes to the problem of clustering the unlabelled images; and, (3) we train the data representation by optimizing a joint objective function on the labelled and unlabelled subsets of the data, improving both the supervised classification of the labelled data, and the clustering of the unlabelled data. Moreover, we propose a method to estimate the number of classes for the case where the number of new categories is not known a priori. We evaluate AutoNovel on standard classification benchmarks and substantially outperform current methods for novel category discovery. In addition, we also show that AutoNovel can be used for fully unsupervised image clustering, achieving promising results., TPAMI 2021, code: http://www.robots.ox.ac.uk/~vgg/research/auto_novel/. arXiv admin note: substantial text overlap with arXiv:2002.05714

Current work on lane detection relies on large manually annotated datasets. We reduce the dependency on annotations by leveraging massive cheaply available unlabelled data. We propose a novel loss function exploiting geometric knowledge of lanes in Hough space, where a lane can be identified as a local maximum. By splitting lanes into separate channels, we can localize each lane via simple global max-pooling. The location of the maximum encodes the layout of a lane, while the intensity indicates the the probability of a lane being present. Maximizing the log-probability of the maximal bins helps neural networks find lanes without labels. On the CULane and TuSimple datasets, we show that the proposed Hough Transform loss improves performance significantly by learning from large amounts of unlabelled images., ICIP2021

Subspace clustering is a popular method for clustering unlabelled data. However, the computational cost of the subspace clustering algorithm can be unaffordable when dealing with a large data set. Using a set of dimension sketched data instead of the original data set can be helpful for mitigating the computational burden. Thus, finding a way for dimension sketching becomes an important problem. In this paper, a new dimension sketching algorithm is proposed, which aims to select informative dimensions that have significant effects on the clustering results. Experimental results reveal that this method can significantly improve subspace clustering performance on both synthetic and real-world datasets, in comparison with two baseline methods.

In this paper, U-Net-based method for robust adherent cell segmentation for quantitative phase microscopy image is designed and optimised. We designed and evaluated four specific post-processing pipelines. To increase the transferability to different cell types, non-deep learning transfer with adjustable parameters is used in the post-processing step. Additionally, we proposed a self-supervised pretraining technique using nonlabelled data, which is trained to reconstruct multiple image distortions and improved the segmentation performance by from 0.67 to 0.70 of Object-wise Intersection over Union. Moreover, we publish a new dataset of manually labelled images suitable for this task together with the unlabelled data for self-supervised pretraining.Graphical AbstractHighlightsFour strategies for instance cell segmentation with U-Net were compared.Specialised post-processing pipelines with tunable/optimizable parameters were designed for each segmentation strategy.Transferability to different cell types by optimisation of post-processing parameters was tested.The proposed self-supervised pretraining method improved both segmentation performance and transferability to different cell types.A new manually labelled quantitative phase imaging dataset for cell segmentation with unlabelled data for self-supervised pretraining was created.

2021 24th Euromicro Conference on Digital System Design (DSD) : [Proceedings] - IEEE, 2021. - ISBN 978-1-6654-2703-6 - doi:10.1109/DSD53832.2021.00028 24th Euromicro Conference on Digital System Design (DSD), Palermo, Italy, 1 Sep 2021 - 3 Sep 2021; 5 pp. (2021). doi:10.1109/DSD53832.2021.00028, Anomaly detection algorithms solve the problem of identifying unexpected values in data sets. Such algorithms have been classically used for cleaning unlabelled data sets from potentially unwanted values. However, the ability to detect outlying values in data sets can also be used to detect anomalies in systems. Semi-supervised anomaly detection algorithms learn from data for known correct behavior. Such algorithms have been used in various fields, e.g., system security, fault detection, medical applications.In this paper, we use the Area Under the Receiver Operating Characteristic (AUROC) score to evaluate algorithms for semi-supervised anomaly detection when applied to high-integrity distributed digital systems. We identify the relevant parameter for each algorithm and observe how the parameter influences the score and the runtime.

Non-intrusive speech quality assessment is a crucial operation in multimedia applications. The scarcity of annotated data and the lack of a reference signal represent some of the main challenges for designing efficient quality assessment metrics. In this paper, we propose two multi-task models to tackle the problems above. In the first model, we first learn a feature representation with a degradation classifier on a large dataset. Then we perform MOS prediction and degradation classification simultaneously on a small dataset annotated with MOS. In the second approach, the initial stage consists of learning features with a deep clustering-based unsupervised feature representation on the large dataset. Next, we perform MOS prediction and cluster label classification simultaneously on a small dataset. The results show that the deep clustering-based model outperforms the degradation classifier-based model and the 3 baselines (autoencoder features, P.563, and SRMRnorm) on TCD-VoIP. This paper indicates that multi-task learning combined with feature representations from unlabelled data is a promising approach to deal with the lack of large MOS annotated datasets., Comment: Published in 2021 13th International Conference on Quality of Multimedia Experience (QoMEX)

Although deep face recognition benefits significantly from large-scale training data, a current bottleneck is the labelling cost. A feasible solution to this problem is semi-supervised learning, exploiting a small portion of labelled data and large amounts of unlabelled data. The major challenge, however, is the accumulated label errors through auto-labelling, compromising the training. In this paper, we present an effective solution to semi-supervised face recognition that is robust to the label noise aroused by the auto-labelling. Specifically, we introduce a multi-agent method, named GroupNet (GN), to endow our solution with the ability to identify the wrongly-labelled samples and preserve the clean samples. We show that GN alone achieves the leading accuracy in traditional supervised face recognition even when the noisy labels take over 50% of the training data. Further, we develop a semi-supervised face recognition solution, named Noise Robust Learning-Labelling (NRoLL), which is based on the robust training ability empowered by GN. It starts with a small amount of labelled data and consequently conducts high-confidence labelling on a large amount of unlabelled data to boost further training. The more data is labelled by NRoLL, the higher confidence is with the label in the dataset. To evaluate the competitiveness of our method, we run NRoLL with a rough condition that only one-fifth of the labelled MSCeleb is available and the rest is used as unlabelled data. On a wide range of benchmarks, our method compares favorably against the state-of-the-art methods.

The wide diversity of categories encountered in big data applications means certain classes will be labelled while many other classes will remain unlabelled. A classification domain would consequently consist of known source classes for which labelled samples are available, and unknown novel classes that do not have any labelled samples. Given the assumption that novel classes are far more prevalent than source classes, it is appropriate to define an encapsulating ’unknown’ class for all such novel classes. To be practically useful, classification systems must classify over both the source and the unknown novel domains. Regularly available data often includes unlabelled samples that, intuitively, belong to both source and novel classes. Including scattered unlabelled data with source labelled data during model training provides tractable means to learn classification boundaries between source classes and the unknown class. Da et al. were the first to introduce this framework as learning with augmented class by exploiting unlabelled data or LACU [6]. In this work, we promote the LACU paradigm to the neural network research space with the first LACU-enabled neural model. Using simple neural architectures, our proposed method produces state-of-the-art results when compared to previously published LACU works. With neural networks more capable of handling large datasets, this work takes us one step closer to building big data classifiers capable of known and unknown classification.

The Internet of Things (IoT) is a network of distributed devices or sensors connected through the internet to allow gathering and sharing of data. The data generated by these devices is affected by anomalies or abnormal behaviour due to attack issues, or breakdown in devices, as examples. However, most current anomaly detection systems rely on labelled data, while the class labels for IoT data are usually unavailable. Furthermore, the manual labelling task is expensive and time-consuming to perform due to the need for domain experts. More importantly, the volume of data in the IoT is growing rapidly, creating a need to predict the classification labels for future data. This study proposes a Hybrid Learning Model which uses both Clustering and Classification methods (HLMCC) to automate the labelling process and detect anomalies in IoT data. The model consists of two practical phases, automatic labelling and detecting anomalies. First, the HLMCC groups the data into normal and anomaly clusters by adopting Hierarchical Affinity Propagation (HAP) clustering. Second, the labelled data obtained from the clustering phase is used to train the Decision Trees (DTs) and to classify future unseen data. The results show that the HLMCC is able to automate the labelling of data, which is beneficial to minimize human involvement. Moreover, HLMCC outperforms the DTs on the originally labelled datasets and the state-of-the-art model over a wide range of evaluation metrics based on the average ranks. HLMCC produces the highest average ranks against other models in terms of False Positive Rate (FPR), recall, precision and the Area Under the Precision-Recall curve (AUCPR) with 1.8, 1.6, 1.8 and 1.8, respectively.

The cork oak (Quercus suber L.) is periodically harvested for bark (cork) throughout its lifetime. Trees undergo physiological changes as they age which affect stem diameter growth and their sensitivity to climate. However, little is known about trees age- or size-related growth changes and it remains unclear if trees of different ages (sizes) have similar climate-growth relationships. In this study, we examined the increment in stem basal area of 47 randomly selected (large and small) cork oaks over a 12-year period to assess divergent climate-growth relationships. Our approach, using a machine learning algorithm on unlabelled data sets of basal area increments, successfully filtered out tree-clusters that suggested a size (age)-dependent growth response to climate. On average, the basal area increment was more than three times larger in larger-trees clusters compared with smaller-trees clusters. A large tree (diameter >75 cm) on average added 105 cm² y⁻¹ to its basal area against 25 cm² y⁻¹ in a small tree (diameter

Prior work on plant species classification predominantly focuses on building models from isolated plant attributes. Hence, there is a need for tools that can assist in species identification in the natural world. We present a novel and robust two-fold approach capable of identifying trees in a real-world natural setting. Additionally, we leverage unlabelled data through deep semi-supervised learning and demonstrate superior performance to supervised learning. Our single-GPU implementation for feature recognition uses minimal annotated data and achieves accuracies of 93.96% and 93.11% for leaves and bark, respectively. Further, we extract feature-specific datasets of 50 species by employing this technique. Finally, our semi-supervised species classification method attains 94.04% top-5 accuracy for leaves and 83.04% top-5 accuracy for bark.

Political instability, conflicts and inequalities result into significant flows of people worldwide, moving to different countries in search of a better life, safety or to be reunited with their families. Irregular crossings into Europe via sea routes, despite not being new, have recently increased together with the loss of lives of people in the attempt to reach EU shores. This highlights the need to find ways to improve the understanding of what is happening at sea. This paper, intends to expand the knowledge available on practices among smugglers and contribute to early warning and maritime situational awareness. By identifying smuggling techniques and based on anomaly detection methods, behaviours of interest are modelled and one class support vector machines are used to classify unlabelled data and detect potential smuggling vessels. Nine vessels are identified as potentially carrying irregular migrants and refugees. Though, further inspection of the results highlights possible misclassifications caused by data gaps and limited knowledge on smuggling tactics. Accepted classifications are considered subject to further investigation by the authorities.

Using lobbying data from OpenSecrets.org, we offer several experiments applying machine learning techniques to predict if a piece of legislation (US bill) has been subjected to lobbying activities or not. We also investigate the influence of the intensity of the lobbying activity on how discernible a lobbied bill is from one that was not subject to lobbying. We compare the performance of a number of different models (logistic regression, random forest, CNN and LSTM) and text embedding representations (BOW, TF-IDF, GloVe, Law2Vec). We report results of above 0.85\% ROC AUC scores, and 78\% accuracy. Model performance significantly improves (95\% ROC AUC, and 88\% accuracy) when bills with higher lobbying intensity are looked at. We also propose a method that could be used for unlabelled data. Through this we show that there is a considerably large number of previously unlabelled US bills where our predictions suggest that some lobbying activity took place. We believe our method could potentially contribute to the enforcement of the US Lobbying Disclosure Act (LDA) by indicating the bills that were likely to have been affected by lobbying but were not filed as such.

In the paper we revisit the problem of fitting logistic regression to positive and unlabelled data. There are two key contributions. First, a new light is shed on the properties of frequently used naive method (in which unlabelled examples are treated as negative). In particular we show that naive method is related to incorrect specification of the logistic model and consequently the parameters in naive method are shrunk towards zero. An interesting relationship between shrinkage parameter and label frequency is established. Second, we introduce a novel method of fitting logistic model based on simultaneous estimation of vector of coefficients and label frequency. Importantly, the proposed method does not require prior estimation, which is a major obstacle in positive unlabelled learning. The method is superior in predicting posterior probability to both naive method and weighted likelihood method for several benchmark data sets. Moreover, it yields consistently better estimator of label frequency than other two known methods. We also introduce simple but powerful representation of positive and unlabelled data under Selected Completely at Random assumption which yields straightforwardly most properties of such model.

Automated pathology segmentation remains a valuable diagnostic tool in clinical practice. However, collecting training data is challenging. Semi-supervised approaches by combining labelled and unlabelled data can offer a solution to data scarcity. An approach to semi-supervised learning relies on reconstruction objectives (as self-supervision objectives) that learns in a joint fashion suitable representations for the task. Here, we propose Anatomy-Pathology Disentanglement Network (APD-Net), a pathology segmentation model that attempts to learn jointly for the first time: disentanglement of anatomy, modality, and pathology. The model is trained in a semi-supervised fashion with new reconstruction losses directly aiming to improve pathology segmentation with limited annotations. In addition, a joint optimization strategy is proposed to fully take advantage of the available annotations. We evaluate our methods with two private cardiac infarction segmentation datasets with LGE-MRI scans. APD-Net can perform pathology segmentation with few annotations, maintain performance with different amounts of supervision, and outperform related deep learning methods.

Domain pretraining followed by task fine-tuning has become the standard paradigm for NLP tasks, but requires in-domain labelled data for task fine-tuning. To overcome this, we propose to utilise domain unlabelled data by assigning pseudo labels from a general model. We evaluate the approach on two clinical STS datasets, and achieve r= 0.80 on N2C2-STS. Further investigation reveals that if the data distribution of unlabelled sentence pairs is closer to the test data, we can obtain better performance. By leveraging a large general-purpose STS dataset and small-scale in-domain training data, we obtain further improvements to r= 0.90, a new SOTA.

Lack of labelled data in the target domain for training is a common problem in domain adaptation. To overcome this problem, we propose a novel unsupervised domain adaptation method that combines projection and self-training based approaches. Using the labelled data from the source domain, we first learn a projection that maximises the distance among the nearest neighbours with opposite labels in the source domain. Next, we project the source domain labelled data using the learnt projection and train a classifier for the target class prediction. We then use the trained classifier to predict pseudo labels for the target domain unlabelled data. Finally, we learn a projection for the target domain as we did for the source domain using the pseudo-labelled target domain data, where we maximise the distance between nearest neighbours having opposite pseudo labels. Experiments on a standard benchmark dataset for domain adaptation show that the proposed method consistently outperforms numerous baselines and returns competitive results comparable to that of SOTA including self-training, tri-training, and neural adaptations.

Deep learning methods have been applied to Chinese named entity recognition for the online medical consultation. They require a large number of marked samples. However, no such database is available at present. This paper begins with constructing a larger labelled Chinese texts database for the online medical consultation. Second, a basic framework unit is proposed, which is pre-trained by the transfer learning from both Bidirectional language model and Mask language model trained on the larger unlabelled data. Finally, cross domains adversarial learning (CDAL) for Chinese named entity recognition is proposed to further improve the performance, which not only uses the pre-trained basic framework unit, but also uses the adversarial multi-task learning on both electronic medical record texts and online medical consultation texts. Experimental results validate the effectiveness of CDAL.

In this paper we illustrate a system aimed at solving a longstanding and challenging problem: acquiring a classifier to automatically annotate bibliographic records by starting from a huge set of unbalanced and unlabelled data. We illustrate the main features of the dataset, the learning algorithm adopted, and how it was used to discriminate philosophical documents from documents of other disciplines. One strength of our approach lies in the novel combination of a standard learning approach with a semantic one: the results of the acquired classifier are improved by accessing a semantic network containing conceptual information. We illustrate the experimentation by describing the construction rationale of training and test set, we report and discuss the obtained results and conclude by drawing future work.

Improving a semi-supervised image segmentation task has the option of adding more unlabelled images, labelling the unlabelled images or combining both, as neither image acquisition nor expert labelling can be considered trivial in most clinical applications. With a laparoscopic liver image segmentation application, we investigate the performance impact by altering the quantities of labelled and unlabelled training data, using a semi-supervised segmentation algorithm based on the mean teacher learning paradigm. We first report a significantly higher segmentation accuracy, compared with supervised learning. Interestingly, this comparison reveals that the training strategy adopted in the semi-supervised algorithm is also responsible for this observed improvement, in addition to the added unlabelled data. We then compare different combinations of labelled and unlabelled data set sizes for training semi-supervised segmentation networks, to provide a quantitative example of the practically useful trade-off between the two data planning strategies in this surgical guidance application.

We propose an activity recognition model balancing accuracy, overhead, data labelling.We propose a similarity measurement method to effectively discover activity patterns.We perform comprehensive experimental and comparison studies to validate our method. In the past decades, activity recognition had aroused great interest for the community of context-awareness computing and human behaviours monitoring. However, most of the previous works focus on supervised methods in which the data labelling is known to be time-consuming and sometimes error-prone. In addition, due to the randomness and erratic nature of human behaviours in realistic environments, supervised models trained with data from certain subject might not be scaled to others. Further more, unsupervised methods, with little knowledge about the activities to be recognised, might result in poor performance and high clustering overhead. To this end, we propose an activity recognition model with labelled and unlabelled data in smart environments. With small amount of labelled data, we discover activity patterns from unlabelled data based on proposed similarity measurement algorithm. Our system does not require large amount of data to be labelled while the proposed similarity measurement method is effective to discover length-varying, disordered and discontinuous activity patterns in smart environments. Therefore, our methods yield comparable performance with much less labelled data when compared with traditional supervised activity recognition, and achieve higher accuracy with lower clustering overhead compared with unsupervised methods. The experiments based on real datasets from the smart environments demonstrate the effectiveness of our method, being able to discover more than 90% of original activities from the unlabelled data, and the comparative experiments show that our methods are capable of providing a better trade-off, regarding the accuracy, overhead and labelling efforts, between the supervised and unsupervised methods.

Despite the widespread use of supervised learning methods for speech emotion recognition, they are severely restricted due to the lack of sufficient amount of labelled speech data for the training. Considering the wide availability of unlabelled speech data, therefore, this paper proposes semisupervised autoencoders to improve speech emotion recognition. The aim is to reap the benefit from the combination of the labelled data and unlabelled data. The proposed model extends a popular unsupervised autoencoder by carefully adjoining a supervised learning objective. We extensively evaluate the proposed model on the INTERSPEECH 2009 Emotion Challenge database and other four public databases in different scenarios. Experimental results demonstrate that the proposed model achieves state-of-the-art performance with a very small number of labelled data on the challenge task and other tasks, and significantly outperforms other alternative methods.

This paper seeks to examine the effect of including background knowledge in the form of character pre-trained neural language model (LM), and data bootstrapping to overcome the problem of unbalanced limited resources. As a test, we explore the task of language identification in mixed-language short non-edited texts with an under-resourced language, namely the case of Algerian Arabic for which both labelled and unlabelled data are limited. We compare the performance of two traditional machine learning methods and a deep neural networks (DNNs) model. The results show that overall DNNs perform better on labelled data for the majority categories and struggle with the minority ones. While the effect of the untokenised and unlabelled data encoded as LM differs for each category, bootstrapping, however, improves the performance of all systems and all categories. These methods are language independent and could be generalised to other under-resourced languages for which a small labelled data and a larger unlabelled data are available.

We present a machine learning pipeline that identifies key sentences in abstracts of oncological articles to aid evidence-based medicine. This problem is characterized by the lack of gold standard datasets, data imbalance and thematic differences between available silver standard corpora. Additionally, available training and target data differs with regard to their domain (professional summaries vs. sentences in abstracts). This makes supervised machine learning inapplicable. We propose the use of two semi-supervised machine learning approaches: To mitigate difficulties arising from heterogeneous data sources, overcome data imbalance and create reliable training data we propose using transductive learning from positive and unlabelled data (PU Learning). For obtaining a realistic classification model, we propose the use of abstracts summarised in relevant sentences as unlabelled examples through Self-Training. The best model achieves 84% accuracy and 0.84 F1 score on our dataset

Current supervised parsers are limited by the size of their labelled training data, making improving them with unlabelled data an important goal. We show how a state-of-the-art CCG parser can be enhanced, by predicting lexical categories using unsupervised vector-space embeddings of words. The use of word embeddings enables our model to better generalize from the labelled data, and allows us to accurately assign lexical categories without depending on a POS-tagger. Our approach leads to substantial improvements in dependency parsing results over the standard supervised CCG parser when evaluated on Wall Street Journal (0.8%), Wikipedia (1.8%) and biomedical (3.4%) text. We compare the performance of two recently proposed approaches for classification using a wide variety of word embeddings. We also give a detailed error analysis demonstrating where using embeddings outperforms traditional feature sets, and showing how including POS features can decrease accuracy.

Attention Deficit Hyperactive Disorder (ADHD) is a disruptive neurobehavioral disorder characterized by abnormal behavioral patterns in attention, perusing activity, acting impulsively and combined types. It is predominant among school going children and it is tricky to differentiate between an active and an ADHD child. Misdiagnosis and undiagnosed cases are very common. Behavior patterns are identified by the mentors in the academic environment who lack skills in screening those kids. Hence an unsupervised learning algorithm can cluster the behavioral patterns of children at school for diagnosis of ADHD. In this paper, we propose a hierarchical clustering algorithm to partition the dataset based on attribute dependency (HCAD). HCAD forms clusters of data based on the high dependent attributes and their equivalence relation. It is capable of handling large volumes of data with reasonably faster clustering than most of the existing algorithms. It can work on both labeled and unlabelled data sets. Experimental results reveal that this algorithm has higher accuracy in comparison to other algorithms. HCAD achieves 97% of cluster purity in diagnosing ADHD. Empirical analysis of application of HCAD on different data sets from UCI repository is provided.

What is the simplest thing you can do to solve a problem? In the contextof semi-supervised feature selection, we tackle exactly this { how much we can gain from two simple classifier-independent strategies. If we have some binary labelled data and some unlabelled, we could assume the unlabelled data are all positives, or assume them all negatives. These minimalist, seemingly naive, approaches have not previously been studied in depth. However, with theoretical and empirical studies, we show they provide powerful results for feature selection, via hypothesis testing and feature ranking. Combining them with some \soft" prior knowledge of the domain, we derive two novel algorithms (Semi-JMI, Semi-IAMB) that outperform significantly more complex competing methods, showing particularly good performance when the labels are missing-not-at-random. We conclude that simple approaches to this problem can work surprisingly well, and in many situations we can provably recover the exact feature selection dynamics, as if we had labelled the entire dataset.

Searching for extraterrestrial, transient signals in astronomical data sets is an active area of current research. However, machine learning techniques are lacking in the literature concerning single-pulse detection. This paper presents a new, two-stage approach for identifying and classifying dispersed pulse groups (DPGs) in single-pulse search output. The first stage identified DPGs and extracted features to characterize them using a new peak identification algorithm which tracks sloping tendencies around local maxima in plots of signal-to-noise ratio vs. dispersion measure. The second stage used supervised machine learning to classify DPGs. We created four benchmark data sets: one unbalanced and three balanced versions using three different imbalance treatments.We empirically evaluated 48 classifiers by training and testing binary and multiclass versions of six machine learning algorithms on each of the four benchmark versions. While each classifier had advantages and disadvantages, all classifiers with imbalance treatments had higher recall values than those with unbalanced data, regardless of the machine learning algorithm used. Based on the benchmarking results, we selected a subset of classifiers to classify the full, unlabelled data set of over 1.5 million DPGs identified in 42,405 observations made by the Green Bank Telescope. Overall, the classifiers using a multiclass ensemble tree learner in combination with two oversampling imbalance treatments were the most efficient; they identified additional known pulsars not in the benchmark data set and provided six potential discoveries, with significantly less false positives than the other classifiers., 13 pages, accepted for publication in MNRAS, ref. MN-15-1713-MJ.R3

Data on cytomegalovirus infection (CMV) prevalence and course in hospitalized infants are rather scarce, obsolete and considerably inconsistent. AIM: to determine the prevalence, rate of clinical manifestations, risk factors and predictive capacity of clinical manifestations of CMV infection in hospitalized infants during their first year of life. PATIENTS AND METHODS: All 163 infants hospitalized in the Pediatric Ward for Nonrespiratory Pathology in a tertiary hospital were serologically screened for cytomegalovirus infection for 10 months. In infants up to 6 months old that were CMV IgG (+) and CMV IgM (-) we followed up the CMV IgG concentration or compared it with that of their mothers. RESULTS: The CMV prevalence for the entire study sample was 33.1 ± 3.7% (54 seropositive out of 163 examined infants); in newborns it was 19.4 ± 6.7% (7 of 36), in infants aged 1-3 months - 23.8 ± 5.4% (15 of 63), in 4-6-month olds - 28.1 ± 8.1% (9 of 32), and in 7-12-month old - 71.9 ± 8.1% (23 of 32). The rates of clinically apparent infections in the respective groups was 33.3 ± 6.5%, 57.01 ± 20.2%, 53.3 ± 13.3%, 33.3 ± 16.6%, and 13.0 ± 7.17%. The overall rate of clinically apparent CMV infection in all 163 children was between 11.0 ± 2.5% and 17.2 ± 2.9%. The probability of CMV infection increased with age and duration of breastfeeding. Hepatitis, cerebral vasculopathy and pneumonia (alone or combined) turned out to be predictors of CMV infection, but none of these symptoms had a frequency greater than 22%. CONCLUSIONS: We found a high rate of cytomegalovirus infections in hospitalized infants less than one year of age. This infection is the reason why at least 10% of the newborns and 12% of the children aged 1 to 3 months were hospitalised. The course was clinically apparent in over half of the infected children of up to 3 months of age.

This paper is an attempt to raise pertinent questions and act as platform to generate fruitful discussions within the AKBC community about the need for a large scale dataset for relation extraction. For proper training and evaluation of relation extraction tasks, the weaknesses of datasets used so far need to be tackled: mainly the size (too small) and the amount of data that is actually labelled (unlabelled data leading to recall problems). We have the vision of building a new large and fully labelled dataset for entity pairs connected via binary relations from both Freebase as well as other datasets, such as Clueweb. Concerning the process of building, we present pioneering work on a roadmap which will serve as the foundation for the intended discussion within the community. Points to discuss arise within the following steps: first, the source data has to be preprocessed in order to ensure that the set of relations consists of valid relations only; second, we suggest a method to find the most relevant relations for an entity pair; and third, we outline approaches on how to actually label the data. It is necessary to discuss several key issues in the process of generating this dataset. This will enable us to thoroughly create a dataset that will have the potential to serve as a standard to the community.

Unsupervised methods for learning distributed representations of words are ubiquitous in today's NLP research, but far less is known about the best ways to learn distributed phrase or sentence representations from unlabelled data. This paper is a systematic comparison of models that learn such representations. We find that the optimal approach depends critically on the intended application. Deeper, more complex models are preferable for representations to be used in supervised systems, but shallow log-linear models work best for building representation spaces that can be decoded with simple spatial distance metrics. We also propose two new unsupervised representation-learning objectives designed to optimise the trade-off between training time, domain portability and performance.

An open challenge in constructing dialogue systems is developing methods for automatically learning dialogue strategies from large amounts of unlabelled data. Recent work has proposed Next-Utterance-Classification (NUC) as a surrogate task for building dialogue systems from text data. In this paper we investigate the performance of humans on this task to validate the relevance of NUC as a method of evaluation. Our results show three main findings: (1) humans are able to correctly classify responses at a rate much better than chance, thus confirming that the task is feasible, (2) human performance levels vary across task domains (we consider 3 datasets) and expertise levels (novice vs experts), thus showing that a range of performance is possible on this type of task, (3) automated dialogue systems built using state-of-the-art machine learning methods have similar performance to the human novices, but worse than the experts, thus confirming the utility of this class of tasks for driving further research in automated dialogue systems., Comment: Accepted to SIGDIAL 2016 (short paper). 5 pages

This paper introduces a method to build a classifier based on labelled and unlabelled data. We set up the Expectation-Maximization (EM) algorithm steps for the particular case of the naive Bayes approach and show empirical work for the restricted web page database. Original contributions includes the application of the EM algorithm to simulated data in order to see the behavior of the algorithm for different numbers of labelled and unlabelled data, and to study the effect of the sampling mechanism for the unlabelled data on the results.

Semi-supervised learning (SSL) is useful when few labeled and plenty of unlabeled examples are available. This occurs in most of the cases due to labeled instances be difficult, expensive and time consuming to be obtained since human experts are required for the labeling task (Chapelle, Schlkopf, & Zien, 2010). The training set contain labeled data represented by X L = {( x 1 , y 1 ) ... ( x 1 , y 1 )}, and unlabelled data represented as X U = { x l+1 ... x l+u }. The total amount of training data is X = l ∪ u . The set of labeled examples is associated with the labels L Y = { y 1 , ..., y l } where y i ε {1,..., c } and c is the number of classes. The purpose of SSL is to infer the missing labels Y U = { y l+1 ,..., y n } corresponding to the unlabelled set X U .

A major challenge in consumer credit risk portfolio management is to classify households according to their risk profile. In order to build such risk profiles it is necessary to employ an approach that analyses data systematically in order to detect important relationships, interactions, dependencies and associations amongst the available continuous and categorical variables altogether and accurately generate profiles of most interesting household segments according to their credit risk. The objective of this work is to employ a knowledge discovery from database process to identify groups of indebted households and describe their profiles using a database collected by the Consumer Credit Counselling Service (CCCS) in the UK. Employing a framework that allows the usage of both categorical and continuous data altogether to find hidden structures in unlabelled data it was established the ideal number of clusters and such clusters were described in order to identify the households who exhibit a high propensity of excessive debt levels., Comment: Annals of Data Science, 2 (1), pp. 43-59, 2015

BioASQ is a series of challenges that aims to assess the performance of information systems in supporting two tasks that are central to the biomedical question answering process: a the indexing of large volumes of unlabelled data, primarily scientific articles, with biomedical concepts, b the processing of biomedical questions and the generation of answers and supporting material. In this paper, the main results of the first two BioASQ challenges are presented.