Inferring finer-grained human information with multi-modal cross-granularity learning

Linear regression methods are commonly used by both researchers and data scientists due to their interpretability and their reduced likelihood of overfitting. However, these methods can still perform poorly if little labeled training data is available. Typical methods used to overcome a lack of labeled training data somehow involve exploiting an outside source of labeled data or large amounts of unlabeled data. This includes areas such as active learning, semi-supervised learning and transfer learning, but in many domains these approaches are not always applicable because they require either a mechanism to label data, large amounts of unlabeled data or additional sources of sufficiently related data. In this paper we explore an alternative, non-data centric approach. We allow the user to guide the learning system through three forms of feature-level guidance which constrain the parameters of the regression function. Such guidance is unlikely to be perfectly accurate, so we derive methods which are robust to some amounts of noise, a property we formally prove for one of our methods.

Developing ladder network for intelligent evaluation system: Case of remaining useful life prediction for centrifugal pumps

Representation learning with deep sparse auto-encoder for multi-task learning

High spectral dimensionality and the shortage of annotations make hyperspectral image (HSI) classification a challenging problem. Recent studies suggest that convolutional neural networks can learn discriminative spatial features, which play a paramount role in HSI interpretation. However, most of these methods ignore the distinctive spectral-spatial characteristic of hyperspectral data. In addition, a large amount of unlabeled data remains an unexploited gold mine for efficient data use. Therefore, we proposed an integration of generative adversarial networks (GANs) and probabilistic graphical models for HSI classification. Specifically, we used a spectral-spatial generator and a discriminator to identify land cover categories of hyperspectral cubes. Moreover, to take advantage of a large amount of unlabeled data, we adopted a conditional random field to refine the preliminary classification results generated by GANs. Experimental results obtained using two commonly studied datasets demonstrate that the proposed framework achieved encouraging classification accuracy using a small number of data for training.

In the context of automated analysis of eye fundus images, it is an important common fallacy that prior works achieve very high scores in segmentation of lesions, and that fallacy is fueled by some reviews reporting very high scores, and perhaps some confusion with terms. A simple analysis of the detail of the few prior works that really do segmentation reveals scores between 7% and 70% in sensitivity for 1 FPI. That is clearly sub-par with medical doctors trained to detect signs of Diabetic Retinopathy, since they can distinguish well the contours of lesions in Eye Fundus Images (EFI). Still, a full segmentation of lesions could be an important step for both visualization and further automated analysis using rigorous quantification or areas and numbers of lesions to better diagnose. I discuss what prior work really does, using evidence-based analysis, and confront with segmentation networks, comparing on the terms used by prior work to show that the best performing segmentation network outperforms those prior works. I also compare architectures to understand how the network architecture influences the results. I conclude that, with the correct architecture and tuning, the semantic segmentation network improves up to 20 percentage points over prior work in the real task of segmentation of lesions. I also conclude that the network architecture and optimizations are important factors and that there are still important limitations in current work.

A semi-supervised social relationships inferred model based on mobile phone data

A longest repeat query on a string, motivated by its applications in many subfields including computational biology, asks for the longest repetitive substring(s) covering a particular string position (point query). In this paper, we extend the longest repeat query from point query to \emph{interval query}, allowing the search for longest repeat(s) covering any position interval, and thus significantly improve the usability of the solution. Our method for interval query takes a different approach using the insight from a recent work on \emph{shortest unique substrings} [1], as the prior work's approach for point query becomes infeasible in the setting of interval query. Using the critical insight from [1], we propose an indexing structure, which can be constructed in the optimal $O(n)$ time and space for a string of size $n$, such that any future interval query can be answered in $O(1)$ time. Further, our solution can find \emph{all} longest repeats covering any given interval using optimal $O(occ)$ time, where $occ$ is the number of longest repeats covering that given interval, whereas the prior $O(n)$-time and space work can find only one candidate for each point query. Experiments with real-world biological data show that our proposal is competitive with prior works, both time and space wise, while providing with the new functionality of interval queries as opposed to point queries provided by prior works.

It is a common scenario in industrial applications that though a large amount of monitoring data of mechanical machines are available, only a few of them are labeled due to the lack of expert knowledge and labor. This leads to the difficulty of developing powerful supervised fault diagnosis methods, which requires a relatively large fully-labeled dataset containing machine monitoring data collected under healthy and different faulty states. In terms of this issue, a novel few-label learning method for fault diagnosis is proposed in this work, which can first learn useful representations from a large amount of unlabeled data with the help of a contrastive learning technique, based on which a fault diagnosis model can be constructed with the support of only a few labeled data. To validate the effectiveness of The proposed method is applied to a benchmark bearing fault diagnosis dataset to validate its effectiveness in few-label scenarios. Results show that the proposed method obtains better accuracy than other state-of-the-art methods.

Trajectory point classification can be described as a supervised sequence labeling problem, in which a model is trained by labeling data to predict the category of unknown points and identify key events in the trajectory. Due to the difficulty of labeling trajectory point, a large amount of trajectory data is either unlabeled or labeled in an imbalanced way. To make matters worse, traditional trajectory point classification methods are generally constrained to utilize the statistical features of the labeled data and the semantic features as well as the large amount of unlabeled data have not been well studied yet. For this reason, the performance of traditional trajectory point classification methods is far from satisfactory. To solve this problem, we transfer existing language model knowledge to construct the semantic features and construct a trajectory point classification model by combining both the motion features and semantic features. The simulation results show that, compared with the traditional methods, our method has improved the accuracy of trajectory point classification by three and seven percentage points in the classification of circular and turning movements respectively.KeywordsTrajectory point classificationSequence labelingMotion features and semantic featuresLanguage model

An important challenge for asset management is to analyze large amounts of data in a short period of time to provide insightful information for decision making in a timely fashion. Analyzing all available data manually is impractical and inefficient. It is advantageous to develop pattern recognition algorithms to recognize events-of-interest to achieve effective asset management.Conventional pattern recognition algorithms usually require a fairly large training set in which data points are carefully prepared and "labeled". Examples include designating an equipment’s status as healthy or faulty by subject-matter experts. This can make the process time consuming and error-prone when the training set is large. For most applications, a small amount of data points are already labeled by the experts through their routine activities. While these data points are usually not enough to form a training set for conventional pattern recognition methods, some of the newer methods can take advantage of them along with the hidden manifold structures manifested by the unlabeled data. Moreover, subject matter experts may be willing to provide more input if a clear indication of the value of information and a manageable subset of data is pre-selected for their inspection. In fact, many other industries are facing the same challenge where the cost of acquiring labels is too expensive to be practical and large amounts of unlabeled data and limited expert time are available. A suite of advanced machine learning algorithms (e.g., semi-supervised learning, active learning) have been developed to tackle this challenge, and many of them have been successfully used for various applications in the past few years. In this paper, we will review the concepts and report our observations about the effectiveness of these methods in a real-world asset management scenario. We consider well test validation in an asset with a large number of tests as an example of a label-rich data set that can serve as the basis for our numerical review of existing methods. In this example we will specifically look at the task of building a statistical model to recognize the validity of rate measurement tests in a test separator. In this case, through their daily activities, the operators have labeled most of these tests as valid or invalid. The extensive amount of well test validation data provides sufficient information to assess the newer approaches under review. The plan then is to apply a similar approach to tasks such as equipment health monitoring to identify pump failures with limited expert input.Exxon Mobil Corporation has numerous subsidiaries, many with names that include ExxonMobil, Exxon, Esso and Mobil. For convenience and simplicity in this paper, the parent company and its subsidiaries may be referenced separately or collectively as "ExxonMobil." Abbreviated references describing global or regional operational organizations and global or regional business lines are also sometimes used for convenience and simplicity. Nothing in this paper is intended to override the corporate separateness of these separate legal entities. Working relationships discussed in this paper do not necessarily represent a reporting connection, but may reflect a functional guidance, stewardship, or service relationship.Conceptually, reduction in labeled input can be achieved by combining the information from the labels and the statistical distribution of the data (e.g., clusters). As an extreme example, consider that the pump measurement data may show two distinct clusters and the operators have labeled a few data points in one cluster as pump failures when reports had to be made due to wells being shut in. This information is sufficient to label one of the clusters as healthy and the other one as faulty. For a new measurement, a prediction may be made by first determining the cluster to which the measurement belongs and then assigning it the corresponding label. While most real world problems are much more challenging than this example due to the number of data points, dimensionality of the data, lack of clear cluster structure and potential ambiguity of data structures, similar ideas can be used to develop highly accurate statistical models with a limited number of labels.

Human Activity Recognition Using Ambient Sensor Data

Activity recognition is one of the preliminary steps in designing and implementing assistive services in smart homes. Such services help identify abnormality or automate events generated while occupants do, as well as intend to do, their desired activities of daily living (ADLs) inside a smart home environment. However, most of the existing systems are applied for single-resident homes. Multiple people living together create additional complexity in modeling numbers of overlapping and concurrent activities. In this paper, we introduce a hybrid mechanism between ontology-based and unsupervised machine learning strategies in creating activity models used for activity recognition in the context of multi-resident homes. On comparing with related data-driven approaches, the proposed technique is technically and practically scalable to real-world scenarios due to fast training time and easy implementation. An average activity recognition rate of 95.83% on CASAS Spring dataset was achieved and the run time per recognition operation was measured as 12.86 ms.

The logging of Activities of Daily Living (ADLs) is becoming increasingly popular mainly thanks to wearable devices. Currently, most sensors used for ADLs logging are queried and filtered mainly by location and time. However, in an Internet of Things future, a query will return a large amount of sensor data. Therefore, existing approaches will not be feasible because of resource constraints and performance issues. Hence more fine-grained queries will be necessary. We propose to filter on the likelihood that a sensor is relevant for the currently sensed activity. Our aim is to improve system efficiency by reducing the amount of data to query, store and process by identifying which sensors are relevant for different activities during the ADLs logging by relying on Distributional Semantics over public text corpora and unsupervised hierarchical clustering. We have evaluated our system over a public dataset for activity recognition and compared our clusters of sensors with the sensors involved in the logging of manually-annotated activities. Our results show an average precision of 89% and an overall accuracy of 69%, thus outperforming the state of the art by 5% and 32% respectively. To support the uptake of our approach and to allow replication of our experiments, a Web service has been developed and open sourced.

Recently, egocentric activity recognition has attracted considerable attention in the pattern recognition and artificial intelligence communities because of its widespread applicability to human systems, including the evaluation of dietary and physical activity and the monitoring of patients and older adults. In this paper, we present a knowledge-driven multisource fusion framework for the recognition of egocentric activities in daily living (ADL). This framework employs Dezert–Smarandache theory across three information sources: the wearer’s knowledge, images acquired by a wearable camera, and sensor data from wearable inertial measurement units and GPS. A simple likelihood table is designed to provide routine ADL information for each individual. A well-trained convolutional neural network is then used to produce a set of textual tags that, along with routine information and other sensor data, are used to recognize ADLs based on information theory-based statistics and a support vector machine. Our experiments show that the proposed method accurately recognizes 15 predefined ADL classes, including a variety of sedentary activities that have previously been difficult to recognize. When applied to real-life data recorded using a self-constructed wearable device, our method outperforms previous approaches, and an average accuracy of 85.4% is achieved for the 15 ADLs.

Fast and automated image quality assessment (IQA) for diffusion MR images is crucial so that a rescan decision can be made swiftly during or after the scanning session. However, learning this task is challenging as the number of annotated data is limited and the annotated label is not always perfect. To this end, we introduce an automatic multistage IQA method for pediatric diffusion MR images. Our IQA is performed in three sequential stages, i.e., slice-wise quality assessment (QA) using a nonlocal residual network, volume-wise QA by agglomerating the extracted features of slices belonging to one volume using a nonlocal network, and subject-wise QA by ensembling the QA results of volumes belonging to one subject. In addition, we employ semi-supervised learning to make full use of a small amount of annotated data and a large amount of unlabeled data to train our network. Specifically, we first pretrain our network using labeled data, which are iteratively expanded by labeling the unlabeled data with the trained network. Furthermore, we devise a self-training strategy which iteratively relabels and prunes the labeled dataset when training the network to deal with noisy labels. Experimental results demonstrate that our network, trained using only samples of modest size, exhibits great generalizability and is capable of conducting large-scale rapid IQA with near-perfect accuracy.