Fast Support Vector Classification for Large-Scale Problems.

The support vector machine (SVM) is a very important machine learning algorithm with state-of-the-art performance on many classification problems. However, on large datasets it is very slow and requires much memory. To solve this defficiency, we propose the fast support vector classifier (FSVC) that includes: 1) an efficient closed-form training free of any numerical iterative procedure; 2) a small collection of class prototypes that avoids to store in memory an excessive number of support vectors; and 3) a fast method that selects the spread of the radial basis function kernel directly from data, without classifier execution nor iterative hyper-parameter tuning. The memory requirements of FSVC are very low, spending in average only 6 ·10-7 sec. per pattern, input and class, and processing datasets up to 31 millions of patterns, 30,000 inputs and 131 classes in less than 1.5 hours (less than 3 hours with only 2GB of RAM). In average, the FSVC is 10 times faster, requires 12 times less memory and achieves 4.7 percent more performance than Liblinear, that fails on the 4 largest datasets by lack of memory, being 100 times faster and achieving only 6.7 percent less performance than Libsvm. The time spent by FSVC only depends on the dataset size and thus it can be accurately estimated for new datasets, while Libsvm or Liblinear are much slower on "difficult" datasets, even if they are small. The FSVC adjusts its requirements to the available memory, classifying large datasets in computers with limited memory. Code for the proposed algorithm in the Octave scientific programming language is provided.1.

Walking Recognition in Mobile Devices.

Presently, smartphones are used more and more for purposes that have nothing to do withphone calls or simple data transfers. One example is the recognition of human activity, which isrelevant information for many applications in the domains of medical diagnosis, elderly assistance,indoor localization, and navigation. The information captured by the inertial sensors of the phone(accelerometer, gyroscope, and magnetometer) can be analyzed to determine the activity performedby the person who is carrying the device, in particular in the activity of walking. Nevertheless,the development of a standalone application able to detect the walking activity starting only fromthe data provided by these inertial sensors is a complex task. This complexity lies in the hardwaredisparity, noise on data, and mostly the many movements that the smartphone can experience andwhich have nothing to do with the physical displacement of the owner. In this work, we exploreand compare several approaches for identifying the walking activity. We categorize them into twomain groups: the first one uses features extracted from the inertial data, whereas the second oneanalyzes the characteristic shape of the time series made up of the sensors readings. Due to the lackof public datasets of inertial data from smartphones for the recognition of human activity underno constraints, we collected data from 77 different people who were not connected to this research.Using this dataset, which we published online, we performed an extensive experimental validationand comparison of our proposals.

Robust and Fast Scene Recognition in Robotics Through the Automatic Identification of Meaningful Images.

Scene recognition is still a very important topic in many fields, and that is definitely the case in robotics. Nevertheless, this task is view-dependent, which implies the existence of preferable directions when recognizing a particular scene. Both in human and computer vision-based classification, this actually often turns out to be biased. In our case, instead of trying to improve the generalization capability for different view directions, we have opted for the development of a system capable of filtering out noisy or meaningless images while, on the contrary, retaining those views from which is likely feasible that the correct identification of the scene can be made. Our proposal works with a heuristic metric based on the detection of key points in 3D meshes (Harris 3D). This metric is later used to build a model that combines a Minimum Spanning Tree and a Support Vector Machine (SVM). We have performed an extensive number of experiments through which we have addressed (a) the search for efficient visual descriptors, (b) the analysis of the extent to which our heuristic metric resembles the human criteria for relevance and, finally, (c) the experimental validation of our complete proposal. In the experiments, we have used both a public image database and images collected at our research center.

Direct Kernel Perceptron (DKP): ultra-fast kernel ELM-based classification with non-iterative closed-form weight calculation.

The Direct Kernel Perceptron (DKP) (Fernández-Delgado et al., 2010) is a very simple and fast kernel-based classifier, related to the Support Vector Machine (SVM) and to the Extreme Learning Machine (ELM) (Huang, Wang, & Lan, 2011), whose α-coefficients are calculated directly, without any iterative training, using an analytical closed-form expression which involves only the training patterns. The DKP, which is inspired by the Direct Parallel Perceptron, (Auer et al., 2008), uses a Gaussian kernel and a linear classifier (perceptron). The weight vector of this classifier in the feature space minimizes an error measure which combines the training error and the hyperplane margin, without any tunable regularization parameter. This weight vector can be translated, using a variable change, to the α-coefficients, and both are determined without iterative calculations. We calculate solutions using several error functions, achieving the best trade-off between accuracy and efficiency with the linear function. These solutions for the α coefficients can be considered alternatives to the ELM with a new physical meaning in terms of error and margin: in fact, the linear and quadratic DKP are special cases of the two-class ELM when the regularization parameter C takes the values C=0 and C=∞. The linear DKP is extremely efficient and much faster (over a vast collection of 42 benchmark and real-life data sets) than 12 very popular and accurate classifiers including SVM, Multi-Layer Perceptron, Adaboost, Random Forest and Bagging of RPART decision trees, Linear Discriminant Analysis, K-Nearest Neighbors, ELM, Probabilistic Neural Networks, Radial Basis Function neural networks and Generalized ART. Besides, despite its simplicity and extreme efficiency, DKP achieves higher accuracies than 7 out of 12 classifiers, exhibiting small differences with respect to the best ones (SVM, ELM, Adaboost and Random Forest), which are much slower. Thus, the DKP provides an easy and fast way to achieve classification accuracies which are not too far from the best one for a given problem. The C and Matlab code of DKP are freely available.

Addressing the flaws of current critical alarms: a fuzzy constraint satisfaction approach.

OBJECTIVES: Threshold alarms, the support supplied by commercial monitoring devices to supervise the signs that pathologies produce over physiological variables, generate a large amount of false positives, owing to the high number of artifacts in monitoring signals, and they are not capable of satisfactorily representing and identifying all monitoring criteria used by healthcare staff. The lack of an adequate support for monitoring the evolution of physical variables prevents the suitable exploitation of the information obtained when monitoring critical patients. This work proposes a solution for designing intelligent alarms capable of addressing the flaws and limitations of threshold alarms. MATERIALS AND METHODS: The solution proposed is based on the multivariable fuzzy temporal profile (MFTP) model, a formal model for describing certain monitoring criteria as a set of morphologies defined over the temporal evolution of the patient's physiological variables, and a set of relations between them. The MFTP model represents these morphologies through a network of fuzzy constraints between a set of points in the evolution of the variables which the physician considers especially relevant. We also provide a knowledge acquisition tool, TRACE, with which clinical staff can design and edit alarms based on the MFTP model. RESULTS: Sixteen alarms were designed using the MFTP model; these were capable of supervising monitoring criteria that could be satisfactorily supervised with commercial monitoring devices. The alarms were validated over a total of 196h of recordings of physiological variables from 78 different patients admitted to an intensive care unit. Of the 912 alarm triggerings, only 7% were false positives. A study of the usability of the tool TRACE was also carried out. After a brief training seminar, five physicians and four nurses designed a number of alarms with this tool. They were then asked to fill in the standard System Usability Scale test. The average score was 68.2. CONCLUSION: The proposal presented herein for describing monitoring criteria, comprising the MFTP model and TRACE, permits the supervision of monitoring criteria that cannot be represented by means of thresholds, and makes it possible to construct alarms that give a rate of false positives far below that for threshold alarms.

Cortical modulation of dorsal column nuclei: a computational study.

We present a computational study aimed at exploring the sensorimotor cortex modulation of the behaviour of dorsal column nuclei, specifically the impact of synaptic parameters, during both sleep and waking conditions. On the basis of the circuit proposed by Canedo et al. (2000), we have developed realistic computational models that have been tested with simultaneous electrocorticographic as well as intracellular cuneate recordings performed in anaesthetized cats. The results show that, (1) under sleep conditions, the model can block the transmission of afferent sensory information and, (2) operations expected during wakefulness, such as filtering and facilitation, can be performed if synaptic parameters are appropriately tuned.

A computational model of cuneothalamic projection neurons.

The dorsal column nuclei, cuneatus and gracilis, play a fundamental role in the processing and integration of somesthetic ascending information. Intracellular and patch-clamp recordings obtained in cat in vivo have shown that cuneothalamic projection neurons present two modes of activity: oscillatory and tonic (Canedo et al 1998 Neuroscience 84 603-17). The former is the basis of generating, in sleep and anaesthetized states, slow, delta and spindle rhythms under the control of the cerebral cortex (Mariño et al 2000 Neuroscience 95 657-73). The latter is needed, during wakefulness, to process somesthetic information in real time. To study this behaviour we have developed the first realistic computational model of the cuneothalamic projection neurons. The modelling was guided by experimental recordings, which suggest the existence of hyperpolarization-activated inward currents, transient low- and high-threshold calcium currents, and calcium-activated potassium currents. The neuronal responses were simulated during (1) sleep, (2) transition from sleep to wakefulness and (3) wakefulness under both excitatory and inhibitory synaptic input. In wakefulness the model predicts a set of synaptically driven firing modes that could be associated with information processing strategies in the middle cuneate nucleus.