Vigneshwaran Subbaraju

The case for smartwatch-based diet monitoring

We explore the use of gesture recognition on a wrist-worn smartwatch as an enabler of an automated eating activity (and diet monitoring) system. We show, using small-scale user studies, how it is possible to use the accelerometer and gyroscope data from a smartwatch to accurately separate eating episodes from similar non-eating activities, and to additionally identify the mode of eating (i.e., using a spoon, bare hands or chopsticks). Additionally, we investigate the likelihood of automatically triggering the smartwatchs camera to capture clear images of the food being consumed, for possible offline analysis to identify what (and how much) the user is eating. Our results show both the promise and challenges of this vision: while opportune moments for capturing such useful images almost always exist in an eating episode, significant further work is needed to both (a) correctly identify the appropriate instant when the camera should be triggered and (b) reliably identify the type of food via automated analyses of such images.

Accommodating user diversity for in-store shopping behavior recognition

This paper explores the possibility of using mobile sensing data to detect certain in-store shopping intentions or behaviours of shoppers. We propose a person-independent activity recognition technique called CROSDAC, which captures the diversity in the manifestation of such intentions or behaviours in a heterogeneous set of users in a data-driven manner via a 2-stage clustering-cum-classification technique. Using smartphone based sensor data (accelerometer, compass and Wi-Fi) from a directed, but real-life study involving 86 shopping episodes from 30 users in a malls food court, we show that CROSDACs mobile sensing-based approach can offer reasonably high accuracy (77:6% for a 2-class identification problem) and outperforms the traditional community-driven approaches that unquestioningly segment users on the basis of underlying demographic or lifestyle attributes.

Accurate detection of autism spectrum disorder from structural MRI using extended metacognitive radial basis function network

A study on ASD detection in females using MRI is presented using the ABIDE dataset.VBM is used to study differences in gray matter composition.Different regions within the motor cortex area are affected for female ASD patients.Age-specific approach improves accuracy (by 2.5% for adolescents, 15% for adults).Proposed EMcRBFN classifier achieves higher accuracy than SVM and PBL-McRBFN. In this paper, we present an accurate detection of Autism Spectrum Disorder (ASD) from structural MRI using an Extended Metacognitive Radial Basis Function Neural Classifier (EMcRBFN). An automatic whole brain Voxel Based Morphometry (VBM) approach is used to identify gray matter composition in the brain from structural Magnetic Resonance Imaging (MRI) and an improved q-Gaussian classifier and its metacognitive learning algorithm has been proposed to approximate the functional relationship between the high dimensional VBM features and the true class labels. Recent genetic studies indicate that ASD manifests in different ways between males and females and also between adolescents and adults. Accordingly, the proposed EMcRBFN classifier has been evaluated using the publicly available Autism Brain Imaging Data Exchange dataset with a comprehensive study on both males and females and also between adolescents and adults in both categories. EMcRBFN classifier performance is compared with currently existing results for ASD classification in the literature and also with well known standard classifiers. The results clearly indicate that the performance of the EMcRBFN classifier is better than that of the other classifiers considered in this study. Further, the comprehensive study also indicates that the following subregions in the brain viz., premotor cortex and supplementary motor cortex are affected for adult-females while the somatosensory cortex subregion is affected for adolescent-females with ASD. Similar results indicate that the precentral gyrus, motor cortex, medial frontal gyrus and the paracentral lobule areas are affected for adolescent males while the superior frontal gyrus and the frontal eye fields areas are affected for adult males with ASD.

Identifying differences in brain activities and an accurate detection of autism spectrum disorder using resting state functional-magnetic resonance imaging : A spatial filtering approach

&NA; This paper presents a new approach for detecting major differences in brain activities between Autism Spectrum Disorder (ASD) patients and neurotypical subjects using the resting state fMRI. Further the method also extracts discriminative features for an accurate diagnosis of ASD. The proposed approach determines a spatial filter that projects the covariance matrices of the Blood Oxygen Level Dependent (BOLD) time‐series signals from both the ASD patients and neurotypical subjects in orthogonal directions such that they are highly separable. The inverse of this filter also provides a spatial pattern map within the brain that highlights those regions responsible for the distinguishable activities between the ASD patients and neurotypical subjects. For a better classification, highly discriminative log‐variance features providing the maximum separation between the two classes are extracted from the projected BOLD time‐series data. A detailed study has been carried out using the publicly available data from the Autism Brain Imaging Data Exchange (ABIDE) consortium for the different gender and age‐groups. The study results indicate that for all the above categories, the regional differences in resting state activities are more commonly found in the right hemisphere compared to the left hemisphere of the brain. Among males, a clear shift in activities to the prefrontal cortex is observed for ASD patients while other parts of the brain show diminished activities compared to neurotypical subjects. Among females, such a clear shift is not evident; however, several regions, especially in the posterior and medial portions of the brain show diminished activities due to ASD. Finally, the classification performance obtained using the log‐variance features is found to be better when compared to earlier studies in the literature. HighlightsIdentify neural activity differences using fMRI and an accurate diagnosis of ASD.Spatial filtering for discriminative features and spatial map of activity differences.Distinguishable activities more prevalent in the right hemisphere for ASD patients.For male ASD patients, shift in resting state activities to prefrontal cortex.Better classification performance on ABIDE data set using discriminative features. Graphical abstract Figure. No caption available.

Identification of lateralized compensatory neural activities within the social brain due to autism spectrum disorder in adolescent males

Socio‐behavioral impairments are important characteristics of autism spectrum disorders (ASD) and MRI‐based studies are pursued to identify a neurobiological basis behind these conditions. This paper presents an MRI‐based study undertaken to (i) identify the differences in brain activities due to ASD, (ii) verify whether such differences exist within the ‘social brain’ circuit which is hypothesized to be responsible for social functions, and (iii) uncover potential compensatory mechanisms within the identified differences in brain activities. In this study, a whole‐brain voxel‐wise analysis is performed using resting‐state fMRI data from 598 adolescent males, that is openly available from the ABIDE consortium. A new method is developed, which can (i) extract the discriminative brain activities, that provide high separability between the blood oxygenation time‐series signals from ASD and neurotypical populations, (ii) select the activities that are relevant to ASD by evaluating the correlation between the separability and traditional severity scores, and (iii) map the spatial pattern of regions responsible for generating the discriminative activities. The results show that the most discriminative brain activities occur within a subset of the social brain that is involved with affective aspects of social processing, thereby supporting the idea of the social brain and also its fractionalization in ASD. Further, it has also been found that the diminished activities in the posterior cingulate area are potentially compensated by enhanced activities in the ventromedial prefrontal and anterior temporal areas within the social brain. Hemispherical lateralization is also observed on such compensatory activities.

Experiences in Building a Real-World Eating Recogniser

In this paper, we describe the progressive design of the gesture recognition module of an automated food journaling system -- Annapurna. Annapurna runs on a smartwatch and utilises data from the inertial sensors to first identify eating gestures, and then captures food images which are presented to the user in the form of a food journal. We detail the lessons we learnt from multiple in-the-wild studies, and show how eating recognizer is refined to tackle challenges such as (i) high gestural diversity, and (ii) non-eating activities with similar gestural signatures. Annapurna is finally robust (identifying eating across a wide diversity in food content, eating styles and environments) and accurate (false-positive and false-negative rates of 6.5% and 3.3% respectively)

Inferring smartphone keypress via smartwatch inertial sensing

Due to numerous benefits, sensor-rich smartwatches and wrist-worn wearable devices are quickly gaining popularity. The popularity of these devices also raises privacy concerns. In this paper we explore one such privacy concern: the possibility of extracting the location of a users touch-event on a smartphone, using the inertial sensor data of a smartwatch worn by the user on the same arm. This is a major concern not only because it might be possible for an attacker to extract private and sensitive information from the inputs provided but also because the attack mode utilises a device (smartwatch) that is distinct from the device being attacked (smartphone). Through a user study we find that such attacks are possible. Specifically, we can infer the users entry pattern on a qwerty keyboard, with an error bound of ±2 neighboring keys, with 73.85% accuracy. As a possible preventive mechanism, we also show that adding a little white noise to inertial sensor data can reduce the inference accuracy by almost 30%, without affecting the accuracy of macro-gesture recognition.

MiPAL: Multiple-instance passive aggressive learning for identification of attention deficit hyperactive disorder from fMRI

This paper proposes a new algorithm for the multiple instance learning problem (MIL) and investigates its application for detecting Attention Deficit Hyperactive Disorder (ADHD) from resting-state functional Magnetic Resonance Imaging data. The core component of many kernel-based MIL algorithms is usually an SVM-like batch optimization framework, hence scaling to large datasets like fMRI is often difficult. On the other hand, a family of on-line kernel classification algorithms widely known as “perceptron-like” kernel classifiers demonstrate efficient and accurate solutions. This paper presents MiPAL — Multiple-instance Passive Aggressive Learning algorithm, based on such a perceptron-like kernel classifier. First, MiPAL builds a labeller by inputting negative bags into PA algorithm. Second, this labeller helps to train a separate PA classifier to predict binary class labels such that least-negative instances are regarded as positive. Due to the on-line PA algorithms fast adaptation, the impact of invalid positive support-vectors could be attenuated by the new, accurate support-set over time. Our experimental results reveal performance gains in several MIL datasets including state-of-the-art performance in Muskl, Fox, and comparable accuracy in the preprocessed ADHD-200 dataset.

Discovering anomalous events from urban informatics data

Singapore’s “smart city” agenda is driving the government to provide public access to a broader variety of urban informatics sources, such as images from traffic cameras and information about buses servicing different bus stops. Such informatics data serves as probes of evolving conditions at different spatiotemporal scales. This paper explores how such multi-modal informatics data can be used to establish the normal operating conditions at different city locations, and then apply appropriate outlier-based analysis techniques to identify anomalous events at these selected locations. We will introduce the overall architecture of sociophysical analytics, where such infrastructural data sources can be combined with social media analytics to not only detect such anomalous events, but also localize and explain them. Using the annual Formula-1 race as our candidate event, we demonstrate a key difference between the discriminative capabilities of different sensing modes: while social media streams provide discriminative signals during or prior to the occurrence of such an event, urban informatics data can often reveal patterns that have higher persistence, including before and after the event. In particular, we shall demonstrate how combining data from (i) publicly available Tweets, (ii) crowd levels aboard buses, and (iii) traffic cameras can help identify the Formula-1 driven anomalies, across different spatiotemporal boundaries.

Measurement-driven performance analysis of indoor femtocellular networks

This paper describes initial empirical studies, performed on a 6-node 3G indoor femtocellular testbed, that investigate the impact of pedestrian mobility on network parameters, such as handoff behavior and data throughput. The studies establish that, owing to the small radii of cells, even modest changes in movement speed can have disproportionately large impact on handoff patterns and network throughput. By also revealing a strong temporal dependency effect, the studies motivate the need for algorithms to accurately predict RF signal strength distributions in dynamic indoor environments. We present such an RF prediction algorithm, based on crowd-sourced signal strength readings, and show that the algorithm can predict RF signal strengths with an average estimation error of 3 dBm.