OBJECTIVE: While non-invasive, cuffless blood pressure (BP) measurement has demonstrated relevancy in controlled environments, ambulatory measurement is important for hypertension diagnosis and control. We present both in-lab and ambulatory BP estimation results from a diverse cohort of participants. METHODS: Participants (N=1125, aged 21-85, 49.2% female, multiple hypertensive categories) had BP measured in-lab over a 24-hour period with a subset also receiving ambulatory measurements. Radial tonometry, photoplethysmography (PPG), electrocardiography (ECG), and accelerometry signals were collected simultaneously with auscultatory or oscillometric references for systolic (SBP) and diastolic blood pressure (DBP). Predictive models to estimate BP using a variety of sensor-based feature groups were evaluated against challenging baselines. RESULTS: Despite limited availability, tonometry-derived features showed superior performance compared to other feature groups and baselines, yieldingprediction errors of 0.32 ±9.8 mmHg SBP and 0.54 ±7.7 mmHg DBP in-lab, and 0.86 ±8.7 mmHg SBP and 0.75 ±5.9 mmHg DBP for 24-hour averages. SBP error standard deviation (SD) was reduced in normotensive (in-lab: 8.1 mmHg, 24-hr: 7.2 mmHg) and younger (in-lab: 7.8 mmHg, 24-hr: 6.7 mmHg) subpopulations. SBP SD was further reduced 15-20% when constrained to the calibration posture alone. CONCLUSION: Performance for normotensive and younger participants was superior to the general population across all feature groups. Reference type, posture relative to calibration, and controlled vs. ambulatory setting all impacted BP errors. SIGNIFICANCE: Results highlight the need for demographically diverse populations and challenging evaluation settings for BP estimation studies. We present the first public dataset of ambulatory tonometry and cuffless BP over a 24-hour period to aid in future cardiovascular research.

Igor Labutov, Sumit Basu, Lucy Vanderwende. Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2015.

In comparison to multiple choice or other recognition-oriented forms of assessment, short answer questions have been shown to offer greater value for both students and teachers; for students they can improve retention of knowledge, while for teachers they provide more insight into student understanding. Unfortunately, the same open-ended nature which makes them so valuable also makes them more difficult to grade at scale. To address this, we propose a cluster-based interface that allows teachers to read, grade, and provide feedback on large groups of answers at once. We evaluated this interface against an unclustered baseline in a within-subjects study with 25 teachers, and found that the clustered interface allows teachers to grade substantially faster, to give more feedback to students, and to develop a high-level view of students' understanding and misconceptions.

Given a classification task, what is the best way to teach the resulting boundary to a human? While machine learning techniques can provide excellent methods for finding the boundary, including the selection of examples in an online setting, they tell us little about how we would teach a human the same task. We propose to investigate the problem of example selection and presentation in the context of teaching humans, and explore a variety of mechanisms in the interests of finding what may work best. In particular, we begin with the baseline of random presentation and then examine combinations of several mechanisms: the indication of an example’s relative difficulty, the use of the shaping heuristic from the cognitive science literature (moving from easier examples to harder ones), and a novel kernel-based “coverage model” of the subject’s mastery of the task. From our experiments on 54 human subjects learning and performing a pair of synthetic classification tasks via our teaching system, we found that we can achieve the greatest gains with a combination of shaping and the coverage model.

We introduce a new approach to the machine-assisted grading of short answer questions. We follow past work in automated grading by first training a similarity metric between student responses, but then go on to use this metric to group responses into clusters and subclusters. The resulting groupings allow teachers to grade multiple responses with a single action, provide rich feedback to groups of similar answers, and discover modalities of misunderstanding among students; we refer to this amplification of grader effort as “powergrading.” We develop the means to further reduce teacher effort by automatically performing actions when an answer key is available. We show results in terms of grading progress with a small “budget” of human actions, both from our method and an LDA-based approach, on a test corpus of 10 questions answered by 698 respondents.

Real-time recognition may be limited by scarce memory and computing resources for performing classification. Although, prior research has addressed the problem of training classifiers with limited data and computation, few efforts have tackled the problem of memory constraints on recognition. We explore methods that can guide the allocation of limited storage resources for classifying streaming data so as to maximize discriminatory power. We focus on computation of the expected value of information with nearest neighbor classifiers for online face recognition. Experiments on real-world datasets show the effectiveness and power of the approach. The methods provide a principled approach to vision under bounded resources, and have immediate application to enhancing recognition capabilities in consumer devices with limited memory.

An important way to make large training sets is to gather noisy labels from crowds of nonexperts. We propose a minimax entropy principle to improve the quality of these labels. Our method assumes that labels are generated by a probability distribution over workers, items, and labels. By maximizing the entropy of this distribution, the method naturally infers item confusability and worker expertise. We infer the ground truth by minimizing the entropy of this distribution, which we show minimizes the Kullback-Leibler (KL) divergence between the probability distribution and the unknown truth. We show that a simple coordinate descent scheme can optimize minimax entropy. Empirically, our results are substantially better than previously published methods for the same problem.

Interactive clustering refers to situations in which a human labeler is willing to assist a learning algorithm in automatically clustering items. We present a related but somewhat different task, assisted clustering, in which a user creates explicit groups of items from a large set and wants suggestions on what items to add to each group. While the traditional approach to interactive clustering has been to use metric learning to induce a distance metric, our situation seems equally amenable to classification. Using clusterings of documents from human subjects, we found that one or the other method proved to be superior for a given cluster, but not uniformly so. We thus developed a hybrid mechanism for combining the metric learner and the classifier. We present results from a large number of trials based on human clusterings, in which we show that our combination scheme matches and often exceeds the performance of a method which exclusively uses either type of learner.

We present data-driven methods for supporting musical creativity by capturing the statistics of a musical database. Specifically, we introduce a system that supports users in exploring the high-dimensional space of musical chord sequences by parameterizing the variation among chord sequences in popular music. We provide a novel user interface that exposes these learned parameters as control axes, and we propose two automatic approaches for defining these axes. One approach is based on a novel clustering procedure, the other on principal components analysis. A user study compares our approaches for defining control axes both to each other and to an approach based on manually-assigned genre labels. Results show that our automatic methods for defining control axes provide a subjectively better user experience than axes based on manual genre labeling.

Selection tasks are common in modern computer interfaces: we are often required to select a set of files, emails, data entries, and the like. File and data browsers have sorting and block selection facilities to make these tasks easier, but for complex selections there is little to aid the user without writing complex search queries. We propose an interactive machine learning solution to this problem called "smart selection," in which the user selects and deselects items as inputs to a selection classifier which attempts at each step to correctly generalize to the user's target state. Furthermore, we take advantage of our data on how users perform selection tasks over many sessions, and use it to train a label regressor that models their generalization behavior: we call this process learning to generalize. We then combine the user's explicit labels as well the label regressor outputs in the selection classifier to predict the user's desired selections. We show that the selection classifier alone takes dramatically fewer mouse clicks than the standard file browser, and when used in conjunction with the label regressor, the predictions of the classifier are significantly more accurate with respect to the target selection state.

We introduce MySong, a system that automatically chooses chords to accompany a vocal melody. A user with no musical experience can create a song with instrumental accompaniment just by singing into a microphone, and can experiment with different styles and chord patterns using interactions designed to be intuitive to non-musicians.

In this work, we present a novel means of browsing recorded audio conversations. The method we develop produces scalable summaries of the recognized speech, in which we can increase the amount of text continuously with the desired level of detail to best fill the available space. We present an interface in which a user can view an entire conversation in one screen, but can also quickly zoom in to see the full transcript; the corresponding audio can be easily played as well. The scaling is achieved via a combination of topic segmentation and informative phrase selection, where the threshold for informativeness decreases with increasing level of detail. Finally, we evaluate our method and interface against a baseline interface with a user study.

We present a method for performing acoustic echo cancellation in a channel with rapidly varying gain and thus a rapidly varying channel characteristic. This is a situation in which standard AEC approaches perform poorly. Our method involves learning a scale-free channel characteristic (Htilde). We then apply this to a windowed version of the signal and remove the projection of the transformed signal from the output signal. We also develop a "ramp projection" method that allows for a linear variation in gain within the window. We show results in a telephony application with 3 dB to more than 8 dB of improvement over conventional AEC using the simple projection and an additional 1 dB using the ramp projection

In this paper, we present a speaker identification algorithm for a microphone array based on a first-order joint hidden Markov model (HMM) where the observations correspond to the angle of arrival of the speech and the speech spectrum. The goal of the research is to investigate whether including angle of arrival information improves the speaker identification error rates compared to an algorithm based on the speech spectrum only. The spectral model consists of a Gaussian mixture model (GMM) using multiple discriminant analysis (MDA) coefficients and the angle model includes a separate histogram for each participant. The convergence time of the joint HMM is improved by estimating the GMM for each of the meeting participants prior to the start of the meeting and initializing each participant's spectral GMM in the joint HMM to the pretrained parameter values. The performance of the algorithm is analyzed from data collected during live meetings recorded using an eight element, circular microphone array. For meetings where the participants are stationary, the results show significant improvement over a single channel speaker ID algorithms based on spectrum only

The paper presents a novel representation for auditory environments that can be used for classifying events of interest, such as speech, cars, etc., and potentially used to classify the environments themselves. We propose a novel discriminative framework that is based on the audio epitome, an audio extension of the image representation developed by N. Jojic et al. (see Proc. Int. Conf. Comp. Vision, 2003). We also develop an informative patch sampling procedure to train the epitomes. This procedure reduces the computational complexity and increases the quality of the epitome. For classification, the training data is used to learn distributions over the epitomes to model the different classes; the distributions for new inputs are then compared to these models. On a task of distinguishing between 4 auditory classes in the context of environmental sounds (car, speech, birds, utensils), our method outperforms the conventional approaches of nearest neighbor and mixture of Gaussians on three out of the four classes.

Researchers have noted conflicting trends in collaboration technologies between delivering more information on larger displays and exploiting mobility on smaller devices. Large, shared displays provide greater choice in the presentation of information, but mobile devices offer greater flexibility in the access of information. We describe a platform that leverages the best of both worlds by allowing multiple users to access and interact with a large, shared display using their own personal mobile devices, such as a cell phone, laptop, or wireless PDA. We highlight three applications built on top of the platform that demonstrate its generality and utility in a variety of group settings: namely, web browsing, polling, and entertainment.

influences In this work, we quantitatively investigate the ways in which a given person the joint turn-taking behavior in a conversation. After collecting an auditory database of social interactions among a group of twenty-three people via wearable sensors (66 hours of data each over two weeks), we apply speech and conversation detection methods to the auditory streams. These methods automatically locate the conversations, determine their participants, and mark which participant was speaking when. We then model the joint turn-taking behavior as a Mixed-Memory Markov Model [1] that combines the statistics of the individual subjects' self-transitions and the partners ' cross-transitions. The mixture parameters in this model describe how much each person's individual behavior contributes to the joint turn-taking behavior of the pair. By estimating these parameters, we thus estimate how much influence each participant has in determining the joint turn(cid:173) this measure correlates taking behavior. We significantly with betweenness centrality [2], an independent measure of an individual's importance in a social network. This result suggests that our estimate of conversational influence is predictive of social influence.

We present a novel method for simultaneous voicing and speech detection based on a linked-HMM architecture, with robust features that are independent of the signal energy. Because this approach models the change in dynamics between speech and nonspeech regions, it is robust to low sampling rates, significant levels of additive noise, and large distances from the microphone. We demonstrate the performance of our method in a variety of testing conditions and also compare it to other methods reported in the literature.

Intelligent agents need to learn how the communication structure evolves within interacting groups and how to influence the groups overall behavior. We are developing methods to automatically and unobtrusively learn the social network structure that arises within a human group based on wearable sensors. Computational models of group interaction dynamics are derived from data gathered using wearable sensors. The questions we are exploring are: Can we tell who influences whom? Can we quantify this amount of influence? How can we modify group interactions to promote better information diffusion? The goal is real-time learning and modification of social network relationships by applying statistical machine learning techniques to data obtained from unobtrusive wearable sensors.

We describe a method for enhancing auditory awareness by selectively passing speech sounds in the environment to the user. We develop a robust far-field speech detection algorithm for noisy environments and a source localization algorithm for flexible arrays. We then combine these methods to give a user control over the spatial regions from which speech will be passed through. Using this technique, we have implemented a "smart headphones" system in which a user can be listening to music over headphones and hear speech from specified directions mixed in. We show our preliminary results on the algorithms and describe initial user feedback about the system.

We address the problem of tracking and reconstructing 3D human lip motions from a 2D view. This problem is challenging due both to the complex nature of lip motions and the minimal data available from a raw video stream of the face. We counter both of these difficulties with statistical approaches. We first build a physically-based 3D model of lips and train it to cover only the subspace of lip motions. We then track this model in video by finding the shape within the subspace that maximizes the posterior probability of the model given the observed features. In this study, the features are the likelihoods of the lip and non-lip color classes: we iteratively derive forces from these values to apply to the physical model and converge to the final solution. Because of the full 3D nature of the model, this framework allows us to track the lips from any head pose. In addition, because of the constraints imposed by the learned subspace of the model, we are able to accurately estimate the full 3D lip shape from the 2D view.

We describe tools that use measurements from video for the extraction of facial modeling and animation parameters, head tracking, and real time interactive facial animation. These tools share common goals but rely on varying details of physical and geometric modeling and in their input measurement system. Accurate facial modeling involves fine details of geometry and muscle coarticulation. By coupling pixel by pixel measurements of surface motion to a physically based face model and a muscle control model, we have been able to obtain detailed spatio temporal records of both the displacement of each point on the facial surface and the muscle control required to produce the observed facial motion. We discuss the importance of this visually extracted representation in terms of realistic facial motion synthesis. A similar method that uses an ellipsoidal model of the head coupled with detailed estimates of visual motion allows accurate tracking of head motion in 3D. Additionally, by coupling sparse, fast visual measurements with our physically based model via an interpolation process, we have produced a real time interactive facial animation/mimicking system.

This paper describes a method for the robust tracking of rigid head motion from video. This method uses a 3D ellipsoidal model of the head and interprets the optical flow in terms of the possible rigid motions of the model. This method is robust to large angular and translational motions of the head and is not subject to the singularities of a 2D model. The method has been successfully applied to heads with a variety of shapes, hair styles, etc. This method also has the advantage of accurately capturing the 3D motion parameters of the head. This accuracy is shown through comparison with a ground truth synthetic sequence (a rendered 3D animation of a model head). In addition, the ellipsoidal model is robust to small variations in the initial fit, enabling the automation of the model initialization. Lastly, due to its consideration of the entire 3D aspect of the head, the tracking is very stable over a large number of frames. This robustness extends even to sequences with very low frame rates and noisy camera images.