Ran Gilad-Bachrach

Full body gait analysis with Kinect

Human gait is an important indicator of health, with applications ranging from diagnosis, monitoring, and rehabilitation. In practice, the use of gait analysis has been limited. Existing gait analysis systems are either expensive, intrusive, or require well-controlled environments such as a clinic or a laboratory. We present an accurate gait analysis system that is economical and non-intrusive. Our system is based on the Kinect sensor and thus can extract comprehensive gait information from all parts of the body. Beyond standard stride information, we also measure arm kinematics, demonstrating the wide range of parameters that can be extracted. We further improve over existing work by using information from the entire body to more accurately measure stride intervals. Our system requires no markers or battery-powered sensors, and instead relies on a single, inexpensive commodity 3D sensor with a large preexisting install base. We suggest that the proposed technique can be used for continuous gait tracking at home.

On composition of a federated web search result page: using online users to provide pairwise preference for heterogeneous verticals

Modern web search engines are federated --- a user query is sent to the numerous specialized search engines called verticals like web (text documents), News, Image, Video, etc. and the results returned by these engines are then aggregated and composed into a search result page (SERP) and presented to the user. For a specific query, multiple verticals could be relevant, which makes the placement of these vertical results within blocks of textual web results challenging: how do we represent, assess, and compare the relevance of these heterogeneous entities?n In this paper we present a machine-learning framework for SERP composition in the presence of multiple relevant verticals. First, instead of using the traditional label generation method of human judgment guidelines and trained judges, we use a randomized online auditioning system that allows us to evaluate triples of the form query, web block, vertical>. We use a pairwise click preference to evaluate whether the web block or the vertical block had a better users engagement. Next, we use a hinged feature vector that contains features from the web block to create a common reference frame and augment it with features representing the specific vertical judged by the user. A gradient boosted decision tree is then learned from the training data. For the final composition of the SERP, we place a vertical result at a slot if the score is higher than a computed threshold. The thresholds are algorithmically determined to guarantee specific coverage for verticals at each slot.n We use correlation of clicks as our offline metric and show that click-preference target has a better correlation than human judgments based models. Furthermore, on online tests for News and Image verticals we show higher user engagement for both head and tail queries.

Speeding up the Xbox recommender system using a euclidean transformation for inner-product spaces

A prominent approach in collaborative filtering based recommender systems is using dimensionality reduction (matrix factorization) techniques to map users and items into low-dimensional vectors. In such systems, a higher inner product between a user vector and an item vector indicates that the item better suits the users preference. Traditionally, retrieving the most suitable items is done by scoring and sorting all items. Real world online recommender systems must adhere to strict response-time constraints, so when the number of items is large, scoring all items is intractable.n We propose a novel order preserving transformation, mapping the maximum inner product search problem to Euclidean space nearest neighbor search problem. Utilizing this transformation, we study the efficiency of several (approximate) nearest neighbor data structures. Our final solution is based on a novel use of the PCA-Tree data structure in which results are augmented using paths one hamming distance away from the query (neighborhood boosting). The end result is a system which allows approximate matches (items with relatively high inner product, but not necessarily the highest one). We evaluate our techniques on two large-scale recommendation datasets, Xbox Movies and Yahoo~Music, and show that this technique allows trading off a slight degradation in the recommendation quality for a significant improvement in the retrieval time.

Manual for Using Homomorphic Encryption for Bioinformatics

Biological data science is an emerging field facing multiple challenges for hosting, sharing, computing on, and interacting with large data sets. Privacy regulations and concerns about the risks of leaking sensitive personal health and genomic data add another layer of complexity to the problem. Recent advances in cryptography over the last five years have yielded a tool, homomorphic encryption, which can be used to encrypt data in such a way that storage can be outsourced to an untrusted cloud, and the data can be computed on in a meaningful way in encrypted form, without access to decryption keys. This paper introduces homomorphic encryption to the bioinformatics community, and presents an informal “manual” for using the Simple Encrypted Arithmetic Library (SEAL), which we have made publicly available for bioinformatic, genomic, and other research purposes.

Predicting "About-to-Eat" Moments for Just-in-Time Eating Intervention

Various wearable sensors capturing body vibration, jaw movement, hand gesture, etc., have shown promise in detecting when one is currently eating. However, based on existing literature and user surveys conducted in this study, we argue that a Just-in-Time eating intervention, triggered upon detecting a current eating event, is sub-optimal. An eating intervention triggered at About-to-Eat moments could provide users with a further opportunity to adopt a better and healthier eating behavior. In this work, we present a wearable sensing framework that predicts About-to-Eat moments and the Time until the Next Eating Event. The wearable sensing framework consists of an array of sensors that capture physical activity, location, heart rate, electrodermal activity, skin temperature and caloric expenditure. Using signal processing and machine learning on this raw multimodal sensor stream, we train an About-to-Eat moment classifier that reaches an average recall of 77%. The Time until the Next Eating Event regression model attains a correlation coefficient of 0.49. Personalization further increases the performance of both of the models to an average recall of 85% and correlation coefficient of 0.65. The contributions of this paper include user surveys related to this problem, the design of a system to predict about to eat moments and a regression model used to train multimodal sensory data in real time for potential eating interventions for the user.

Designing Social and Emotional Skills Training: The Challenges and Opportunities for Technology Support

Social and emotional skills are crucial for all aspects of our everyday life. However, understanding how digital technology can facilitate the development and learning of such skills is yet an under-researched area in HCI. To start addressing this gap, this paper reports on a series of interviews and design workshops with the leading researchers and developers of Social and Emotional Learning (SEL) curricula. SEL is a subfield of educational psychology with a long history of teaching such skills, and a range of evidence based curricula that are widely deployed in primary and secondary schools. We identify the shared challenges across existing curricula that digital technology might help address: the support for out-of-session learning, scaffolding for parental engagement, and feedback for the curricula developers. We argue how this presents an opportunity for mutually beneficial collaborations, with the potential for significant real-world impact of novel HCI systems, and can inform HCI work on supporting social and emotional skills development in other domains.

Challenges for designing notifications for affective computing systems

Affective computing systems blend technical and social elements and present challenges for designing notifications. As the number of applications expand that utilize notifications, we need to consider that competing for attention is a concern. With respect to promoting positive health behaviors, notifications need to be adapted to times when a user is vulnerable and receptive for a health intervention but also to users attentional states. In this paper we outline affective computing projects and in the workshop will discuss challenges of designing interventions to promote positive behaviors in real world environments.

Scaffolding the scaffolding: Supporting children's social-emotional learning at home

The development of strong social and emotional skills is central to personal wellbeing. Increasingly, these skills are being taught in schools through well researched curricula. Such social-emotional learning (SEL) curricula are most effective if reinforced by parents, thus transferring the skills into everyday contexts. Traditional SEL programs have however had limited success in engaging parents, and we argue that technology might be able to help bridge this school-home divide. Through interviews with SEL experts we identified central design considerations for technology and SEL content: the reliance on experiential learning and the need to scaffold the parents in scaffolding the interaction for their children. This informed the design of a technology probe comprising a magnet card and online SEL activities, deployed in a school and via Mturk. The results provide a nuanced understanding of how technology-based interventions could bridge the school-home gap in real-world settings and support at-home reinforcement of childrens social-emotional skills.

Learning from mistakes: towards a correctable learning algorithm

Many learning algorithms generate complex models that are difficult for a human to interpret, debug, and extend. In this paper, we address this challenge by proposing a new learning paradigm called correctable learning, where the learning algorithm receives external feedback about which data examples are incorrectly learned. We define a set of metrics which measure the correctability of a learning algorithm. We then propose a simple and efficient correctable learning algorithm which learns local models for different regions of the data space. Given an incorrect example, our method samples data in the neighborhood of that example and learns a new, more correct local model over that region. Experiments over multiple classification and ranking datasets show that our correctable learning algorithm offers significant improvements over the state-of-the-art techniques.

Logistic regression over encrypted data from fully homomorphic encryption

BackgroundOne of the tasks in the 2017 iDASH secure genome analysis competition was to enable training of logistic regression models over encrypted genomic data. More precisely, given a list of approximately 1500 patient records, each with 18 binary features containing information on specific mutations, the idea was for the data holder to encrypt the records using homomorphic encryption, and send them to an untrusted cloud for storage. The cloud could then homomorphically apply a training algorithm on the encrypted data to obtain an encrypted logistic regression model, which can be sent to the data holder for decryption. In this way, the data holder could successfully outsource the training process without revealing either her sensitive data, or the trained model, to the cloud.MethodsOur solution to this problem has several novelties: we use a multi-bit plaintext space in fully homomorphic encryption together with fixed point number encoding; we combine bootstrapping in fully homomorphic encryption with a scaling operation in fixed point arithmetic; we use a minimax polynomial approximation to the sigmoid function and the 1-bit gradient descent method to reduce the plaintext growth in the training process.ResultsOur algorithm for training over encrypted data takes 0.4–3.2 hours per iteration of gradient descent.ConclusionsWe demonstrate the feasibility but high computational cost of training over encrypted data. On the other hand, our method can guarantee the highest level of data privacy in critical applications.