Tong Yu

FootprintID: Indoor Pedestrian Identification through Ambient Structural Vibration Sensing

We present FootprintID, an indoor pedestrian identification system that utilizes footstep-induced structural vibration to infer pedestrian identities for enabling various smart building applications. Previous studies have explored other sensing methods, including vision-, RF-, mobile-, and acoustic-based methods. They often require specific sensing conditions, including line-of-sight, high sensor density, and carrying wearable devices. Vibration-based methods, on the other hand, provide easy-to-install sparse sensing and utilize gait to distinguish different individuals. However, the challenge for these methods is that the signals are sensitive to the gait variations caused by different walking speeds and the floor variations caused by structural heterogeneity. We present FootprintID, a vibration-based approach that achieves robust pedestrian identification. The system uses vibration sensors to detect footstep-induced vibrations. It then selects vibration signals and classifiers to accommodate sensing variations, taking step location and frequency into account. We utilize the physical insight on how individual step signal changes with walking speeds and introduce an iterative transductive learning algorithm (ITSVM) to achieve robust classification with limited labeled training data. When trained only on the average walking speed and tested on different walking speeds, FootprintID achieves up to 96% accuracy and a 3X improvement in extreme speeds compared to the Support Vector Machine. Furthermore, it achieves up to 90% accuracy (1.5X improvement) in uncontrolled experiments.

Hybridizing Personal and Impersonal Machine Learning Models for Activity Recognition on Mobile Devices

Recognition of human activities, using smart phones and wearable devices, has attracted much attention recently. The machine learning (ML) approach to human activity recognition can broadly be classified into two categories: training an ML model on (i) an impersonal dataset or (ii) a personal dataset. Previous research shows that models learned from personal datasets can provide better activity recognition accuracy compared to models trained on impersonal datasets. In this paper, we develop a hybrid incremental (HI) method with logistic regression models. This method uses incremental learning of logistic regression, to combine the advantages of both impersonal and personal approaches. We investigate two essential issues in this method, which are the selection of the learning rate schedule and the class imbalance problem. Our experiments show that the model learned using our HI method give better accuracy than the model learned from personal or impersonal data only. Besides, the techniques of adaptive learning rate and cost-sensitive learning give faster updates and more robust ML models in incremental learning. Our method also has potential benefits in the area of privacy preservation.

Towards continuous and passive authentication across mobile devices: an empirical study

Mobile devices, such as smartphones and tablets, have become prevalent given their ample functionality brought by a variety of applications. Unfortunately, these devices face security and privacy threats due to unauthorized access. Ordinary protection mechanisms such as passcode and fingerprint verification are widely employed to mitigate the threats. To achieve strong security without sacrificing usability, extensive research efforts have been devoted to continuous authentication through passive sensing and behavior modeling. Nowadays, more and more users own multiple devices. This trend presents opportunities for further optimization of authentication across devices. In this paper, we conduct an empirical study on how a behavioral model created on one device can be transferred to other devices to bootstrap continuous authentication. To pursue this goal, we collect 160 sets of usage data on multiple mobile devices and perform a proof-of-concept experiment. The results demonstrate that we can leverage the similarity between user behaviors on different devices to enable cross-device authentication and anomaly detection.

Thompson Sampling for Optimizing Stochastic Local Search

Stochastic local search (SLS), like many other stochastic optimization algorithms, has several parameters that need to be optimized in order for the algorithm to find high quality solutions within a short amount of time. In this paper, we formulate a stochastic local search bandit (\(\mathtt{SLSB}\)), which is a novel learning variant of SLS based on multi-armed bandits. \(\mathtt{SLSB}\) optimizes SLS over a sequence of stochastic optimization problems and achieves high average cumulative reward. In \(\mathtt{SLSB}\), we study how SLS can be optimized via low degree polynomials in its noise and restart parameters. To determine the coefficients of the polynomials, we present polynomial Thompson Sampling (\(\mathtt{PolyTS}\)). We derive a regret bound for \(\mathtt{PolyTS}\) and validate its performance on synthetic problems of varying difficulty as well as on feature selection problems. Compared to bandits with no assumptions of the reward function and other parameter optimization approaches, our \(\mathtt{PolyTS}\) assuming polynomial structure can provide substantially better parameter optimization for SLS. In addition, due to its simple model update, \(\mathtt{PolyTS}\) has low computational cost compared to other SLS parameter optimization methods.

XRec: Behavior-Based User Recognition Across Mobile Devices

As smartphones and tablets become increasingly prevalent, more customers have multiple devices. The multi-user, multi-device interactions inspire many problems worthy of investigation, among which recognizing users across devices has significant implications on recommendation, advertising and user experience. Unlike the binary classification problem in user identification on a single device, cross-device user recognition is essentially a set partition problem. The app back-end aims to divide user activities on devices hosting the app into groups each associated with one user. In this paper, we present XRec which leverages user behavioral patterns, namely when, where and how a user uses the app, to achieve the recognition. To address the user-device partition problem, we propose a classification-plus-refinement algorithm. To validate our approach, we conduct a field study with an Android app. We instrument the app to collect usage data from real users. We provide proof-of-concept experimental results to demonstrate how XRec can provide added value to mobile apps, with the ability to correctly match a user across multiple devices with 70% recall and 90% precision.

Mitigating Multi-tenant Interference in Continuous Mobile Offloading

Offloading computation to resource-rich servers is effective in improving application performance on resource constrained mobile devices. Despite a rich body of research on mobile offloading frameworks, most previous works are evaluated in a single-tenant setting, i.e., a server is assigned to a single client. In this paper we consider that multiple clients offload various continuous mobile sensing applications with end-to-end delay constraints, to a cluster of machines as the server. Contention for shared computing resources on a server can unfortunately result in delays and application malfunctions. We present a two-phase Plan-Schedule approach to mitigate multi-tenant resource contention, thus to reduce offloading delays. The planning phase predicts future workloads from all clients, estimates contention, and devises offloading schedule to remove or reduce contention. The scheduling phase dispatches arriving offloaded workloads to the server machine that minimizes contention, according to the running workloads on each machine. We implement the methods into ATOMS (Accurate Timing prediction and Offloading for Mobile Systems), a framework that adopts prediction of workload computing times, estimation of network delays, and mobile-server clock synchronization techniques. Using several mobile vision applications, we evaluate ATOMS under diverse configurations and prove its effectiveness.

Mitigating multi-tenant interference on mobile offloading servers: poster abstract

This work considers that multiple mobile clients offload various continuous sensing applications with end-to-end delay constraints, to a cluster of machines as the server. Contention for shared computing resources on a server can result in delay degradation and application malfunction. We present ATOMS (Accurate Timing prediction and Offloading for Mobile Systems), a framework to mitigate multi-tenant resource contention and to improve delay using a two-phase Plan-Schedule approach. The planning phase includes methods to predict future workloads from all clients, to estimate contention, and to devise offloading schedule to reduce contention. The scheduling phase dispatches arriving offloaded workload to the server machine that minimizes contention, based on the running workloads on each machine.

QoS-Aware Scheduling of Heterogeneous Servers for Inference in Deep Neural Networks

Deep neural networks (DNNs) are popular in diverse fields such as computer vision and natural language processing. DNN inference tasks are emerging as a service provided by cloud computing environments. However, cloud-hosted DNN inference faces new challenges in workload scheduling for the best Quality of Service (QoS), due to dependence on batch size, model complexity and resource allocation. This paper represents the QoS metric as a utility function of response delay and inference accuracy. We first propose a simple and effective heuristic approach that keeps low response delay and satisfies the requirement on processing throughput. Then we describe an advanced deep reinforcement learning (RL) approach that learns to schedule from experience. The RL scheduler is trained to maximize QoS, using a set of system statuses as the input to the RL policy model. Our approach performs scheduling actions only when there are free GPUs, thus reduces scheduling overhead over common RL schedulers that run at every continuous time step. We evaluate the schedulers on a simulation platform and demonstrate the advantages of RL over heuristics.