Since November 2018, I work as a research scientist at Lyft. Before that, I worked at Microsoft as a senior data scientist and at Amazon.com as a research scientist.
Until then I was a postdoctoral scholar at Department of Computer Science and Engineering of University of California, San Diego. I belonged to Microelectronic Embedded Systems Laboratory led by Prof. Rajesh Gupta.
I received my Ph.D. (2009) in Computer Science from UCSD and M.Eng (1999) and B.Eng (1997) in Mechano-Informatics from The University of Tokyo, Japan. Previously I have worked as a researcher at IBM Research - Tokyo.
My research interests were in theoretical aspects of sensor networks and mobile ad hoc networks.
As a postdoc, I worked on clock synchronization in sensor networks (Presented at EWSN'11). Although clock synchronization has been studied for a long time in sensor networks as well as in distributed systems, recent synchronization protocols for sensor networks have been mostly focused on achieving good average accuracy. While these protocols achieve microsecond-order accuracy, many of them do not have any guarantees on the worst case. However, as sensor networks become used more widely or incorporated into larger systems in the form of CPS (cyber-physical system), the notion of guaranteed performance increases importance much more than ever. To deal with this, our work focused on providing guaranteed accuracy in clock synchronization. Based on observation on the characteristics of crystal oscillators on sensor nodes, our synchronization protocol achieved much tighter accuracy guarantee than the classical methods (sometimes called "interval-based clock synchronization algorithms") do.
Another work during my postdoc was on localization (Presented at INFOCOM'11). When we cannot use GPS-based techniques (e.g., indoor environment), an alternative practical solution is to obtain distance information based on distance measurement between sensor nodes by using acoustic signals, signal strength, etc, and estimate the location of nodes. This is formulated as a problem often called "sensor localization problem." In sensor localization, there is an important subproblem of "localizability" of network or node, i.e., whether there exists a unique position for the whole network or a single node that satisfies all distance constraints. While rigidity theory plays an important role in identifying several localizability conditions, major limitations are that the results are only applicable to so-called generic frameworks and that the distance measurements need to be error-free. These limitations, in addition to the hardness of finding the node locations for a uniquely localizable graph, miss large portions of practical application scenarios that require sensor localization. To cope with this situation, we devised a novel SDP-based formulation for analyzing node localizability and providing a deterministic upper bound of localization error. Unlike other optimization-based formulations for solving localization problem for the whole network, our formulation allows fine-grained evaluation on the localization accuracy per each node. Our formulation gives a sufficient condition for unique node localizability for any frameworks, i.e., not only for generic frameworks. Furthermore, we extend it for the case with measurement errors and for computing directional error bounds. We also design an iterative algorithm for large-scale networks and demonstrate the effectiveness by simulation experiments.
In my PhD research, I have studied the use of controlled mobility for data collection applications in wireless sensor networks. Collecting data from spatially distributed sensors is a generic form of sensor network applications. Multihop forwarding approach has been broadly used for this purpose, but it can be inefficient in terms of energy consumption especially when the nodes are deployed sparsely. On the other hand, using a mobile device (often called a "data mule") to traverse the sensor field and collect data from each sensor is an alternative approach. This data mule approach significantly reduces energy consumption at each node. However, a downside of this approach is an increased data delivery latency, and thus we should optimize the motion of data mule so that we can minimize the latency. To optimize the motion of data mule, we have devised a novel formulation of the problem, which we call the Data Mule Scheduling (DMS) problem. One of the major benefits of the DMS problem framework is that it can handle various assumptions on the mobility and communication capabilities of data mule in a very structured way. We have analyzed the computational complexity of the DMS problem for different mobility cases and designed exact/approximation/heuristic algorithms. We have also considered an extension of the DMS problem for the hybrid case of multihop forwarding and data mule.