Theophilus A. Benson And George Konidaris Earn Salomon Awards
- Posted by Jesse Polhemus
- on March 19, 2018
Click the links that follow for more news items about Theophilus A. Benson, George Konidaris, and Richard B. Salomon Faculty Awards.
Less than a year after Professor Theophilus (Theo) A. Benson's arrival at Brown University’s Computer Science Department (Brown CS), and less than two after Professor George Konidaris joined the Brown CS faculty, they've each received a Richard B. Salomon Faculty Research Award. A full list of awardees is available here. The award, given annually by Brown’s Office of the Vice-President for Research, was established to support excellence in scholarly work by providing funding for selected faculty research projects of exceptional merit with preference given to junior faculty who are in the process of building their research portfolio. They join multiple previous Brown CS winners, including Jeff Huang, Stefanie Tellex, Rodrigo Fonseca, Sherief Reda, and Ugur Cetintemel.
"Mobile devices," says Theo when asked to explain his research, "have become the primary mode of Internet access in both developed and developing countries. Yet, in developing regions, driven by their low cost, mobile devices are often equipped with small memory sizes and slow CPUs. For example, according to a recent study of mobile devices in Pakistan, 90% of devices were equipped with at most 1024MB of RAM, and 89% had 1GHz or slower processors. Despite the prevalence of such low-end devices in these markets, there are few systematic studies of the differences between web performance on low-end and high-end smartphones. Additionally, we lack principled techniques to analyze and improve performance on these low-end phones."
The goal of his proposal is to fill that vital gap:
- Using low-end smartphones prevalent in developing regions, they will conduct a large-scale measurement study to identify bottleneck resources (e.g., CPU, memory, and network) in the page load process and analyze how these bottlenecks may change over time based on device characteristics, network connectivity, server configuration and page structure. The insights from this study will help in understanding the effectiveness of various infrastructure design choices (e.g., the role of CDNs), operating system principles (e.g., Android One) and page load optimization techniques (e.g., Polaris) for developing regions.
- Developing a set of data-driven optimizations that enables, both large providers, e.g., Facebook, and small and local startups, to leverage their insights and improve end-user performance. In designing these optimizations, they will explore various architectural design choices to enable efficient continuous analysis of large data sets while providing fine-grained control over end-user connection characteristics.
George explains his research as follows: "Solar panel tracking improves the solar panel energy production by pointing them towards the sun throughout the day. Existing tracking algorithms compute the location of the sun in the sky via astronomical calculations, and move solar panels to match that angle. However, these calculations do not account for reflective and diffuse radiance or weather conditions that impact the efficiency of tracking algorithms. We reframe the problem of collecting energy from the sun as a contextual bandit problem, where the solar panel is controlled by a learning program attempting to maximize the 'reward' (i.e. energy) it collects. A small working group of graduate and undergraduate students at Brown has performed extensive simulated experiments in a variety of scenarios, indicating that in many locations on the Earth contextual bandit approaches outperform existing baselines. The group has also constructed a small, low-cost prototype of a single-axis solar tracking system, leading to preliminary results and publications accepted to RLDM, EnviroInfo, and IAAI conferences."
They propose three goals:
- construct a larger, more reliable dual-axis tracker to validate their approach over longer timescales
- develop advanced algorithmic solutions for contextual bandit problems, with sustainability as a motivating application and
- expand their computational sustainability working group within the AI/robotics research labs at Brown. They aim to place Brown at the forefront of computational sustainability research, combining our existing focus on socially-conscious robotics with computational efforts to ameliorate environmental problems facing modern society.
George and Theo will receive their awards at Brown's annual Celebration of Research on Thursday, April 19th, at the Faculty Club.
For more information, click the link that follows to contact Brown CS Communication Outreach Specialist Jesse C. Polhemus.