August 2017

NSF Funds Cross-Cutting CEAS Collaboration to Optimize Cloud Computing for Real-Life Applications

Researchers in the Departments of Computer Science, and Applied Mathematics and Statistics awarded $449k through the National Science Foundation’s NeTS program.

Anshul Gandhi, left, and Zhenhua Liu

Even for the personal smartphone or home computer user there is no avoiding the use of cloud computing. Cloud computing is low in cost, easily available, and offers access to useful services that would otherwise be out of reach. Services such as Netflix, Amazon Fire, and Expedia are only some of the popular online services being hosted on the cloud. On the backend, dynamic applications in the cloud are more lucrative if their deployments grow through dynamic capacity provisioning. Software deployments must be carefully provisioned to meet their performance requirements without wasting resources.

Most resource provisioning solutions today employ predictions to estimate demand and provide resources, accordingly. At times, this process could be fraught with errors. With the support of a National Science Foundation (NSF) Networking Technology and Systems (NeTS) award of $449k researchers Anshul Gandhi and Zhenhua Liu, seek to bridge the gap between predictors and provisioning solutions.

The goal of their NeTS project, Demystifying the Role of Prediction Models: Bridging Prediction Algorithms and Resource Provisioning, is to develop and leverage error models to fully realize the potential of predictors. According to Gandhi, “Our research will allow businesses to maximize resource utilization despite prediction errors.”

“This is the kind of cross-cutting research we encourage among our faculty, to advance technologies that push the boundaries and challenge traditional thinking,” said Fotis Sotiropoulos, Dean of the College of Engineering and Applied Sciences. “I applaud Professors Gandhi and Liu on their collaborative approach to this research, and congratulate them on this recognition from the NSF. I look forward to following the progress of their research.”

Gandhi and Liu will investigate the prediction error model which includes constructing models that capture the structure of prediction errors; developing an algorithmic framework; and designing systems to exploit the new prediction error-aware algorithms.

Joseph Mitchell, chair of the Department of Applied Mathematics and Statistics, said, “This is a compelling project that requires collective expertise from computer science, optimization, probability, and statistics, and represents an ideal collaboration between Computer Science and Applied Mathematics and Statistics, both in research and in educational impact.”

Gandhi and Liu will realize additional benefits of their research through technology transfer opportunities with industrial partners.

NSF identified this NeTS project as “transformative research” related to fundamental scientific and technological advances in networking as well as systems research. NSF funds projects such as these in the hope that it will lead to “the development of future-generation, high-performance networks and future internet architectures”.

Arie E. Kaufman, chair of the Department of Computer Science, said, “In addition to finding the answer to network performance challenges, this project is especially interesting because it will directly contribute to interdisciplinary courses taught in two major departments on campus.”

About the Researchers
Anshul Gandhi is an assistant professor in the Department of Computer Science and he is affiliated with the Department of Applied Mathematics and Statistics, and Stony Brook University’s Smart Energy Technologies Cluster. He earned his PhD in computer science from Carnegie Mellon University, where he was advised by Prof. Mor Harchol-Balter. Prior to joining Stony Brook, he spent a year as a post-doctoral researcher in the Cloud Optimization and Analytics group at the IBM T.J. Watson Research Center.

Assistant Professor Zhenhua Liu is currently based in the Department of Applied Mathematics and Statistics, and he is affiliated with Department of Computer Science, and Smart Energy Technology Cluster. He was recently on leave for the ITRI-Rosenfeld Fellowship in the Energy and Environmental Technology Division at Lawrence Berkeley National Laboratory during the year 2014-2015. Dr. Liu earned his PhD in computer science at the California Institute of Technology.

Gandhi and Liu also recently received another NSF award, titled Enhancing the Parasol Experimental Testbed for Sustainable Computing, as part of an infrastructure grant led by Rutgers University to study sustainable computing in datacenters, which is aligned with the Smart Energy Technology cluster’s objectives.

Both the Department of Computer Science and the Department of Applied Mathematics and Statistics are part of the College of Engineering and Applied Sciences at Stony Brook University.

–Christine Cesaria

Risks to Stony Brook's Research Infrastructure from Proposed F&A Rate Reduction

Dear Faculty Colleagues,

I write today concerning a disturbing proposal that is being discussed in Washington that could have serious repercussions for the funding that we receive from federal agencies in support of research that is vital to the well-being of our citizens and our nation.

Recent discussions within the White House Office of Management and Budget (OMB) and the U.S. Department of Health and Human Services (HHS) have suggested that the federal science budget could be dramatically reduced by slashing facilities and administrative (F&A) cost reimbursements to universities. This proposal - which some in Congress may endorse - should concern all of us.

Federal research funding for universities includes two major components. First, grants or contracts awarded to universities (almost always through a competitive process) contain direct costs attributed to individual projects. These include items such as salary support, research staff and students, supplies, equipment, travel, and publication costs.

The other component of a grant or contract is the facilities and administrative cost rate (F&A or ‘indirect’ costs). These costs cannot be assigned to a single project because they include items such as laboratory space, heat, lights, IT infrastructure, animal care facilities, hazardous waste disposal, power, insurance, and the support staff required to manage grants and to ensure compliance with a myriad of federal and state regulations (human subjects protection, export controls, conflict of interest, etc). In negotiating these rates, Stony Brook, like other research universities, includes only those resources actually used to support research. Independent analyses have demonstrated repeatedly that the federal government only partially reimburses universities, including Stony Brook, for these expenses, many of which have been invested prior to the awarding of the grant. These are also costs that Stony Brook would not incur if we were not a research-intensive university.

I am well aware that faculty often question the F&A rate, how it is calculated, and who does the negotiation. Stony Brook’s current on-campus F&A rate is 59.5%, which falls close to the median rate of research institutions.  This rate is set through a comprehensive process guided by strict OMB rules, called Uniform Guidance 2 CFR 200. Our staff produce and file an extensive report in which they calculate our actual F&A expenditures based on prior years and apportioned to research, instruction, or other. Our actual rate is then negotiated with staff from the U.S. Department of Health and Human Services Division of Cost Allocation Services, and set based upon a comprehensive review and assessment of these costs. Compliance with the rate and OMB and agencies’ rules regarding these F&A reimbursements is then audited by the federal government every year under the terms established in the Single Audit Act. Importantly, many direct cost items are excluded from the base used to determine the final F&A reimbursement level (tuition, equipment, major renovations or repairs, and subcontract awards).

It is important to understand that Stony Brook is already subsidizing the actual F&A costs that federal research grants incur. This is due to two factors. The first is that since 1991, the federal government has imposed an administrative cap of 26% on the total F&A rate for administrative costs. At the same time, the cumulative number of regulatory changes relating to research with which universities must comply has dramatically increased. Ensuring compliance with these additional regulations—whether associated with human subject protections, animal care, export controls, effort reporting, conflict of interest, scientific fraud, and misconduct investigations—costs money and employment of additional administrative staff. The independent General Accounting Office (GAO) has estimated that research-intensive universities are already contributing about 25% of total dollars in support of faculty-led research projects.

Another question faculty often ask is why we accept funding from foundations that do not pay the federal F&A rate? There are several explanations. For some foundations, what would normally be considered an F&A expense may be charged as a direct cost since foundations are not required to use federal rules. In other instances, university funds are used to cover these costs. Finally, total research funding from private foundations amounts to a small percentage of our total research volume so the impact is less evident and often absorbed. Federal funds, however, account for more than 70% of our external funding. 

What is the risk to Stony Brook if the federal government drastically reduces F&A reimbursement for our research costs without any meaningful reduction in regulations and administrative compliance costs? Currently, our F&A reimbursement supports salaries for grants and contracts support staff, IRB and IACUC compliance specialists, IT specialists that support grant proposal submission and management, and technology transfer specialists, just to mention a few.  As a researcher, you are already aware that Stony Brook has undertaken a major assessment of administrative burdens that investigators shoulder during the conduct of research.  Initial findings call for investment in several of these key areas to better support faculty in obtaining and managing federal research funding. A significant reduction in F&A reimbursement to Stony Brook would not only preclude such investment, but would severely undermine existing research support services.  Negatively affected would be cost sharing, graduate student tuition subsidies, lab renovations, faculty start-up packages, benefits, and beyond. Almost every aspect of how our university supports research would be seriously impacted.

This is where all of us can help in protecting and sustaining the phenomenally successful 70 year university-federal government partnership for American science. While the Stony Brook leadership and our colleagues at other AAU institutions have been trying to educate all the relevant sectors about the importance of sustaining the partnership through robust funding of research and F&A reimbursement, we need the support of individual faculty researchers. It can be especially helpful for individual faculty members to understand the threat to Stony Brook’s research enterprise and to speak out to others to encourage their support.

With my best regards,
Richard J. Reeder
Vice President for Research

Computer Science PhD Alum Wins Best Dissertation Among Data Science Community 0

Alum Bryan Perozzi, now a research scientist at Google, won the Association of Computing Machinery SIGKDD, KDD 2017 Doctoral Dissertation award for his work at Stony Brook University. The annual award acknowledges excellent doctoral research in the field of data mining and knowledge discovery.

Bryan Perozzi and Steven Skiena

Perozzi thesis, Local Modeling of Attributed Graphs: Algorithms and Applications, was recognized as the best dissertation of the year in the data science community. His work involves graph embeddings — ways of representing the knowledge encoded in the structure of networks to make them accessible for machine learning models.

Focused on developing scalable algorithms and models for attributed graphs, Perozzi presented an online learning algorithm utilizing recent advances in deep learning to result in rich graph embeddings. The applications of this research are far reaching for the fields of data mining, information retrieval, profiling and demographic inference, online advertising and fraud detection.

Perozzi, whose advisor was Stony Brook Professor Steven Skiena, defended his thesis in May 2016. Upon learning of the award, he said, “Wow, what an honor! I’m humbled to have my work recognized by this prestigious early career award, and I am looking forward to giving a talk during the Doctoral Dissertation Award session on August 15.” The KDD 2017 Conference takes place in Nova Scotia, August 13-17.

Professor Skiena is especially proud of Perozzi’s research accomplishments, and they are collaborators on a number of published works. Their paper on DeepWalk graph embeddings has already been cited 270 times in Google Scholar since its publication in 2014.

“Bryan was a very creative, hardworking and independent graduate student here at Stony Brook, and his work on DeepWalk has proven extremely influential in the data science and machine learning communities. They got the right man for this award,” said Skiena.

At Google, Perozzi’s research relates to the intersection of data mining, machine learning, graph theory, and network science with a particular focus on local graph algorithms. In January 2017, he published and presented Ties that Bind: Characterizing Classes by Attributes and Social Ties, a collaboration with Stony Brook PhD student Aria Rezaei and Carnegie Melon faculty Leman Akoglu.

Perozzi is the first PhD student in the Department of Computer Science, which is part of the College of Engineering and Applied Sciences at Stony Brook University, to receive this award.

About the Association for Computing Machinery
Founded in 1947, the ACM is the largest and oldest scientific and industrial computing society. SIGKDD is the ACM’s Special Interest Group on Knowledge Discovery and Data Mining. SIGKDD selects one winner and two runner-ups each year to receive the award. Selections are based on the relevance to KDD, originality, scientific significance, technical depth and soundness, and overall presentation and readability.

Go to top