A research collaboration has found an efficient way to expand the collective behavior of swarming microrobots: Mixing different sizes of the micron-scale ‘bots enables them to self-organize into... Read more about Swarming microrobots self-organize into diverse patterns
Five ECE Faculty win Google Faculty Research Awards
Five ECE faculty have received Google Faculty Research Awards. The program provides unrestricted gifts to support research at institutions around the world, recognizing and supporting world-class faculty pursuing cutting-edge research in areas of mutual interest. Proposals are categorized into broad research topics. The goal of the Google Faculty Research Awards program is to identify and strengthen long-term collaborative relationships with faculty working on problems that will impact how future generations use technology.
The five award recipients from ECE are: Jayadev Acharya, assistant professor; Christina Delimitrou, assistant professor; Daniel Lee, professor, Cornell Tech; Mahsa Shoaran, assistant professor; Zhiru Zhang, associate professor.
Jayadev Acharya, Assistant Professor
Proposal: Private Heavy Hitters - Simplest and Optimal
Acharya’s proposal is part of the broader theme of developing new algorithms and understanding the fundamental limits for learning while ensuring privacy of the data samples. While user data helps design better systems, personal and potentially sensitive data (such as websites browsed by individuals) needs to remain private. One particular task that Acharya plans to track is heavy hitter detection, such as when Google receives various search requests from users (or spammers) all occurring at the same time. The proposal will support the development of new and simple algorithms for detecting heavy hitters and other learning tasks.
Christina Delimitrou, Assistant Professor
Proposal: Using machine learning to improve datacenter server management
Delimitrou’s proposal aims to address the increasing complexity in the hardware and software of warehouse-scale computers. Manually tuning the many parameters that modern server platforms expose is both ineffective and unscalable. This becomes even more challenging when cloud computing research happens in academic environments where access to production applications is non-trivial. The proposal includes three directions to improve the efficiency and performance predictability of cloud systems in a reproducible manner. First, Delimitrou proposes using practical machine learning techniques to determine the sensitivity of cloud services to mechanisms, such as power management and cache partitioning, offering a systematic way to reason about their impact on performance. Second, she proposes to apply a similar data-driven approach to diagnosing the root causes of unpredictable performance in the cloud. Finally, to improve the relevance and representativeness of academic research on cloud computing, she proposes constructing and releasing a benchmark suite, which although consisting of open-source services, convincingly resembles the architectural characteristics of real cloud applications.
Daniel Lee, Professor, Cornell Tech
Proposal: Joint Learning of Continuous Natural Language Control for Quadcopters in Simulation and Reality
In collaboration with Cornell Tech Assistant Professor Yoav Artzi, this project proposes learning a direct mapping of natural language instructions and raw observations for the continuous control of a quadrotor.
Mahsa Shoaran, Assistant Professor
Proposal: Toward Minimally-Invasive Brain Implants with Embedded Classification of Real-time Neural Data
Topic: Machine Learning and Data Mining
Shoaran’s proposal aims to employ state-of-the-art machine learning principles and efficient hardware architectures to enable real-time neural data processing on devices and prostheses for neurological diseases. She has previously designed a microchip for neural data classification, achieving a remarkable energy efficiency and high accuracy in epileptic seizure detection. In this project, her team will develop both algorithmic and hardware techniques to embed classifiers in applications such as closed-loop stimulation for Parkinson’s disease, migraine attack prediction, and motor decoding for brain-machine interfaces. She has proposed several ideas to improve the scalability, energy, and area efficiency of the embedded classifier, toward building minimally-invasive and high-performance brain implants in future.
Zhiru Zhang, Associate Professor
Proposal: Automatic Synthesis for Programmable Hardware Specialization
Prof. Zhang proposes to develop a new compilation framework that can automatically synthesize a high-quality programmable hardware accelerator from instruction set specifications. For proof of concept, his team has developed a preliminary flow that produces a complete in-order pipelined RISC-V processor, with competitive QoR against widely used manual designs. If successful, this project would significantly reduce both hardware and software design complexity to enable wider and more rapid adoption of the specialized computing paradigm.