Department of Computing & Mathematical Sciences
California Institute of Technology
Yorie Nakahira is a PhD student at the Department of Computing and Mathematical Sciences, California Institute of Technology. She works with Professor John Doyle on optimization, control, and information theory. She collaborates with researchers in diverse elds and strives to provide theoretical insights into neuroscience and cell biology as well as scalable algorithms for the power and cloud systems. During her PhD program, she has also visited the University of Cambridge (3 months), the University of Minnesota (2 months), RIKEN Brain Science Institute (2 months), Nippon Telegraph and Telephone Corporation R&D (4 months), and Huawei Technologies R&D (3 months). She graduated from the Department of Control systems, Tokyo Institute of Technology.Control under Information Constraints
A common challenge underlying various control systems is to achieve robustness and eciency under information constraints. From the sensorimotor system to biomolecular systems to service systems, control decisions are generally determined under communication constraints and with only causal information. In this thesis, we develop tools to design control policies that account for information constraints by integrating optimization, control, scheduling, and information theory.
The rst part of this thesis studies control under communication constraints and its applications in human sensorimotor control and biomolecular control. We show closed-form formulas that characterize the impact of delay and data rate on the control performance and propose a linear program to design distributed controllers in the presence of time-varying delays, quantization, saturation, and sampling errors. When applied to sensorimotor control and biomolecular control, the theory explains the extreme heterogeneities exist in the human nervous system as well as analytical insights that complement intuitions obtained by simulating Chemical Master Equations.
The second part of this thesis studies online algorithms for deadline scheduling. In particular, we propose distributed algorithms that minimize service capacity variability when scheduling jobs with deadlines. Specically, we show that Exact Scheduling minimizes the variance of service capacity subject to strict demand and deadline requirements under stationary Poisson arrivals. We also consider more general settings with soft demand requirements, soft deadline requirements, or both, characterizing the optimal distributed policies for each setting. Additionally, we show how close the performance of optimal distributed policies is to that the optimal centralized policy by deriving competitive-ratio-like bounds.