I am broadly interested in algorithms, game theory, and online optimization. Currently, my focus is on applications in power systems and neuroscience.
For the list of publications, check here or my Google Scholar profile.
Effective integration of renewable energy
Slowing global warming is a common and imminent task for mankind. In recent years, we have witnessed a strong trend of using renewable energy to reduce emissions. However, lots of research on renewable integration is overly optimistic about future clean policies and does not recognize the issues associated with blindly increasing the usage of renewable sources. My research addresses such challenges and develops mathematically robust frameworks to have healthy and stable renewable growth.
So far, we have developed alternative dispatching algorithms that hugely reduce the long-term generation cost while maintaining a healthy grid resilience [link]. Currently, we are working on incorporating risk management into power generation so the regional transmission organizations are able to create power portfolio as desired. In particular, we want to formally define various levels of systemic risks and construct a coherent risk measure for generation portfolios.
Fairness & incentives in electricity rate structure design
Currently, energy consumers split the grid overhead cost by paying a share that is proportionate to their net energy consumption, instead of gross consumption. We showed that this structure does not incentivize energy prosumers (i.e. consumers who also produce power, usually solar panel owners) to generate more power than their demand. To encourage prosumers to produce as much energy as possible and promote a fairer cost-sharing scheme, our team proposed a rate structure that requires every consumer to pay an amount proportionate to their gross demand, usually private information [link].
The above problem is our initial motivation for the information-eliciting game with partial signals, where each player has an external value x (observable) and an internal value y >= x (not observable). The player can report any number between x and y with the goal of minimizing her expected cost throughout all rounds. Our goal, as the mechanism designer, is to design a payment scheme to induce truthful reporting. In our upcoming paper [link], we propose a penalty mechanism such that a penalty is charged whenever an inconsistent report is observed. We analyze the strategies and equilibria for multiple agents and prove that truthful or near-truthful behavior could be achieved under certain natural distributions.
Bayesian statistical methods for vocal syntax
An important field of neuroscience is to study the mechanism of human vocal communication. With the absence of high-quality data on humans, mouse vocalization experiments are conducted to obtain valuable insights into human vocal behavior. Such data sets usually consist of categorical sound syllables with continuous inter-syllable intervals (ISIs). The latter is of particular importance as increased ISIs could indicate possible vocal impairment. However, there have been few statistical methods for analyzing both the syllable transitions and the length of ISIs.
We propose a novel class of Bayesian renewal mixed models for analyzing the transition probabilities and ISIs via Dirichlet and gamma mixtures, respectively, allowing the mixture probabilities in both cases to vary flexibly with fixed covariate effects as well as random individual-specific effects [link]. We apply our model to the FoxP2 data set to evaluate the influence on vocal behavior of a mutation of the FoxP2 gene. The R package for the proposed Bayesian renewal models is published on CRAN [link].
