- Approximate computing: An effective likelihood-free method with statistical guarantees
- Project Year:
2018
- REU Student (s):
Ryan Gross | Rutgers University-New Brunswick NJ
- Student 1 Institution:
Rutgers University-New Brunswick
- Project Mentor:
Minge Xie
- Project Mentor Area:
Statistics and Biostatistics
- Project Abstract:
Approximate computing is a field of statistical inference techniques that can be used to produce estimates given complex sets of data. Approximate Bayesian Computing (ABC) and Approximate Confidence Distribution Computing (ACC) are two such methods that do not require the specification of a likelihood function, and hence can be used to estimate posterior distributions of parameters for simulation-based models. We look to apply these methods to a large data set, namely one containing pedestrian entrance and exit data for Madison Square Park in New York City. Given the technical and financial restraints of the counting procedure, much of the data is either missing or otherwise prone to error. Therefore, we propose several models to characterize the pedestrian traffic flow throughout the park. Then, simulation techniques are applied to replicate the missing data and produce distributions of simulated data. This leads to the application of ABC and ACC methods, which are used to ultimately produce accurate estimates of the total number of park users over a given time period.
