Dr. Pavan Aduri, Professor of Computer Science, was awarded a $262,300 grant from the National Science Foundation for his research in studying new directions in algorithmic replicability. Aduri and his collaborators will address the problem of ensuring replicability and reproducibility in scientific research, particularly for studies that rely on randomized computations.
Challenges of Reproducibility
It is critical for scientific results and experiments be reproducibility. If research findings cannot be replicated, it means that other scientists cannot build upon them or verify their accuracy. This results in unreliable findings – which create confusion and slows down scientific progress. Ultimately, this can further the erosion of the public trust in science. The reproducibility issue has received attention from a wide spectrum of entities ranging from popular media, scientific publications as well as from the professional and scientific bodies.
Computations over massive data sets critically rely on randomized computations and randomness in computations may arise either due to the input data coming from an unknown distribution (such as in learning tasks) or from internal randomness used by the algorithms (such as in Monte Carlo simulations) or both. Inspite of their wide usage, randomized computations lack replicability.
Building a Solid Foundation for Data-Driven Scientific Progress
Aduri’s research will test new methods to achieve replicability of randomized computations. They will define what replicability means for different types of computations, develop ways to design randomized algorithms that can be easily reproduced and understand the limitations and costs associated with ensuring replicability.
Through this research, they will ensure that algorithms and models are replicable, which leads to more robust applications. Beyond data-intensive fields such as machine learning and big data analysis, the principles and techniques developed for replicability can influence other scientific and technological domains.