Jayadev Acharya receives Best Paper Award at ALT 2020

Jayadev Acharya, assistant professor of electrical and computer engineering, received the Best Paper Award at the 31st International Conference on Algorithmic Learning Theory (ALT 2020), held in San Diego, California earlier this month. The paper was co-authored by Ananda Theertha Suresh, a research scientist at Google.

Their paper, “Optimal multiclass overfitting by sequence reconstruction from Hamming queries,” provides a resolution to the problem of overfitting in machine learning. 

“Often machine learning models are designed and tested on benchmark datasets,” Acharya explained. “Repeatedly improving the model with the same dataset can lead to overfitting, where the model fits too well with the dataset, and fails to perform well with new data. This leads to the question of how valuable a dataset is, and how many times can it be used.”

Researchers have been trying to determine the extent to which a dataset for multi-class classification (for example, classifying an image into one of 100 different categories) can be overfit.  Acharya and Suresh resolve this by designing new overfitting attacks that try to extract as much information from the dataset as possible. 

 “ALT is one of the top conferences studying the theoretical aspects of machine learning, and we were very honored to receive this award,” Acharya said. 

Other Articles of Interest