Researchers from the Programming Languages Lab recently presented their paper "Detecting Floating-Point Errors via Atomic Conditions" at POPL'20, which was held in USA this year. The paper proposed a new method for detecting large floating-point errors in numerical code. It outperforms the state-of-the-art in efficiency by several orders of magnitude. 

Note: POPL is short for "Principles of Programming Languages Symposium", and is one of the flagship conferences in programming languages.