Chronically, this is Hanlin’s second piece of work done during his internship at IBM last summer. The work tries, for the first time, to combine coreset and quantization to jointly reduce two dimensions of the data cube: the number of data points, and the number of bits in representing each feature of each data point. The result is much better accuracy in using the reduced data to train ML models or much smaller data for achieving the same accuracy, compared to reducing the raw data by using coreset construction alone or quantization alone. Congratulations, Hanlin!