How to represent and compress sparse feature for feature engineering?

Question

There is a number denoted as A (uint_16 type), and the value of A represents a particular meaning, and the number has 400+ possible values(each one is unique).

A sample has to use a feature f to represent A , but A could use multiple values for a sample, eg, A = {0x10, 0x2FC0, ...}

How to represent this feature? One naive approach is hot-encoding, but it's high dimensional(400+ columns), so is that possible to compress? How about bit-vector?

Thank you.