Abstract: Neural architectures tend to fit their data with relatively simple functions. This "simplicity bias" is widely regarded as key to their success. This paper explores the limits of this ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results
Feedback