A Rigorous Study Of The Deep Taylor Decomposition
This work highlights critical flaws in a widely used saliency method for interpretable AI, cautioning against over-reliance on its explanations.
The study investigated the Deep Taylor Decomposition (DTD) method for explaining deep neural networks and found that it is equivalent to the basic gradient×input method under certain conditions, and its theoretical assumptions are often violated, making explanations unreliable.
Saliency methods attempt to explain deep neural networks by highlighting the most salient features of a sample. Some widely used methods are based on a theoretical framework called Deep Taylor Decomposition (DTD), which formalizes the recursive application of the Taylor Theorem to the network's layers. However, recent work has found these methods to be independent of the network's deeper layers and appear to respond only to lower-level image structure. Here, we investigate the DTD theory to better understand this perplexing behavior and found that the Deep Taylor Decomposition is equivalent to the basic gradient$\times$input method when the Taylor root points (an important parameter of the algorithm chosen by the user) are locally constant. If the root points are locally input-dependent, then one can justify any explanation. In this case, the theory is under-constrained. In an empirical evaluation, we find that DTD roots do not lie in the same linear regions as the input - contrary to a fundamental assumption of the Taylor theorem. The theoretical foundations of DTD were cited as a source of reliability for the explanations. However, our findings urge caution in making such claims.