Another important element in TRUST topic (see also AI_7) is the “Consistency“: if two AI-based models are trained on the same task and both produce very similar predictions, the explanations should be very similar (to be highly consistent). It is worth to note here that a high level of consistency is desirable if the models really rely on similar relationships.
See (Hendrycks et al., 2019).
Main Question
Has the “consistency” parameter been considered properly?
Sub-Questions:
- Do models, which are trained on the same task and produce similar outputs, have different explanations?
- Alternatively, are those explanations similar?
- Are there methods to ensure that changes in the input data do not significantly affect the conssistency of the models?
References
- Hendrycks, D. and Dietterich, T. (2018). ‘Benchmarking Neural Network Robustness to Common Corruptions and Perturbations’. 7th International Conference on Learning Representations, Available at: https://openreview.net/forum?id=HJz6tiCqYm (Accessed 22 May 2024)