Can you trust your Machine Learning model?
Supervised machine learning models boast remarkable predictive capabilities.
But can you trust your model?
Will it work in deployment?
What else can it tell you about the world?
We want models to be not only good, but interpretable.
I’ll consider only supervised learning in this post and not other machine learning paradigms, such as reinforcement learning and interactive learning.
At present, interpretability has no formal technical meaning. Before we can determine which meanings might be appropriate, we must ask what the real world objectives of interpretability?
The demand for interpretability arises when there is a mismatch between the formal objectives of supervised learning (test set predictive performance) and the real world costs in a deployment setting.
Consider that most common evaluation metrics for supervised learning require only predictions, together with ground truth, to produce a score. These metrics can be be assessed for every supervised learning model.
So, the desire for an interpretation suggests that in some scenarios, predictions alone and metrics calculated on these predictions are not adequate to characterize the model.
We should then ask, what are these other desire and under what circumstances are they sought?
Understanding the word “Trust”
Is it simply confidence that a model will perform well?
If so, a sufficiently accurate model should be demonstrably trustworthy and interpretability would serve no purpose.
Trust might also be defined subjectively.
For example, a person might feel more at ease with a well understood model, even if this understanding served no obvious purpose.
We may trust the model to make accurate predictions but not to account for biases in the training data for the model’s own effect in perpetuating a cycle of incarceration.
To reduce such biases refer to:
Casualness in Learning
The associations learned by supervised learning algorithms are not guaranteed to reflect causal relationships. There could always exist unobserved causes responsible for both associated variables. One might hope, however, that by interpreting supervised learning models, we could generate hypotheses that scientists could then test experimentally.
Transfer of the Learning
Typically we choose training and test data by randomly partitioning examples from the same distribution. We then judge a model’s generalization error by the gap between its performance on training and test data. However, humans exhibit a far richer capacity to generalize, transferring learned skills to unfamiliar situations. We already use machine learning algorithms in situations where such abilities are required, such as when the environment is non stationary.
Now that we have seen the need of interpretability in models, here are some takeaways for free!
Linear models are not strictly more interpretable than deep neural networks
When choosing between linear and deep models, we must often make a trade-off between algorithmic transparency and decomposability. This is because deep neural networks tend to operate on raw or lightly processed features. So if nothing else, the features are intuitively meaningful, and post-hoc reasoning is sensible.
However, in order to get comparable performance, linear models often must operate on heavily hand-engineered features.
Claims about interpretability must be qualified
To be meaningful, any assertion regarding interpretability should fix a specific definition. If the model satisfies a form of transparency, this can be shown directly.
Post-hoc interpretations can potentially mislead
Caution against blindly embracing post-hoc notions of interpretability, especially when optimized to placate subjective demands. In such cases, one might deliberately or not optimize an algorithm to present misleading but plausible explanations.
For Technicality, the various evaluation matrices in ML we use apart from score are:
They are very easy to understand, I hope you’ll have fun reading about them!
Vivek Gupta: https://www.linkedin.com/in/vivekg-/
Follow me on Quora: https://www.quora.com/profile/Vivek-Gupta-1493
Check out my legal space here: https://easylaw.quora.com
