You can add location information to your Tweets, such as your city or precise location, from the web and via third-party applications. You always have the option to delete your Tweet location history. Learn more
A common, important way in which many real-world ML models are biased (beyond data bias) is that they privilege "canonical" samples and incorrectly process edge cases. This is the main failure mode of recommender systems, for instance. Not a matter of data representativeness.https://twitter.com/yoavgo/status/1274826491436687360 …
Pattern recognition is a poor fit for fat long tails of exceptions. It's not just the data, it's the model category. You could say that ML models are stereotyping machines.
the easiest description i would use: unsummarizable data. the data can be fully representative of reality, but diverse enough as to be unamenable to summary... and complex realities tend to require diverse samples, even within subcategories.
Every AI researcher should read Nassim Taleb
Maybe Taleb should read every AI researcher, just like Karpathyhttps://twitter.com/karpathy/status/1273890466467966976 …
Well said! Can’t agree more.
ML models don’t stereotype, and that might be the issue. A stereotype is a caricature: a synthetic data point where some features are at or beyond the limits of the empiric range. Stereotypes are in the tails.
Presumably stereotyping helps generalization, otherwise why would humans make such a heavy use of them? Alternatively, it might be an efficient strategy for encoding knowledge without encoding for the full empirical distribution
Just call them 'Intuition Machines'.
Yes! I couldn't agree more! Although I can only provide a taste of our work (with a shameless plugin ;-)), we showed on two different distributions that this is the case for the categorical approach, we also explored beyond the models presented there in https://www.aclweb.org/anthology/2020.winlp-1.31.pdf …
But the thing I like the most is that we have found a way out, to overcome this bias, or at least a step in a direction different than the common approach.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.
