miniblog.

← Back to all posts
May 13, 2019 at 19:08
Really interesting paper exploring adversarial inputs to ML models: https://arxiv.org/abs/1905.02175 They conclude: * It's a property of the input data, not the training * You can even train a model on non-robust features and obtain a model that works well on the original input data!
Adversarial Examples Are Not Bugs, They Are Features
Adversarial Examples Are Not Bugs, They Are Features
Adversarial examples have attracted significant attention in machine learning, but the reasons for their existence and pervasiveness remain unclear. We demonstrate that adversarial examples can be directly attributed to the presence of non-robust features: features derived from patterns in the data distribution that are highly predictive, yet brittle and incomprehensible to humans. After capturing these features within a theoretical framework, we establish their widespread existence in standard datasets. Finally, we present a simple setting where we can rigorously tie the phenomena we observe in practice to a misalignment between the (human-specified) notion of robustness and the inherent geometry of the data.

Related Posts

Mar 5, 2026 at 10:27
12
Is there a good term for "after using this feature we realised that the best design is different from the current design"? In casual conversation I generally hear "bug" but there was nothing wrong with the original implementation.
Jun 25, 2025 at 23:39
22
I'm changing method definition syntax in my language: // old fun (this: Int) inc(): Int { this + 1 } // new method inc(this: Int): Int { this + 1 } The original syntax was inspired by Go, but the new syntax is more grep-friendly and perhaps more readable. Not sure about the verbosity though. Thoughts?
Jan 12, 2025 at 07:48
The games console market is fascinating: there's incentive to *not* provide upgraded models. You want the guarantee that a game for $X just works on any $X purchased. E.g. the Switch OLED has a bigger screen, and a better CPU than the original, but it's downclocked to match the original Switch's CPU.