Search Results for author:
Found 4 papers, 2 papers with code
Towards Principled Evaluations of Sparse Autoencoders for Interpretability and Control
However, the lack of ground-truth for these features in realistic scenarios makes the validation of recent approaches, such as sparse dictionary learning, elusive.
Is This the Subspace You Are Looking for? An Interpretability Illusion for Subspace Activation Patching
We demonstrate this phenomenon in a distilled mathematical example, in two real-world domains (the indirect object identification task and factual recall), and present evidence for its prevalence in practice.
Rethinking Backdoor Attacks
In a backdoor attack, an adversary inserts maliciously constructed backdoor examples into a training set to make the resulting model vulnerable to manipulation.
Towards Deep Learning Models Resistant to Adversarial Attacks
Its principled nature also enables us to identify methods for both training and attacking neural networks that are reliable and, in a certain sense, universal.
Cannot find the paper you are looking for? You can
Submit a new open access paper.
