REVIEW 5 cited by
Steering CLIP's vision transformer with sparse autoencoders
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
While vision models are highly capable, their internal mechanisms remain poorly understood -- a challenge which sparse autoencoders (SAEs) have helped address in language, but which remains underexplored in vision. We address this gap by training SAEs on CLIP's vision transformer and uncover key differences between vision and language processing, including distinct sparsity patterns for SAEs trained across layers and token types. We then provide the first systematic analysis on the steerability of CLIP's vision transformer by introducing metrics to quantify how precisely SAE features can be steered to affect the model's output. We find that 10-15\% of neurons and features are steerable, with SAEs providing thousands more steerable features than the base model. Through targeted suppression of SAE features, we then demonstrate improved performance on three vision disentanglement tasks (CelebA, Waterbirds, and typographic attacks), finding optimal disentanglement in middle model layers, and achieving state-of-the-art performance on defense against typographic attacks.
Forward citations
Cited by 5 Pith papers
-
The Signs Were Always There: Training-Free Concept Detection and Steering in Raw Transformer Dimensions
Sign patterns in the unrotated standard basis of transformer activations form independent binary feature registers that support training-free detection, prediction, and causal intervention across language, vision, and...
-
From Attribution to Action: A Human-Centered Application of Activation Steering
Activation steering paired with attribution enables intervention-based debugging in vision models, as all 8 interviewed experts shifted to hypothesis testing, most trusted observed responses, and highlighted risks lik...
-
IRIS: A Visual Cortex-Inspired Framework for Analyzing Orientation Selectivity in Vision Transformers
IRIS measures orientation selectivity in vision transformers and shows that a representational similarity score's peak predicts the best layer depth for fine-tuning.
-
Interpreting Video Representations with Spatio-Temporal Sparse Autoencoders
Spatio-temporal contrastive SAEs recover temporal coherence lost by hard TopK, improve action probes by +3.9% and retrieval by up to 2.8× R@1, and expose a monosemanticity metric artifact.
-
Beyond Interpretability: When, Why, and How Sparse Autoencoders Enable Label-Free Visual Steering
VS2 constructs steering vectors from sparse SAE features on unlabeled in-domain activations to improve zero-shot accuracy of CLIP models by 0.93-4.12% on CIFAR-100, CUB-200, and Tiny-ImageNet while remaining forward-p...
Discussion (0). Continue with ORCID to comment.