Searcharxiv⌕ Search

arXiv subjects

Janhavi Khindkar

Publications and source records attributed to Janhavi Khindkar.

2 recordsLinked to original sources

Refusal Register is Causally Steerable: Decomposing the Refusal Direction into Category-Aligned Components in Gemma-2-9B-IT

Prior work shows that refusal in instruction-tuned language models can be ablated by intervening on a single residual-stream direction (Arditi et al., 2024), and more recent work finds multiple refusal-mediating directions and concept cones (Wollschläger et al., 2025). We ask a different question: do refusal-related directions carry information about the register of refusal language - how the model refuses - and not only whether it refuses? In Gemma-2-9B-IT we identify two SVM-derived linear directions (B1 for harmful-content vs. neutral, B2 for self-reference vs. neutral) at layer 20 with cos(B1, B2) = 0.252 (about 75 degrees, substantially non-parallel). Adding the harmful-vs-neutral difference-of-means direction (cosine 0.68 with B1) to the residual stream of self-reference prompts at alpha = 2.0 shifts content-safety vocabulary from 2.55 to 10.37 hits per 100 tokens, while a magnitude-matched random direction produces 1.13-2.16; the model continues to refuse but switches register from meta-style ("as a language model...") to content-style ("this would be harmful"). The same register-injection effect reproduces qualitatively in Gemma-2-2B-IT at layer 12 (content vocabulary 1.72 -> 6.37 at alpha = 1.0, vs. random control 1.35), with a narrower coherent-injection window than in 9B. The two directions are not symmetric: the B2 -> harmful intervention produces a much smaller register effect, and across ablation, sufficiency, and injection B1's causal effects are consistently larger than B2's. We frame this asymmetry as part of the finding: B1 appears to encode both category detection and refusal-vocabulary information, while B2 encodes primarily category detection. We frame the contribution as identifying category-aligned components related to the Arditi-style difference-of-means direction that play distinct functional roles in refusal register.

cs.CL↗

GAMMA: Generative Augmentation for Attentive Marine Debris Detection

We propose an efficient and generative augmentation approach to solve the inadequacy concern of underwater debris data for visual detection. We use cycleGAN as a data augmentation technique to convert openly available, abundant data of terrestrial plastic to underwater-style images. Prior works just focus on augmenting or enhancing existing data, which moreover adds bias to the dataset. Compared to our technique, which devises variation, transforming additional in-air plastic data to the marine background. We also propose a novel architecture for underwater debris detection using an attention mechanism. Our method helps to focus only on relevant instances of the image, thereby enhancing the detector performance, which is highly obliged while detecting the marine debris using Autonomous Underwater Vehicle (AUV). We perform extensive experiments for marine debris detection using our approach. Quantitative and qualitative results demonstrate the potential of our framework that significantly outperforms the state-of-the-art methods.

cs.CV↗