Papers by Dana Arad
CRISP: Persistent Concept Unlearning via Sparse Autoencoders (2026.acl-long)
Copied to clipboard
| Challenge: | Recent work has explored sparse autoencoders (SAEs) to perform precise interventions on monosemantic features, but most SAE-based methods operate at inference time, which does not create persistent changes in the model’s parameters. |
| Approach: | They propose a parameter-efficient method for persistent concept unlearning using SAEs that automatically identifies salient SAE features across multiple layers and suppresses their activations. |
| Outcome: | The proposed method outperforms previous methods on safety-critical unlearning tasks from the WMDP benchmark, successfully removing harmful knowledge while preserving general and in-domain capabilities. |
SAEs Are Good for Steering – If You Select the Right Features (2025.emnlp-main)
Copied to clipboard
| Challenge: | Sparse Autoencoders (SAEs) can learn a decomposition of a model’s latent space by analyzing the input tokens that activate them. |
| Approach: | They propose an unsupervised approach to learn a decomposition of a model’s latent space by analyzing the input tokens that activate them. |
| Outcome: | The proposed approach matches the performance of existing supervised methods by identifying features with low output scores and identifying them with input and output scores. |
Diffusion Lens: Interpreting Text Encoders in Text-to-Image Pipelines (2024.acl-long)
Copied to clipboard
| Challenge: | Text-to-image diffusion models use a latent text prompt to guide image generation . however, the process by which the encoder produces the text representation is unknown . |
| Approach: | They propose a method for analyzing the text encoder of T2I models by generating images from its intermediate representations. |
| Outcome: | The proposed method provides valuable insights into the text encoder component in T2I pipelines. |
Mechanisms of Prompt-Induced Hallucination in Vision–Language Models (2026.acl-long)
Copied to clipboard
William Rudman, Michal Golovanevsky, Dana Arad, Yonatan Belinkov, Carsten Eickhoff, Ritambhara Singh, Kyle Mahowald
| Challenge: | Large vision–language models (VLMs) often hallucinate by favoring textual prompts over visual evidence. |
| Approach: | They study the failure mode of large vision–language models by focusing on textual prompts over visual evidence. |
| Outcome: | The proposed model overestimates the number of objects in an image . it hallucinates additional waterlilies when asked to describe a mismatched number of items . the model ablation reduces prompt-induced hallucinosities by at least 40% without additional training . |
ReFACT: Updating Text-to-Image Models by Editing the Text Encoder (2024.naacl-long)
Copied to clipboard
| Challenge: | Text-to-image generative models encode factual associations that can quickly become outdated, diminishing their utility for end-users. |
| Approach: | They propose a method for editing factual associations in text-to-image models without retraining or explicit input from end-users. |
| Outcome: | The proposed method improves generalization and preservation of unrelated concepts on an existing dataset and compares with other methods. |