paper-with-me

Papers

Controlling Language and Diffusion Models by Transporting Activations

2024-10-30 · Pau Rodriguez, Arno Blaas, Michal Klein, Luca Zappella, Nicholas Apostoloff, Marco Cuturi, Xavier Suau

The increasing capabilities of large generative models and their ever more widespread deployment have raised concerns about their reliability, safety, and potential misuse. To address these issues, recent works have proposed to control model generation by steering model activations in order to effectively induce or prevent the emergence of concepts or behaviors in the generated output. In this paper we introduce Activation Transport (AcT), a general framework to steer activations guided by optimal transport theory that generalizes many previous activation-steering works. AcT is modality-agnostic and provides fine-grained control over the model behavior with negligible computational overhead, while minimally impacting model abilities. We experimentally show the effectiveness and versatility of our approach by addressing key challenges in large language models (LLMs) and text-to-image diffusion models (T2Is). For LLMs, we show that AcT can effectively mitigate toxicity, induce arbitrary concepts, and increase their truthfulness. In T2Is, we show how AcT enables fine-grained style control and concept negation.

📄 PDF Abstract BibTeX arXiv:2410.23054

Code (1)

apple/ml-act 공식 구현 pytorch

Tasks

Negation

Methods 이 논문이 사용한 방법론

Diffusion Diffusion models generate samples by gradually removing noise from a signal, and their training objective can be expressed as a reweighted variational lower-bound…

Similar Papers 제목 키워드 기반

Few Channels Draw The Whole Picture: Revealing Massive Activations in Diffusion Transformers

2026-05-13 · Evelyn Turri, Davide Bucciarelli, Sara Sarto, Lorenzo Baraldi 외 arxiv

Diffusion Transformers (DiTs) and related flow-based architectures are now among the strongest text-to-image generators, yet the internal mechanisms through which prompts shape image semantics remain poorly understood. I…

UniSteer: Text-Guided Flow Matching in Activation Space for Versatile LLM Steering

2026-05-28 · Yingdong Shi, Ruiming Zhang, Changming Li, Zhiyu Yang 외 arxiv

Activation-based control steers large language models (LLMs) by intervening on their internal representations during inference, and has emerged as an effective paradigm for controlling behaviors such as persona and style…

Instruction Following

Transporting Task Vectors across Different Architectures without Training

2026-02-13 · Filippo Rinaldi, Aniello Panariello, Giacomo Salici, Angelo Porrello 외 arxiv

Adapting large pre-trained models to downstream tasks often produces task-specific parameter updates that are expensive to relearn for every model variant. While recent work has shown that such updates can be transferred…

LLM-guided Instance-level Image Manipulation with Diffusion U-Net Cross-Attention Maps

2025-01-23 · Andrey Palaev, Adil Khan, Syed M. Ahsan Kazmi

The advancement of text-to-image synthesis has introduced powerful generative models capable of creating realistic images from textual prompts. However, precise control over image attributes remains challenging, especial…

Image GenerationImage Manipulation

Are language models aware of the road not taken? Token-level uncertainty and hidden state dynamics

2025-11-06 · Amir Zur, Atticus Geiger, Ekdeep Singh Lubana, Eric Bigelow arxiv

When a language model generates text, the selection of individual tokens might lead it down very different reasoning paths, making uncertainty difficult to quantify. In this work, we consider whether reasoning language m…