跳到正文
HuggingFace Daily Papers(社區熱門論文)·· 18 天前AI 評分57

研究發現視覺語言模型可透過無關視覺任務傳遞隱藏偏好

Subliminal Learning in Vision-Language Models

AI 導讀

研究發現,視覺語言模型可透過與偏好無關的視覺任務,把教師模型的隱藏偏好傳給學生模型。研究涵蓋參數量由 2B 至 7B 的模型,在動物偏好任務中,學生模型選擇目標的比例最高比對照組高約 10 個百分點。即使模型選擇相同圖像,受偏好教師影響的學生模型,其解釋亦會使用更多與教師相同的詞語和短語。

正文

Published on Nov 4

Authors:

,

Abstract

Subliminal learning can transfer hidden preferences from teacher to student vision-language models through unrelated visual tasks, affecting image choices and explanations.

Subliminal learning is a phenomenon in large language models (LLMs) where a teacher passes a hidden preference to a student through training data unrelated to that preference. We study whether subliminal learning also occurs in vision-language models (VLMs). We finetune a teacher VLM to prefer a specific animal or facial attribute and use a control teacher from the same base model. Both teachers generate numbers for an unrelated visual task involving geometric shapes. Two students initialized from the same base VLM are finetuned on the same images, using numbers generated by either the biased or control teacher. Across VLMs ranging from 2B to 7B parameters, we find that both animal and facial preferences can transfer through this unrelated task. Students trained on biased-teacher outputs select the target more often than control students, with a difference of up to sim10 percentage points in the animal preference task. Transfer can become stronger with more training data and varies with the target preference, model, and how the numbers are generated. Beyond preferences, we show that the biased student's explanations use more of the same words and phrases as the biased teacher's explanations than those of the control student, even when all three models select the same image. These results matter for training VLMs and show that training on unrelated data may change both a VLM's image preferences and how it explains its choices, and that transfer can become stronger with more training samples.

View arXiv page View PDF Project page Add to collection

Get this paper in your agent:

hf papers read 2611.08538

Don't have the latest CLI?

curl -LsSf https://hf.co/cli/install.sh | bash

Models citing this paper 0

No model linking this paper

Cite arxiv.org/abs/2611.08538 in a model README.md to link it from this page.

Datasets citing this paper 0

No dataset linking this paper

Cite arxiv.org/abs/2611.08538 in a dataset README.md to link it from this page.

Spaces citing this paper 0

No Space linking this paper

Cite arxiv.org/abs/2611.08538 in a Space README.md to link it from this page.

Collections including this paper 0

No Collection including this paper

Add this paper to a collection to link it from this page.

來源:HuggingFace Daily Papers(社區熱門論文) · huggingface.co