arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.03038cs.CVcs.AI
收藏

Bongards在感知与推理边界上的问题:程序还是语言?

Bongards at the Boundary of Perception and Reasoning: Programs or Language?

Cornell University(康奈尔大学);Shanghai Jiao Tong University(上海交通大学);Nanyang Technological University(南洋理工大学)

Cassidy Langenfeld, Claas Beger, Gloria Geng, Wasu Top Piriyakulkij, Keya Hu, Yewen Pu, Kevin Ellis

AI总结:

本文提出了一种神经符号方法,利用大语言模型和贝叶斯优化解决Bongard问题,通过生成参数化程序化表示来提升视觉推理能力。

摘要:

视觉语言模型(VLMs)在日常视觉任务中取得了重大进展,例如对自然图像进行描述,或回答关于此类图像的常识性问题。但人类拥有令人困惑的能力,能够将它们的视觉推理能力应用于截然不同的新情境,这种技能通过经典的视觉推理挑战集——Bongard问题进行严格测试。我们提出了一种神经符号方法来解决这些问题:给定一个Bongard问题的假设解决方案规则,我们利用大语言模型生成参数化的程序化表示,并使用贝叶斯优化进行参数拟合。我们评估了我们的方法在根据真实规则分类Bongard问题图像以及从头开始解决这些问题上的表现。

英文摘要:

Vision-Language Models (VLMs) have made great strides in everyday visual tasks, such as captioning a natural image, or answering commonsense questions about such images. But humans possess the puzzling ability to deploy their visual reasoning abilities in radically new situations, a skill rigorously tested by the classic set of visual reasoning challenges known as the Bongard problems. We present a neurosymbolic approach to solving these problems: given a hypothesized solution rule for a Bongard problem, we leverage LLMs to generate parameterized programmatic representations for the rule and perform parameter fitting using Bayesian optimization. We evaluate our method on classifying Bongard problem images given the ground truth rule, as well as on solving the problems from scratch.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑