Saved in:
Bibliographic Details
Main Authors: Peng, Yifei, Zha, Zijie, Jin, Yu, Luo, Zhexu, Dai, Wang-Zhou, Ren, Zhong, Ding, Yao-Xiang, Zhou, Kun
Format: Preprint
Published: 2023
Subjects:
Online Access:https://arxiv.org/abs/2310.17451
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913875466649600
author Peng, Yifei
Zha, Zijie
Jin, Yu
Luo, Zhexu
Dai, Wang-Zhou
Ren, Zhong
Ding, Yao-Xiang
Zhou, Kun
author_facet Peng, Yifei
Zha, Zijie
Jin, Yu
Luo, Zhexu
Dai, Wang-Zhou
Ren, Zhong
Ding, Yao-Xiang
Zhou, Kun
contents Making neural visual generative models controllable by logical reasoning systems is promising for improving faithfulness, transparency, and generalizability. We propose the Abductive visual Generation (AbdGen) approach to build such logic-integrated models. A vector-quantized symbol grounding mechanism and the corresponding disentanglement training method are introduced to enhance the controllability of logical symbols over generation. Furthermore, we propose two logical abduction methods to make our approach require few labeled training data and support the induction of latent logical generative rules from data. We experimentally show that our approach can be utilized to integrate various neural generative models with logical reasoning systems, by both learning from scratch or utilizing pre-trained models directly. The code is released at https://github.com/future-item/AbdGen.
format Preprint
id arxiv_https___arxiv_org_abs_2310_17451
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Generating by Understanding: Neural Visual Generation with Logical Symbol Groundings
Peng, Yifei
Zha, Zijie
Jin, Yu
Luo, Zhexu
Dai, Wang-Zhou
Ren, Zhong
Ding, Yao-Xiang
Zhou, Kun
Artificial Intelligence
Computer Vision and Pattern Recognition
Graphics
Making neural visual generative models controllable by logical reasoning systems is promising for improving faithfulness, transparency, and generalizability. We propose the Abductive visual Generation (AbdGen) approach to build such logic-integrated models. A vector-quantized symbol grounding mechanism and the corresponding disentanglement training method are introduced to enhance the controllability of logical symbols over generation. Furthermore, we propose two logical abduction methods to make our approach require few labeled training data and support the induction of latent logical generative rules from data. We experimentally show that our approach can be utilized to integrate various neural generative models with logical reasoning systems, by both learning from scratch or utilizing pre-trained models directly. The code is released at https://github.com/future-item/AbdGen.
title Generating by Understanding: Neural Visual Generation with Logical Symbol Groundings
topic Artificial Intelligence
Computer Vision and Pattern Recognition
Graphics
url https://arxiv.org/abs/2310.17451