المفاهيم الأساسية
A neurosymbolic system learns to infer programs capturing visual concepts across multiple domains.
الملخص
The content introduces a neurosymbolic system that learns to infer Template Programs, capturing visual concepts in a domain-general fashion. It explores tasks like few-shot generation and co-segmentation, showcasing superior performance compared to task-specific alternatives. The method is validated across 2D layouts, Omniglot characters, and 3D shapes, demonstrating its flexibility and capabilities.
Directory:
- Abstract
- Introduces a neurosymbolic system for inferring Template Programs.
- Explores tasks like few-shot generation and co-segmentation.
- Introduction
- Humans understand the visual world through concepts.
- Desires machines with similar abilities for flexible concept learning.
- Related Work
- Discusses methods focusing on concept learning and related tasks.
- Method
- Describes the framework for inferring Template Programs.
- Results
- Validates benefits of the method through comparisons in various visual domains.
- Discussion
- Explores out-of-distribution generalization, ablations, and additional capabilities of Template Programs.
الإحصائيات
We run experiments across multiple visual domains: 2D layouts, Omniglot characters, and 3D shapes.
Our method outperforms task-specific alternatives for few-shot generation tasks.
اقتباسات
"We introduce Template Programs: programmatic expressions from a domain-specific language."
"Our neurosymbolic system learns how to infer programs that capture visual concepts."