Declarative Networks: Towards Robust and Explainable Deep Learning

这个项目的目的是开发声明性机器学习技术，利用世界的固有结构和模型。深度学习已经成为机器学习的主要方法，许多产品和承诺都建立在这项技术之上。但深度学习成本高昂、不透明、脆弱，而且完全依赖于人类标记的数据。该项目旨在通过建立理论和算法使深度学习更加可靠，这些理论和算法允许将物理和数学模型嵌入深度学习框架中，提供性能保证和可解释性。这可能有利于基于机器学习的产品，这些产品可以通过视觉和语言自然地理解世界并与人类互动。

The aim of this project is to develop declarative machine learning techniques that exploit inherent structure and models of the world. Deep learning has become the dominant approach for machine learning with many products and promises built on this technology. But deep learning is expensive, opaque, brittle and relies solely on human labelled data. This project intends to make deep learning more reliable by establishing theory and algorithms that allow physical and mathematical models to be embedded within a deep learning framework, providing performance guarantees and interpretability. This would likely benefit machine learning based products that can understand the world and interact with humans naturally through vision and language.