Abstract:We introduce a new method for integrating neural networks with logic programming in Neural-Symbolic AI (NeSy), aimed at learning with distant supervision, in which direct labels are unavailable. Unlike prior methods, our approach does not depend on symbolic solvers for reasoning about missing labels. Instead, it evaluates logical implications and constraints in a differentiable manner by embedding both neural network outputs and logic programs into matrices. This method facilitates more efficient learning under distant supervision. We evaluated our approach against existing methods while maintaining a constant volume of training data. The findings indicate that our method not only matches or exceeds the accuracy of other methods across various tasks but also speeds up the learning process. These results highlight the potential of our approach to enhance both accuracy and learning efficiency in NeSy applications.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to solve the problem of how to effectively combine neural networks with logic programming in the **Distant Supervision** environment. Specifically, the author proposes a new method to achieve the learning tasks in Neural - Symbolic AI (NeSy) by introducing **Differentiable Logic Programming**. This method is especially suitable for situations where label information is not directly available. #### Main problems and background 1. **Combination of neural networks and symbolic reasoning**: - Neural networks are good at learning complex representations from input - output pairs, but they require a large amount of training data and perform poorly on tasks that require logical reasoning. - Symbolic reasoning can learn from a small amount of data, but it is sensitive when dealing with noise and non - symbolic data. 2. **Learning in the distant supervision environment**: - In the distant supervision environment, direct label information is not available, and learning is carried out through labels generated by rules, external data or knowledge bases. For example, in the MNIST Addition task, the label is the sum of two handwritten digits, not the individual label of each image. 3. **Limitations of existing methods**: - Existing methods usually rely on symbolic solvers for reasoning, which will lead to high computational costs, especially when operations such as weighted model counting or enumerating stable models need to be performed in each iteration. - These methods are difficult to directly handle logical constraints and are not suitable for direct application to the learning of neural networks. #### Main contributions of the paper 1. **Proposing a new architecture**: - The author proposes a new architecture that combines neural networks with logic programming in a differentiable way. This method can directly evaluate logical implications and constraints without relying on symbolic solvers for reasoning, thus achieving more efficient learning in the distant supervision environment. 2. **Experimental verification**: - Through experimental verification, the author's method not only matches or even exceeds the existing methods based on symbolic solvers in terms of accuracy, but also significantly reduces the training time of neural networks and improves computational efficiency. 3. **Application potential**: - This method shows the potential to improve accuracy and learning efficiency in Neural - Symbolic AI applications, especially in the distant supervision environment. ### Summary This paper mainly solves the problem of how to effectively combine neural networks with symbolic reasoning by introducing Differentiable Logic Programming in the distant supervision environment. The new method proposed by the author not only performs well in terms of accuracy, but also greatly improves the learning efficiency, providing new research directions and technical means for the field of Neural - Symbolic AI.

Differentiable Logic Programming for Distant Supervision

Differentiable Inductive Logic Programming for Structured Examples

A Study on Differentiable Logic and LLMs for EPIC-KITCHENS-100 Unsupervised Domain Adaptation Challenge for Action Recognition 2023

Differentiable Logic Programming for Probabilistic Reasoning

Neural Probabilistic Logic Programming in Discrete-Continuous Domains

Learning from Interpretation Transition Using Differentiable Logic Programming Semantics

Learning Differentiable Logic Programs for Abstract Visual Reasoning

Learning First-Order Rules with Differentiable Logic Program Semantics

Differentiable Logic Machines

Analyzing Differentiable Fuzzy Logic Operators

dPASP: A Comprehensive Differentiable Probabilistic Answer Set Programming Environment For Neurosymbolic Learning and Reasoning

DeepLogic: Joint Learning of Neural Perception and Logical Reasoning.

Differentiable Inductive Logic Programming in High-Dimensional Space

Differentiable Learning of Logical Rules for Knowledge Base Reasoning

Deep Inductive Logic Programming meets Reinforcement Learning

A Differentiable First-Order Rule Learner for Inductive Logic Programming

Learning Interpretable Differentiable Logic Networks

Neuro-Symbolic Inductive Logic Programming with Logical Neural Networks

Beyond Logic Programming for Legal Reasoning

Logical Learning Through a Hybrid Neural Network with Auxiliary Inputs