Knowledge Infused Learning (K-IL): Towards Deep Incorporation of Knowledge in Deep Learning

Combining neural networks with structured knowledge improves AI learning: The paper argues that deep learning models can be significantly enhanced by integrating external knowledge from structured knowledge graphs (like DBPedia and Yago) rather than relying solely on learning patterns from raw data, suggesting a hybrid approach produces better results.

Neuro-symbolic AI could make AI systems more explainable: By combining deep learning's pattern recognition with symbolic computing's reasoning capabilities, the authors propose that AI systems could become more interpretable and trustworthy—knowledge graphs can provide a framework for understanding how neural networks make decisions.

Knowledge graphs serve as a bridge between two AI approaches: The paper identifies structured knowledge graphs as a "common substrate" that can connect probabilistic deep learning methods with symbolic reasoning approaches, potentially creating more robust AI systems that leverage the strengths of both paradigms.

External knowledge can supervise and guide feature learning: Rather than allowing neural networks to learn features independently, incorporating knowledge from external sources helps supervise the learning process within the model's hidden layers, which could lead to more reliable and better-informed AI systems.

This approach supports responsible AI development: By building explainability scaffolding through knowledge graphs and combining symbolic reasoning with neural computing, the paper suggests a pathway toward developing AI systems that are more transparent, reasoned, and potentially more aligned with human understanding and values.