The last decade has witnessed considerable progress in artificial intelligence, marked by breakthroughs in deep learning technologies, accompanied by the proliferation of powerful computing hardware. Although the focus has primarily remained on developing faster and more efficient learning systems through novel mathematical frameworks or clever algorithmic advances, leveraging existing knowledge, be it in the form of symbolic rules, scientific/physics-based models, established empirical relations, or simply common sense knowledge, has often been overlooked. For instance, using models obtained with known dynamics modeling techniques could considerably alleviate the high sample complexity of learning in robotic control tasks; learning from expert demonstrations could inherently imply several problem-specific requirements without having to explicitly specify them; and theoretically established relationships between quantities could inform the types of constraints to be imposed in physics-based learning tasks. Learning and decision-making methods guided by existing domain knowledge could thus be a powerful tool for solving various practical real-world problems in a data-efficient manner to create robust/safe AI systems.
Knowledge-augmented/guided AI, while not a new area, comprises numerous open scientific and technical challenges along several dimensions, from noise modeling to active/passive knowledge representation and elicitation or guided training of AI models to the adaptation of such research in high-impact application domains including healthcare, agriculture, climate, chemistry, and computing systems design etc. Research in this space is even more vital now, as we strive for better, reliable, trustworthy, safe and unbiased AI-driven systems and solutions. The goal of this Research Topic is to motivate and promote contributions that aim to address open research questions in knowledge-guided AI along all the key dimensions. The consequent assortment of articles will, hopefully, not only enrich and highlight the pitfalls of pure data-driven learning and the importance of knowledge in AI but will also motivate technology creators to consider knowledge as a powerful tool in addressing the issues of reliability, trust, safety and bias in real-world deployments of AI-powered technology.
Inviting submission of high quality articles comprising either original research (fundamental or applied), reviews and systematic reviews, hypotheses and theory articles, or technology and code. Potential contributions may be in, but not limited to, the following possible conceptual and applied topics where knowledge-guided modeling and learning are expected to be of significant impact.
1. Human-guided decision making
2. Machine Learning (deep or shallow) with prior & active knowledge
3. Learning / decision-making with causal Knowledge
4. Causal explainable interactive learning
5. Reinforcement learning and Planning under constraints
6. Knowledge + Data in Program Synthesis
7. Synergistic integration of physics-based models and data-driven models
8. Physics-based AI systems
9. Application of knowledge-guided AI in the following areas:
- AI to accelerate scientific discovery and engineering design
- AI for computing systems design and management
- Knowledge-driven intelligent design and architecture
- Knowledge for Safety and Trust in Healthcare AI
- Robotics and configurable design with knowledge
- AI in Logistics, Transportation systems and traffic control
Note, the list is not exhaustive and any article of relevance to the topic is welcome.
The last decade has witnessed considerable progress in artificial intelligence, marked by breakthroughs in deep learning technologies, accompanied by the proliferation of powerful computing hardware. Although the focus has primarily remained on developing faster and more efficient learning systems through novel mathematical frameworks or clever algorithmic advances, leveraging existing knowledge, be it in the form of symbolic rules, scientific/physics-based models, established empirical relations, or simply common sense knowledge, has often been overlooked. For instance, using models obtained with known dynamics modeling techniques could considerably alleviate the high sample complexity of learning in robotic control tasks; learning from expert demonstrations could inherently imply several problem-specific requirements without having to explicitly specify them; and theoretically established relationships between quantities could inform the types of constraints to be imposed in physics-based learning tasks. Learning and decision-making methods guided by existing domain knowledge could thus be a powerful tool for solving various practical real-world problems in a data-efficient manner to create robust/safe AI systems.
Knowledge-augmented/guided AI, while not a new area, comprises numerous open scientific and technical challenges along several dimensions, from noise modeling to active/passive knowledge representation and elicitation or guided training of AI models to the adaptation of such research in high-impact application domains including healthcare, agriculture, climate, chemistry, and computing systems design etc. Research in this space is even more vital now, as we strive for better, reliable, trustworthy, safe and unbiased AI-driven systems and solutions. The goal of this Research Topic is to motivate and promote contributions that aim to address open research questions in knowledge-guided AI along all the key dimensions. The consequent assortment of articles will, hopefully, not only enrich and highlight the pitfalls of pure data-driven learning and the importance of knowledge in AI but will also motivate technology creators to consider knowledge as a powerful tool in addressing the issues of reliability, trust, safety and bias in real-world deployments of AI-powered technology.
Inviting submission of high quality articles comprising either original research (fundamental or applied), reviews and systematic reviews, hypotheses and theory articles, or technology and code. Potential contributions may be in, but not limited to, the following possible conceptual and applied topics where knowledge-guided modeling and learning are expected to be of significant impact.
1. Human-guided decision making
2. Machine Learning (deep or shallow) with prior & active knowledge
3. Learning / decision-making with causal Knowledge
4. Causal explainable interactive learning
5. Reinforcement learning and Planning under constraints
6. Knowledge + Data in Program Synthesis
7. Synergistic integration of physics-based models and data-driven models
8. Physics-based AI systems
9. Application of knowledge-guided AI in the following areas:
- AI to accelerate scientific discovery and engineering design
- AI for computing systems design and management
- Knowledge-driven intelligent design and architecture
- Knowledge for Safety and Trust in Healthcare AI
- Robotics and configurable design with knowledge
- AI in Logistics, Transportation systems and traffic control
Note, the list is not exhaustive and any article of relevance to the topic is welcome.