The field of remote sensing and machine learning is witnessing a significant shift towards weak supervision, enabling models to learn from limited or noisy labels. This trend is driven by the need for more efficient and accurate data acquisition and interpretation methods. Researchers are exploring innovative approaches to integrate weak supervision with symbolic methods, such as inductive logic programming, to improve model interpretability and reliability.

One of the key directions in this area is the development of neuro-symbolic frameworks that combine the strengths of neural networks and symbolic reasoning. These frameworks aim to provide structured relational constraints that guide learning and improve model transparency and accountability.

Another important area of research is the extension of logic programming techniques, such as answer set programming, to accommodate uncertainty and weighted annotations. This allows for the propagation of uncertainty and weights through models and events, enabling more accurate and robust reasoning.

Noteworthy papers in this area include:

• LiDAR Remote Sensing Meets Weak Supervision, which provides a comprehensive review of weakly supervised techniques in LiDAR remote sensing.
• Neuro-symbolic Weak Supervision, which proposes a semantics for neuro-symbolic frameworks to improve multi-instance partial label learning.