Perspectives of neural symbolic integration

Auteurs

• Barbara Hammer

Paramètres

• Informatique, programmation

En savoir plus sur le livre

The human brain's remarkable ability to understand, interpret, and produce language primarily relies on the left hemisphere. Language acquisition is innate, as demonstrated by specific language impairment (SLI), which indicates a lack of grammaticality sense. Language is characterized by strong compositionality and structure, linking biological neural networks to the processing and generation of high-level symbolic structures. In contrast, artificial neural networks and logic do not share this close connection. Two major paradigms in artificial intelligence—symbolic inference mechanisms and statistical machine learning—each have distinct strengths and weaknesses. Statistical methods provide flexible and effective tools for handling corrupted or noisy data, uncertainty, and missing information, as seen in robotics, medical measurements like EEG and EKG, and financial indices. However, these models often function as black boxes, complicating the integration of prior knowledge and human inspection while struggling with complex structures of objects, classes, and relations. Symbolic mechanisms excel in intuitive human-machine interaction and integrating complex prior knowledge, but they are less effective in managing uncertainty and noise in large-scale real-world data sets. Ultimately, the strengths and weaknesses of these two approaches complement each other.

Achat du livre

• Année de publication: 2007