Predictive Coding

Predictive Coding (PC) is a theoretical framework in cognitive science that holds that the human brain processes cognition through spatiotemporal predictions of the visual world. Based on PC theory, existing studies have explored spatiotemporal prediction neural networks and simulated their two core mechanisms: learning prediction errors and hierarchical structures. However, these models have not shown improvements in prediction skills in real-world prediction tasks and have ignored the precision weighting mechanism in PC theory. The precision weighting mechanism holds that the brain pays more attention to signals with lower precision, which helps improve the cognitive ability and efficiency of the human brain.

In neuroscience, predictive coding (also known as predictive processing) is a theory of brain function that posits that the brain continuously generates and updates a "mental model" of the environment. According to the theory, this mental model is used to predict input signals from the senses, which are then compared to the actual input signals from those senses. With the increasing popularity of representation learning, this theory has been actively explored and applied in machine learning and related fields.

References

【1】Wikipedia

【2】HelmFluid: Learning Helmholtz Dynamics for Interpretable Fluid Prediction