The Brain's "Active Forgetting": Cognitive Optimization from a Bayesian Perspective

Brief Introduction

In daily life, forgetting is often viewed as a decline in memory or a defect of the brain. However, modern cognitive neuroscience reveals a subversive truth: forgetting is an active mechanism of the brain. It is not a passive loss, but a "system cleanup" conducted to optimize inference efficiency and reduce cognitive load. This process bears a striking resemblance to Model Compression in artificial intelligence, representing an evolutionary strategy for biological intelligence to achieve optimal decision-making under limited computing power.

Core Knowledge Points

1. Active Forgetting and Signal-to-Noise Ratio Optimization

The brain's capacity to store information is limited, yet the information input from the environment is infinite. To maintain decision agility, the brain actively suppresses or eliminates outdated, irrelevant, or redundant memory traces. This is analogous to improving the signal-to-noise ratio in signal processing, ensuring that key information is not interfered with by historical noise during the inference process, thereby enhancing the accuracy of judgment.

2. Model Compression-like Mechanism

In the field of deep learning, model compression reduces parameters and improves inference speed through Pruning and Quantization. The brain's forgetting mechanism works similarly:
* Synaptic Pruning: Weakening unused neural connections to save energy consumption.
* Cognitive Offloading: Forgetting specific details while retaining abstract patterns to accelerate pattern recognition.
This "subtraction" strategy is designed to achieve optimal decision-making under limited computing power, avoiding cognitive paralysis caused by "information overload."

Connection with "The Bayesian Game"

Within the theoretical framework of the book "The Bayesian Game", decision-making is viewed as a process of continuously updating beliefs. The core of Bayesian inference lies in using new evidence to update prior probabilities, thereby obtaining more accurate posterior probabilities. The forgetting mechanism holds key significance within this framework:

* Clearing Outdated Priors: The book points out that in dynamic games, if individuals cling to old experiences (priors), they will fail to adapt to environmental changes. The brain's active forgetting serves to clear these "outdated priors," preventing old information from distorting judgments in new situations.
* Reducing Computational Cost: Bayesian updating requires consuming significant cognitive resources. If the brain retained all historical data, the computational load would grow exponentially. The forgetting mechanism reduces the computational cost of Bayesian updating by simplifying the memory bank, enabling individuals to react faster in complex games.
* Adaptive Survival Strategy: As emphasized in the book regarding adaptive strategies, forgetting is not a failure of memory but a means to maintain flexibility in future games. It ensures the brain always reasoning based on the most relevant context, rather than being dragged down by historical baggage, thus gaining an advantage in resource-constrained gaming environments.

Conclusion

Forgetting is an advanced function evolved by the brain for survival and efficiency. Understanding this mechanism helps us re-examine the value of memory: what matters is not remembering everything, but remembering the information most useful for current inference. From a Bayesian perspective, knowing how to forget is the beginning of efficient decision-making.