Learning Probabilistic Inference through Spike-Timing-Dependent Plasticity.

Area of Science:

• Computational Neuroscience
• Machine Learning
• Neurobiology

Background:

• The brain processes complex, uncertain information for decision-making via probabilistic inference.
• Existing models explain probabilistic inference in neural networks but not information acquisition.
• Understanding how neural networks learn from examples is crucial.

Purpose of the Study:

• To propose a model explaining how neural networks acquire information for probabilistic inference.
• To demonstrate how neural plasticity mechanisms can form the basis of probabilistic learning.
• To show the network's capability for immediate application of learned models in prediction and decision-making.

Main Methods:

• Utilizing spike-timing-dependent plasticity and intrinsic plasticity.
• Modeling ensembles of pyramidal cells with lateral inhibition.
• Recursively combining adaptive network motifs.

Main Results:

• Demonstrated the formation of probabilistic associations between neurons representing random variables.
• Showcased the network's ability to extract statistical information from complex input streams.
• Established the creation of an internal model for data-generating distributions, including higher-order moments.

Conclusions:

• The proposed model provides a mechanism for neural networks to learn and perform probabilistic inference.
• Spike-timing-dependent plasticity and intrinsic plasticity are key to building adaptive neural networks.
• The network can immediately leverage learned internal models for prediction and decision-making.