General differential Hebbian learning: Capturing temporal relations between events in neural networks and the brain

Area of Science:

• Computational Neuroscience
• Machine Learning
• Neuroscience

Background:

• Traditional Hebbian learning rules update synaptic strength based on coincident neural events.
• Differential Hebbian Learning (DHL) rules incorporate temporal relationships using derivatives but have limited update capabilities.
• Existing DHL rules do not fully capture the diverse learning processes observed in biological spike-timing-dependent plasticity (STDP).

Purpose of the Study:

• To propose a general Differential Hebbian Learning (G-DHL) rule with enhanced expressiveness and flexibility.
• To demonstrate that G-DHL can generate existing DHL rules and novel ones.
• To show G-DHL's utility in modeling diverse artificial neural signals and fitting experimental STDP data.

Main Methods:

• Developed a G-DHL rule combining pre- and post-synaptic neuron signals and their derivatives.
• Applied G-DHL to various artificial neural signals and experimental STDP datasets.
• Proposed signal pre-processing techniques and an automated procedure for rule component identification.

Main Results:

• The G-DHL rule successfully generated existing DHL rules and numerous new ones.
• G-DHL demonstrated flexibility by fitting diverse artificial neural signals and STDP experimental data.
• The identified G-DHL components offered heuristic guidance for exploring biophysical mechanisms of STDP.

Conclusions:

• The G-DHL rule offers a powerful and flexible framework for studying time-sensitive learning in neural networks.
• This approach enhances the modeling of complex synaptic plasticity beyond current DHL limitations.
• G-DHL facilitates a deeper understanding of learning mechanisms in both artificial and biological systems.