A New Hierarchical Temporal Memory Algorithm Based on Activation Intensity.

Area of Science:

• Artificial Intelligence
• Computational Neuroscience

Background:

• Hierarchical Temporal Memory (HTM) is a human-cortex-inspired model effective for sequence learning and time-series analysis.
• Conventional HTM models have limitations in representational capacity due to simplified column states and fixed learning strategies.

Purpose of the Study:

• To propose a new HTM algorithm that enhances sequence learning by incorporating column activation intensity.
• To develop a self-adaptive nonlinear learning strategy for dynamic synaptic adjustment.

Main Methods:

• Introduced column activation intensity to retain finer-grained input information.
• Implemented a self-adaptive nonlinear learning strategy for dynamic synaptic connection adjustments.
• Conducted experiments on real-world time-series datasets.

Main Results:

• The proposed HTM algorithm demonstrated higher accuracy in sequence learning tasks.
• The new method achieved a reduced time overhead compared to existing models.
• Enhanced representation capability and temporal context learning were observed.

Conclusions:

• The activation intensity-based HTM algorithm offers superior performance for time-series applications.
• The self-adaptive learning strategy effectively improves HTM's learning dynamics.
• This approach represents a significant advancement in HTM for sequence learning.