Related Experiment Videos
Mutual information of sparsely coded associative memory with self-control and ternary neurons
D Bollé1, D R Dominguez, S Amari
1Instituut voor Theoretische Fysica, Katholieke Universiteit Leuven, Belgium. desire.bolle@fys.kuleuven.ac.be
Summary
A novel time-dependent threshold enhances attractor associative memory models. This self-control mechanism improves storage capacity and retrieval quality, especially with sparse coding.
Area of Science:
- Computational Neuroscience
- Artificial Intelligence
- Memory Models
Background:
- Attractor associative memory models store and retrieve patterns.
- Ternary neurons (-1, 0, +1) offer unique computational properties.
- Autonomous functioning requires robust retrieval dynamics.
Purpose of the Study:
- To investigate the impact of a macroscopic time-dependent threshold on memory retrieval.
- To determine if an adaptive threshold can ensure autonomous model operation.
- To enhance the performance of ternary neuron associative memory models.
Main Methods:
- Examining retrieval dynamics under a time-dependent threshold.
- Analyzing the influence of cross-talk noise and pattern activity on the threshold.
- Utilizing sparse coding approximations for theoretical analysis.
- Employing mutual information as a key metric for retrieval quality.
Main Results:
- An appropriately adapted time-dependent threshold enables autonomous model functioning.
- Significant improvements in fixed-point retrieval dynamics were observed.
- Storage capacity, basin of attraction, and information content were enhanced.
- Mutual information proved to be a critical parameter for sparsely coded models.
Conclusions:
- Time-dependent, adaptive thresholds are crucial for autonomous associative memory operation.
- This approach substantially boosts model performance, particularly under sparse coding conditions.
- Mutual information effectively quantifies retrieval quality in these advanced memory systems.