Hearing Aid Self-Organizing Map for Acoustic Program Selection

The patent segments the neural network into a single layer of neurons arranged in a grid configuration, eliminating the need for multiple layers. This segmentation reduces the computational complexity while maintaining the essential classification functionality through vector comparison between sensor data and neuron vectors.

The patent uses a simplified copy of the neural network structure - a single layer grid of neurons with fixed weights that are trained offline. This copy maintains the classification capability without requiring intensive real-time calculations for weight updates, as the weights are predetermined and stored in memory.

The patent arranges neurons in a two-dimensional grid configuration with spatial neighbors, adding a dimensional aspect to the single-layer structure. This grid arrangement allows neurons to share information with adjacent neurons, effectively increasing the system's capacity to handle varying acoustic characteristics without requiring multiple layers.

The single-layer grid neuron structure serves multiple functions: it performs classification, stores acoustic situation patterns in its vectors, and allows for offline training. The grid configuration with neighbors enables the system to generalize across varying acoustic values by propagating information spatially, making the simple structure universally applicable to diverse acoustic environments.

The patent implements preliminary offline training of the neural network before actual use. During this phase, the system learns acoustic situations and programs using recorded data, storing the learned patterns in the neuron vectors and labels. This preliminary action enables the system to provide automatic program selection from the start without requiring real-time training capabilities.

The system incorporates feedback mechanisms where user selections during the learning phase are used to adjust neuron labels and vectors. The system monitors which programs users manually select in response to detected acoustic situations and uses this feedback to refine the automatic selection algorithm, improving adaptability over time through iterative refinement rather than real-time learning.