Currently the network architecture is fixed at 81 → 24 → 2, supporting only binary digit recognition (0 vs 1).
Proposed enhancement:
- Allow configurable output neuron count (e.g., 10 for digits 0-9)
- Update the label selector UI to support multiple classes
- Adjust the training data format to support multi-class one-hot encoding
- Consider a larger hidden layer or configurable architecture
This would make the tool useful for exploring how network capacity relates to task complexity.
Currently the network architecture is fixed at 81 → 24 → 2, supporting only binary digit recognition (0 vs 1).
Proposed enhancement:
This would make the tool useful for exploring how network capacity relates to task complexity.