SOTR(Self Organizing Task Representation) is a method we worked on in the continual learning paradigm. This repo includes the code-base of the work.

• GWR - Grow when required
• NC50 - Fifty incremental new class scenario
• Multi-head - Multi-headed deep learning model architecture
• CNN - Convolutional neural network
• SI - Synaptic intelligence
• AP -Average precision
• SE - Squeeze and Excitation
• FC - Fully connected deep neural network
• Task - A set of data provided to the network at once during the incremental training

Fifty incremental new class scenario training and testing

• Here each category of the core50 dataset are incrementally fed to the network.
• File path: core50_hybrid/benchmarks/sonn/NC50_incremental_test.py

Five incremental new class scenario training and testing

• Here all the fifty individual classes of core50 dataset are incrementally fed to the network.
• File path: core50_hybrid/benchmarks/sonn/NC5_incremental_test.py

GWR neural network

• This has different implementations of GWR using networkx module and PyTorch(only one used at each implementation).
• File path: core50_hybrid/benchmarks/sonn/gwr.py