How trentbrick/sdmcontinuallearner works
This repository is research code for continual-learning experiments centered on Sparse Distributed Memory (SDM), with preprocessing and model configuration alongside implementations of NISPA and several comparison baselines. The documented workflow is to prepare and split image datasets, configure a model and experiment, then train/evaluate across tasks; experiment scripts cover pretraining, continual learning, and ablations. NISPA’s entry point and learner flow are visible in excerpts. Other connections are included only where the README or source index supports them; many experiment and training implementations were not sampled.
Main entry points: learner.py NISPA learner · combine_params.py Parameter assembly · main.py NISPA entry point · iBatchLearn.py Baseline runner
Components 33
Data preparation
- Dataset setup
py_scripts/setup_cont_learning_datasets.py - Dataset parameters
py_scripts/dataset_params.py - Data loaders
py_scripts/data_loaders.py
SDM experiments
- Model parameters
py_scripts/model_params.py - Parameter assembly
py_scripts/combine_params.py - SDM model family
models/SDM.py - Model base
models/Base_Model.py - CL benchmarks
models/CL_Benchmarks.py - ConvSDM
models/Conv_SDM.py - Continual-learning run
exp_commands/cont_learn_on_SplitCIFAR10_embeddings.py - Pretraining runs
exp_commands/pretrain_on_ImgNet32_embeddings.py - K-value ablations
exp_commands/ablate_test_k_vals_10K_Neurons_SplitCIFAR10_embeddings.py
NISPA
- NISPA entry point
NISPA/NISPA/main.py - NISPA scenarios
NISPA/NISPA/dataset.py - NISPA learner
NISPA/NISPA/NISPA/learner.py - Masked networks
NISPA/NISPA/NISPA/model.py - Connection growth
NISPA/NISPA/NISPA/grow.py - Learning utilities
NISPA/NISPA/NISPA/utils.py - Task training
NISPA/NISPA/NISPA/training.py - Results reporting
NISPA/NISPA/NISPA/report.py - Replay learner
NISPA/NISPA_Replay/NISPA/learner.py - Replay model
NISPA/NISPA_Replay/NISPA/model.py
Comparison baselines
- CLNP models
NISPA/Baselines/CLNP/models.py - EWC SI MAS agents
NISPA/Baselines/EWC_SI_MAS/agents/default.py - Experience replay agent
NISPA/Baselines/EWC_SI_MAS/agents/exp_replay.py - Baseline data loaders
NISPA/Baselines/EWC_SI_MAS/dataloaders/base.py - Baseline runner
NISPA/Baselines/EWC_SI_MAS/iBatchLearn.py - Single-task model
NISPA/Baselines/STL/model.py
FlyModel
- FlyModel loader
FlyModel/BI_R/load.py - FlyModel continual run
FlyModel/BI_R/main_cl.py - Facebook EWC
FlyModel/codes/EWC_facebook.py - GEM baseline
FlyModel/codes/GEM.py
Other components
- Researcher
Connections 17
- Researcher → Dataset setup: prepares data
README.md - Researcher → Continual-learning run: selects experiments
README.md - Dataset setup → Dataset parameters: prepares splits
- Parameter assembly → Model parameters: combines settings
py_scripts/combine_params.py - Parameter assembly → Dataset parameters: combines settings
py_scripts/combine_params.py - Model base → CL benchmarks: uses benchmarks
models/Base_Model.py - NISPA entry point → NISPA scenarios: loads scenarios
NISPA/NISPA/main.py - NISPA entry point → NISPA learner: creates learner
NISPA/NISPA/main.py - NISPA learner → Masked networks: uses networks
NISPA/NISPA/NISPA/learner.py - NISPA learner → Connection growth: grows connections
NISPA/NISPA/NISPA/learner.py - NISPA learner → Learning utilities: uses utilities
NISPA/NISPA/NISPA/learner.py - NISPA learner → Task training: runs training
NISPA/NISPA/NISPA/learner.py - NISPA learner → Results reporting: logs summaries
NISPA/NISPA/NISPA/learner.py - Masked networks → Learning utilities: shares masks
- Baseline runner → Baseline data loaders: uses loaders
NISPA/Baselines/EWC_SI_MAS/iBatchLearn.py - Baseline runner → EWC SI MAS agents: selects agents
NISPA/Baselines/EWC_SI_MAS/iBatchLearn.py - EWC SI MAS agents → CLNP models: imports models
NISPA/Baselines/EWC_SI_MAS/agents/default.py