How jxsh341/artificial-learning-system works
Artificial Learning System is a research platform for controlled experiments on hybrid graph-and-temporal neural models. The main workflow is to prepare graph tasks, run cue-learning or graph-classification episodes through an environment and trainer, then evaluate learning, retention, reversal, generalization, and state-mediated computation. Experiments also probe ablations and state-to-readout coupling. The graph preserves the documented benchmark and diagnostic branches; some runner relationships are supported by the source index rather than inspected implementation excerpts.
Main entry points: schema.py GraphData interface · graph_classification.py Graph classification probes · runner.py Experiment runner · gnn_ssm.py Hybrid model
Components 21
Graph data
- Data adapters
src/artificial_system/data/adapters.py - Graph task generation
src/artificial_system/data/graph_tasks.py - GraphData interface
src/artificial_system/data/schema.py - Graph perturbations
src/artificial_system/data/perturbations.py
Task environments
- Cue environment
src/artificial_system/environment/cue_environment.py - Graph environment
src/artificial_system/environment/graph_environment.py
Hybrid model
- GNN encoder
src/artificial_system/models/gnn.py - State-space model
src/artificial_system/models/ssm.py - Pooling and readout
src/artificial_system/models/readout.py - Hybrid model
src/artificial_system/models/gnn_ssm.py
Training and experiments
- Episode trainer
src/artificial_system/training/trainer.py - Training losses
src/artificial_system/training/losses.py - Experiment runner
src/artificial_system/experiments/runner.py
Evaluation and diagnostics
- Cue-task evaluation
src/artificial_system/evaluation/metrics.py - Graph classification probes
src/artificial_system/evaluation/graph_classification.py - State-content analysis
src/artificial_system/evaluation/state_content.py - Readout geometry
src/artificial_system/evaluation/readout_geometry.py - Coupling experiments
src/artificial_system/experiments/coupling_rescue.py - Oscillation diagnostics
scripts/run_oscillation_mechanism.py - State-budget experiments
scripts/run_state_budget.py
Other components
- Researcher
Connections 29
- Researcher → Experiment runner: configures runs
- Data adapters → GraphData interface: creates data
src/artificial_system/data/adapters.py - Graph task generation → GraphData interface: builds graphs
src/artificial_system/data/graph_tasks.py - Graph perturbations → GraphData interface: transforms graphs
src/artificial_system/data/perturbations.py - Cue environment → GraphData interface: emits observations
src/artificial_system/environment/cue_environment.py - Graph environment → Graph task generation: uses datasets
src/artificial_system/environment/graph_environment.py - Graph environment → GraphData interface: emits observations
src/artificial_system/environment/graph_environment.py - GNN encoder → GraphData interface: encodes graphs
src/artificial_system/models/gnn.py - Hybrid model → GNN encoder: encodes nodes
src/artificial_system/models/gnn_ssm.py - Hybrid model → Pooling and readout: pools and predicts
src/artificial_system/models/gnn_ssm.py - Hybrid model → State-space model: updates state
src/artificial_system/models/gnn_ssm.py - Hybrid model → GraphData interface: accepts GraphData
src/artificial_system/models/gnn_ssm.py - Episode trainer → Graph environment: resets environment
src/artificial_system/training/trainer.py - Episode trainer → Hybrid model: steps model
src/artificial_system/training/trainer.py - Episode trainer → Training losses: computes loss
src/artificial_system/training/trainer.py - Experiment runner → Cue environment: configures benchmark
src/artificial_system/experiments/runner.py - Experiment runner → Graph environment: configures benchmark
src/artificial_system/experiments/runner.py - Experiment runner → Graph classification probes: runs probes
src/artificial_system/experiments/runner.py - Experiment runner → Cue-task evaluation: runs measures
src/artificial_system/experiments/runner.py - Graph classification probes → Graph task generation: evaluates instances
src/artificial_system/evaluation/graph_classification.py - Graph classification probes → Graph perturbations: tests controls
src/artificial_system/evaluation/graph_classification.py - State-content analysis → Graph classification probes: uses graph probes
src/artificial_system/evaluation/state_content.py - Readout geometry → Graph classification probes: analyzes probes
src/artificial_system/evaluation/readout_geometry.py - Coupling experiments → Graph classification probes: evaluates checkpoints
src/artificial_system/experiments/coupling_rescue.py - Coupling experiments → State-content analysis: measures state
src/artificial_system/experiments/coupling_rescue.py - Oscillation diagnostics → State-content analysis: measures state
scripts/run_oscillation_mechanism.py - Oscillation diagnostics → Coupling experiments: uses experiments
scripts/run_oscillation_mechanism.py - State-budget experiments → State-content analysis: measures state
scripts/run_state_budget.py - State-budget experiments → Coupling experiments: uses experiment tools
scripts/run_state_budget.py