Directory Contents
# Adam Research & Advanced Architectures This directory contains experimental and advanced implementations of next-generation financial technology concepts, integrated into the Adam system. ## 1. Federated Learning (`core/research/federated_learning/`) **Context:** Privacy-preserving distributed training for credit risk models across multiple institutions. **Implementation:** - **Coordinator:** `fl_coordinator.py` implements a standard `FedAvg` (Federated Averaging) algorithm. - **Client:** `fl_client.py` simulates individual banks with private local data (synthetic credit profiles). - **Model:** `model.py` defines a shared PyTorch Neural Network for credit scoring. **Usage:** ```python from core.research.federated_learning.fl_coordinator import FederatedCoordinator coordinator = FederatedCoordinator(num_clients=5) coordinator.run_round(1) ``` ## 2. Graph Neural Networks (`core/research/gnn/`) **Context:** Deep learning on the `UnifiedKnowledgeGraph` to detect systemic risks and hidden contagion paths. **Implementation:** - **Engine:** `engine.py` bridges the Knowledge Graph (NetworkX) with PyTorch, building adjacency matrices and node features. - **Layer:** `layers.py` implements a custom Graph Convolutional Layer (GCN) supporting sparse matrix operations. - **Model:** `model.py` defines a multi-layer GCN for node classification/risk scoring. **Usage:** ```python from core.research.gnn.engine import GraphRiskEngine engine = GraphRiskEngine() risk_scores = engine.predict_risk() ``` ## 3. One-Shot World Models (`core/research/oswm/`) **Context:** Model-Based Reinforcement Learning using Transformers to predict market dynamics and generate counterfactual scenarios via "One-Shot" in-context learning. **Implementation:** - **Inference:** `inference.py` manages the "Pre-training" on synthetic physics priors and "In-Context" generation on real market data. - **Transformer:** `transformer.py` implements a causal Transformer architecture for sequence modeling. **Usage:** ```python from core.research.oswm.inference import OSWMInference oswm = OSWMInference() oswm.pretrain_on_synthetic_prior() prediction = oswm.generate_scenario(market_context) ``` ## Integration Run the demo script to see all modules in action: ```bash python scripts/run_research_demo.py ``` Output is saved to `showcase/data/research_output.json`.