FedCausalCompose
A causal world model framework designed for modular LLM agents, adopting a federated architecture. It identifies causal relationships across module interfaces through local intervention-response evidence, enabling participating modules to jointly reconstruct cross-module causal structures without sharing internal states.
Timeline (last 90 days)
The core idea of FedCausalCompose is to let local actions provide intervention-response evidence to identify interface relationships across modules.
The federated design of FedCausalCompose allows each module to expose only aggregated intervention-response evidence externally, rather than raw internal state-transition data.
The recovery quality of cross-module interfaces improves as intervention-response coverage increases.
Given sufficient coverage and controlled local mechanism error, oracle-level causal composition can surpass the theoretical lower bound achievable by non-causal methods.
All Facts (4)
The core idea of FedCausalCompose is to let local actions provide intervention-response evidence to identify interface relationships across modules.
50%UnverifiedThe federated design of FedCausalCompose allows each module to expose only aggregated intervention-response evidence externally, rather than raw internal state-transition data.
50%UnverifiedThe recovery quality of cross-module interfaces improves as intervention-response coverage increases.
50%UnverifiedGiven sufficient coverage and controlled local mechanism error, oracle-level causal composition can surpass the theoretical lower bound achievable by non-causal methods.
50%