CausalMoE: A Billion-Scale Multimodal Foundation Model for Granger Causal Discovery with Pattern-Routed Heterogeneous Experts
Researchers have introduced CausalMoE, a large-scale multimodal foundation model designed for Granger Causal Discovery (GCD). This model addresses limitations in existing neural GCD methods by employing a Pattern-Routed Mixture of Heterogeneous Experts to dynamically identify temporal patterns and route data to specialized domain experts. CausalMoE also integrates Large Language Models (LLMs) and Vision-Language Models (VLMs) to align numerical signals with textual and visual priors, aiming to improve causal estimation in complex scenarios and establish a new state-of-the-art in both fully supervised and few-shot settings. AI
IMPACT Introduces a novel approach to causal discovery by integrating multimodal data and specialized experts, potentially improving analysis of complex temporal systems.