Tag
This proposal introduces the Dynamic Concept Graph (DCG), a hybrid cognitive architecture that combines neural representation learning, symbolic knowledge structures, multimodal perception, and analogical reasoning to provide persistent, evolving world models for AI, addressing limitations such as inconsistent reasoning and lack of causal understanding in large language models.
The paper introduces CO-ALIGN, a bias mitigation method for text-to-image diffusion models that aligns concept graphs in the text encoder and denoiser, achieving 30% fairness improvement and 11.4 FID gain while reducing incoherent outputs by 88%.