01Theory理論
Predictive coding needs a memory預測編碼需要記憶
Predictive coding supplies a generative hierarchy; modern Hopfield networks supply content-addressable retrieval. A Tweedie–Hopfield correspondence shows that both rely on the same Bayesian denoiser. HOPE combines hierarchical inference, temporal prediction, and layer-wise memory in one energy trained with local Hebbian updates.預測編碼提供生成式階層，現代 Hopfield 網路提供內容定址檢索。Tweedie–Hopfield 對應關係顯示，兩者依賴同一個貝氏去噪器。HOPE 以單一能量函數整合階層推論、時間預測與逐層記憶，並以局部 Hebbian 更新學習。