A dynamic binding mechanism for retrieving and unifying complex predicate-logic knowledge Gadi Pinkas 1 , Priscila Lima 2 , Shimon Cohen 1 1 Center for Academic Studies, Israel, gadip@mla.ac.il; 2 Federal Rural University of Rio de Janeiro, Brazil, priscilamvl@ufrrj.br Abstract. We show how to encode, retrieve and process complex structures equivalent to First-Order Logic (FOL) formulae, with Artificial Neural Networks (ANNs) designed for energy-minimization. The solution constitutes a binding mechanism that uses a neural Working Memory (WM) and a long-term synaptic memory (LTM) that can store both procedural and declarative FOL- like Knowledge-Base (KB). Complex structures stored in LTM are retrieved into the WM only upon need, where they are further processed. The power of our binding mechanism is demonstrated on unification problems: as neurons are dynamically allocated from a pool, most generally unified structures emerge at equilibrium. The network’s size is O(n·k), where n is the size of the retrieved FOL structures and k is the size of the KB. The mechanism is fault-tolerant, as no fatal failures occur when random units fail. The paradigm can be used in a variety of applications, such as language processing, reasoning and planning. Keywords: Binding-Problem, Logic, Unification, Neural-Symbolic Inference 1 Introduction Human cognition processes and produces complex combinatorial structures using neural networks. The human neural circuitry is capable of retrieving knowledge stored in LTM, processing complex structures and reasoning with them, while being extremely robust to failure of neurons. Nevertheless, such systematic and dynamic creation of complex structures presents difficult challenges to theories of neurocognition [5,6], especially in neural modeling of high-level cognitive processes, such as reasoning and language processing. Jackendoff [9] stated 4 major challenges for cognitive neuroscience: The first is related to creating complex structures by binding together simple constituents. The second is related to using bound variables in different roles. The third concerns reasoning with multiple instances of the same knowledge item; and the fourth refers to the relationships between WM and LTM. The binding mechanism seems to be fundamental to neural modeling of high-level cognitive processes, as all challenges seem to depend on it. Our first motivation is to create a plausible neural model capable of complex language processing, reasoning and long-term storage, while offering solutions to the challenges mentioned. We concentrate on FOL because it’s the “lingua-franca” in AI and because its expressiveness and computational aspects are powerful and well understood. Unification is fundamental for FOL realization and demonstrates many of the challenges, mentioned above. A second motivation is to engineer a new, neural