Abstract:

Computational Long Term Memory Architecture is the suggested model's module name. Each of the three modules in the Architecture represents one of the three main categories of human long-term memory: computational semantics, episodic memory, and procedural memory. In order to aid the episodic and procedural modules and to make cognitive judgments, the semantic module is programmed to acquire knowledge about the semantics of various sensory domains. This component learns the semantics of phrases in natural language and translates them into episodic experiences that may serve as a prompt for conversation. A computational mechanism based on a grid and the place neuron is presented, enabling an artificial agent to localize itself in a known environment; this paves the way for the agent to navigate to complex tasks that necessitate learning the spatial semantics of objects for handling. The capabilities of episodic memory are mirrored in the proposed episodic module. This section utilizes the abstract event information, such as event activities and other contextual elements necessary for event encoding and episode generation that has been preprocessed. To save on storage, the episodic module uses a remembering process similar to the forgetting mechanism. By carefully crafting the forgetting decay function, we were able to reduce the rate of event miss relative to state-of-the-art methods like EM ART. The suggested model's third component is a procedural module meant to teach the user how to carry out certain actions. Together, it and the semantic module teach the meaning of actions. This component was developed to learn tasks via the sequential manipulation of object bodies. This module uses deep neural networks to learn the motor level activities that occur in the body in response to interactions with objects

Description:

Computational, Long-Term Memory Architecture, Mimic Human Behaviour