A Top-Down Approach for a Synthetic Autobiographical Memory System

Abstract

Abstract. Autobiographical memory (AM) refers to the organisation of one's experience into a coherent narrative. The exact neural mechanisms responsible for the manifestation of AM in humans are unknown. On the other hand, the field of psychology has provided us with useful understanding about the functionality of a bio-inspired synthetic AM (SAM) system, in a higher level of description. This paper is concerned with a top-down approach to SAM, where known components and organisation guide the architecture but the unknown details of each module are abstracted. By using Bayesian latent variable models we obtain a transparent SAM system with which we can interact in a structured way. This allows us to reveal the properties of specific sub-modules and map them to functionality observed in biological systems. The top-down approach can cope well with the high performance requirements of a bio-inspired cognitive system. This is demonstrated in experiments with faces data.

An integrated probabilistic framework for robot perception, learning and memory

“…This approach is inspired by previous studies on hippocampal memory models for navigation in simulated environments [10], [23]. The SAM model is capable to process data flows from multiple sensory modalities, and chunk them into episodes based on the detection of a new event, to improve the understanding of the changing surrounding environment.…”

“…Secondly, the inclusion of M anchors U ∈ ℜ M×D which are optimized to further compress the signal into a smaller set of variables (M ≪ N ) by being sufficient statistics for the Gaussian process mapping function: p(F|X) = p(F|U, X)p(U)dU. These anchor points are also referred to as "inducing points" [27] and their role in the memory model is further explained in [10].…”

“…The key component of our SAM framework is based on a family of Bayesian latent variable models collectively referred to as a Deep Gaussian process (DGP) [7], [8], [9]. The DGP offers a probabilistic framework for modelling complex data in (un/semi)supervised mode, which has been used as a platform for perception and learning [10], [11].…”

iCub Visual Memory Inspector: Visualising the iCub’s Thoughts

“…The SAM [2] system is a probabilistic framework which approximates functional requirements to Auto-biographical memory as have been identified in previous studies [1]. In detail, denote the N observed sensory data as D multidimensional vectors {y n } N n=1 , i.e.…”

“…The operation of the perceptual systems that provide input to event memory can be analogised to a deep learning process that identifies psychologically meaningful latent variable descriptions [2]. Instantaneous memories then correspond to points in this latent variable space and episodic memories to trajectories through this space.…”