Biologically Inspired Cognitive Architectures最新文献

In this paper we present the validation of the three-dimensional model of emotions by Hugo Lövheim the “cube of emotion” via neurosimulation in the NEST. We also present the extension of original “cube of emotion” with the bridge to computational processes parameters. The neurosimulation is done via re-implementation of DA, 5-HT and NA subsystems of a rat brain to replicate 8 basic psycho-emotional states according to the “cube of emotion”. Results of neurosimulations indicate the incremental influence of DA and NA over computational resources of a psycho-emotional state while 5-HT decreases the computational resources used to calculate a psycho-emotional state. This way we indicate the feasibility of the bio-plausible re-implementation of psycho-emotional states in a computational system. This approach could be useful extension of decision making and load balancing components of modern artificial agents as well as intelligent robotic systems.

This paper presents an adaptive neurologically inspired cognitive model for Major Depressive Disorder. It is based on an (adaptive) temporal-causal network modelling approach incorporating a dynamic perspective on mental states and causal relations. The adaptive network model addresses how a Deep Brain Stimulation treatment used for this disorder can work by a Hebbian learning effect.

Common Model of Cognition (CMC) is a collective attempt to develop a consensus on cognitive architectures. The model needs to be extended to include components and functions that are vital to achieving the goals of Humanlike AI, supporting humanlike learnability, social acceptability and humanlike creativity. Being biologically grounded, together these components will enable social-emotional character reasoning in artifacts and support emotionally-driven behavior generation. Historically, cognitive architectures originated from rule-based systems. Their main building block then evolved to a variety of structures, collectively called here schemas. While a schema is an overloaded term, in the field of biologically inspired cognitive architectures (BICA) it can be given a precise and useful meaning, allowing comparison of different models. Here one particular model is used as the main example: emotional BICA, or eBICA (Samsonovich, BICA, 2013) that extends GMU BICA (Samsonovich & De Jong, 2005) and supports human-like socially-emotional intelligence. This becomes possible with the help of so-called moral schemas. Their operation relies on semantic maps and contributes to the functioning of narrative networks. The present work documents the general formalism of schemas of eBICA, defines moral schemas, and explains their usage on examples. This framework is expected to enable a human-level believability and social compatibility in virtual actors and cobots across a variety of practically important domains and paradigms, thereby contributing to the expected breakthrough in humane artificial intelligence. Expected applications include virtual cobots-assistants and actors-partners in a broad spectrum of tasks. Forming a consensus on goals, paradigms, metrics and target applications for the new framework is equally important in understanding the overarching mission of solving the BICA Challenge.

This article describes the open-source cognitive multi-agent knowledge-based SOSIEL (Self-Organizing Social & Inductive Evolutionary Learning) Platform, designed for building the social components of social-ecological decision support systems, consisting of agents empowered with a cognitive architecture. The platform can simulate the cross-generational progression of one or a large number of agents that can interact among themselves and/or with coupled natural and/or technical systems, learn from their and each other’s experience, create new practices, and make decisions about taking and then take (potentially collective) actions. The platform can also be used for conducting hypothetical experiments that are focused on studying the interactions among: (a) cross-generational population dynamics, (b) self-organizing multi-layered social network structures, (c) evolving place-based knowledge, (d) learning, (e) decision-making, (f) collective action and its potential, and (g) social and (when coupled) social-ecological outcomes. The article describes a simple model that was built with the SOSIEL Platform, which simulates the co-evolution of mental models among socially learning agents.

In this paper, a brain-inspired fuzzy emotional neural network (FUZZ-ENN) is proposed for uncertainty prediction tasks in real world applications. In the proposed FUZZ-ENN, amygdala connections are modeled by fuzzy IF-THEN behavioral rules and orbitofrontal module inhibits the amygdala responses in order to decrease the uncertainty. This computational model is based on the inhibitory connections in the human emotional brain’s nervous system inhibiting the uncertainty. In this paper, genetic algorithm is applied for optimal tuning of crisp numerical and fuzzy parameters of the proposed model. A traditional neural model and a two layered emotional neural network (ENN) are also implemented for comparison purposes on the electrical load and wind power forecasting problem and the prediction of geomagnetic activity indices as two real world case studies. Numerical results indicate the superiority of the proposed approach in term of lower uncertainty in the prediction.

The current work develops the previous model of interaction between learning and evolution (Red’ko, 2017). The previous model investigated this interaction by means of computer simulation. The mechanisms of the main properties of the interaction between learning and evolution (the genetic assimilation, the hiding effect, the influence of the learning load on the interaction between learning and evolution) were analyzed. The results were obtained for the finite size of the population. Fortunately, there is the possibility to analyze the same effect analytically for the case of the infinite size of the population. The current article considers sufficiently large sizes of population. Computer simulation demonstrates that the essential results of the model do not depend on the population size if this size is sufficiently large. Moreover, at such large population size, the results of computer simulation actually coincide with the results of analytical estimations. We consider the processes of learning and evolution for the population of modeled organisms that have genotype and genotype. Genotypes are modified during evolution, phenotypes are optimized by means of learning. At the end of the generation, organisms are selected in accordance with their final phenotype. The main attention is paid to the hiding effect. This effect means that learning can suppress the evolutionary optimization of genotypes: the optimal phenotype can be found by means of learning for a rather large set of different genotypes, so there is no need to find the optimal genotype. The hiding effect is analyzed by both computer simulation and analytically.

Episodic memory plays important role in animal behavior. It allows to reuse general skills for solution of specific tasks in changing environment. This beneficial feature of biological cognitive systems is still not incorporated successfully in an artificial neural architectures. In this paper we propose a neural architecture with shared episodic memory for multi-task reinforcement learning (SEM-PAAC). This architecture extends Parallel Advantage Actor Critic (PAAC) with two recurrent sub-networks for separate tracking of environment and task states. The first subnetwork store episodic memory and the second one allows task specific execution of policy. Experiments in the Taxi domain demonstrated that SEM-PAAC has the same performance as PAAC when subtasks are solved separately. On the other hand when subtasks are solved jointly for completing full Taxi task SEM-PAAC is significantly better due to reuse of episodic memory. Proposed architecture also successfully learned to predict task completion. This is a step towards more autonomous agents for multitask problems.

Taking into account that the human intelligence is the only available intelligence we will find the functional relationship between neuronal processes and psychic phenomena to reproduce intelligence in artificial system. The autonomous behavior of an agent may be the consequence of a gap between physical processes and self-referential meaningful processing of information which is related but not determined by physical processes. This indeterminism can be reproduced in a cognitive architecture through the self-referential processing of information with consideration of itself as a meaningful model. We propose embodiment of cognitive architecture of autonomous intelligent agent as an artificial neural network with a feedback loop in meaningful processing of information.