previous day
next day
all days

View: session overviewtalk overview

10:00-10:30Coffee Break
10:30-12:30 Session 9A: BICA agents and systems

12:15-12:30 - discussion

Location: U311
On the perceptual advantages of visual suppression mechanisms for dynamic robot systems
SPEAKER: João Avelino

ABSTRACT. The use of computer vision based methods to explore the surrounding environment and track individuals, is further enhanced by the ability to move the visual system, a process realized by biological organisms. However, several perceptual issues arise from the rapid movement of the image acquisition system: blurred images and visual-proprioceptive transient delays result in incorrect spatial location estimation. Inspired by the biological mechanism of saccadic suppression, our main contribution is a biologically inspired visual stability mechanism able to deal with problems arising from self-motion. Together with a state-of-the-art pedestrian detection algorithm, our proposed methodology contributes to enhancements in person position estimation, thus improving human-robot interaction behaviors.

Cartesian Abstraction

ABSTRACT. It has been long debated how the so called cognitive map develops in rat hippocampus. The question is relevant since the hippocampus is the key component of the medial temporal lobe memory system, responsible for forming episodic memory and, in humans, also for forming declarative memory, the memory for facts and rules that serves cognition. There is a large number of models spanning from functional to the neuronal level modeling. Here, a novel idea is put forth; we suggest that the cognitive map is a non-linear projection of the egocentric observations to a Cartesian allothetic factor, i.e., to the laboratory coordinate system. This assumption gives rise to a computational deep learning model that can produce place cells. We show numerical results and discuss some emerging features, including the comparator system model of Vinogradova

Computational Locality in HaveNWant Semantic Networks
SPEAKER: Allen King

ABSTRACT. HaveNWant is a low-level cognitive schemata that processes bits of information bidirectionally. It is a common cortical algorithm whose Tinker-Toy® like parts can construct networks that react powerfully in embedded environments. They exhibit many of the animal learning abilities described by Piaget. Learning occurs when unknowns are detected, triggering the addition of new elements to the network. In this way, forward models can be built from experience, and many of these models can be linked together to form large distributed associative memories; HaveNWant’s level of abstraction lies well above neurological models, focusing on functionality and avoiding biological constraints. It also lies above computer AND and OR gates, which operate unidirectionally. In HaveNWant, for every signal going one way, there is another signal coming back. HaveNWant atoms continually reconcile the information on their links, each imposing a particular constraint. Its networks aggregate many single links, to efficiently enforce large sophisticated relationships. HaveNWant operates below most AI architectures, which have algorithms that do not constrain bit-level computational locality. Although the examples given involve toy networks, we have a plan to extend the base algorithms by adding dynamically learned variables and noise tolerance, to produce robust behavior.

Towards a Cognitive Multi-Agent System for Building Control
SPEAKER: Samer Schaat

ABSTRACT. Fitting a bionically inspired cognitive model to a technical application domain is a challenging task. This work explores the initial steps in applying the SiMA-based cognitive model ECABA to the do-main of building automation. The focus of the work is a prototypical proof of concept for applying the human inspired mecha-nisms of drives, and social rules when modeling a cognitive control system for a non-human body. We illustrate the flexibility and extensibility of the distributed cognitive ap-proach by presenting the iterative development of a simple reactive system with different compo-nents. By means of a simple use case, we show how the afore mentioned mechanisms can be imple-mented in a multi-agent system to provide a reactive-based control system.

SiMA-C: A Foundational Mental Architecture
SPEAKER: Samer Schaat

ABSTRACT. Cognitive architectures as frameworks for integrated computational models of the mind often focus on human reasoning capabilities, and sometimes are extended by theories of emotion. The SiMA-C approach starts with low-level mental processing and bases high-level process on its low-level foundation. With the aim of developing a unified model with functional equivalence to the human mind, descriptive concepts of perception, motivation, emotion, and cognition are translated into a functional model of data activation, valuation, mediation, and evaluation. The resulting SiMA-C mental architecture solves the problem of adapting an agent's decision on the current internal state (demands from bodily needs and norms) and the external environment (affordances) and hence mediates between them. Using the SiMA-C model in simulations of environmental-friendly consumer decisions gives an example of a concrete application.

A Case-Driven Methodology for the Interdisciplinary Development and Examination of Mental Architectures
SPEAKER: Samer Schaat

ABSTRACT. This paper provides a theoretical specification of a case-driven methodology to develop and examine mental architectures\footnote{Model details and simulation results are described in \cite{Schaat2016} and \cite{Schaat2014c}}. After analyzing suitable methods from science and engineering and arguing for their combination, in particular a Lakatosian approach to science and computer simulation for the validation of mental architectures, criteria for a methodology to develop mental architectures are justified and the methodology's instruments are described. The challenges of regular interdisciplinary collaboration are tackled by combining casuistry, use-case driven software engineering, and simulation in a novel way. A shared representation of the problem space by exemplary cases supports requirements analysis. By providing such common platform that enables perspectives for different disciplines, the case-driven methodology is able to support interdisciplinary knowledge translation. For requirement and model specification use-case inspired methods are used. To validate and examine mental architectures agent-based simulation is proposed, enabling us to test the specified assumptions and the model's plausibility.

Design of Neuromorphic Cognitive Module based on Hierarchical Temporal Memory and demonstrated on Anomaly Detection
SPEAKER: Marek Otahal

ABSTRACT. Our presented idea is to integrate artificial neural network (probably of BICA type) with a real biological network (ideally in the future with the human brain) in order to extend or enhance cognitive- and sensory- capabilities (e.g. by associating existing and artificial sensory inputs). We propose to design such neuro-module using Hierarchical Temporal Memory (HTM) which is a biologically-inspired model of the mammalian neocortex. A complex task of contextual anomaly detection was chosen as our case-study, where we evaluate capabilities of a HTM module on a specifically designed synthetic dataset and propose improvements to the anomaly model. HTM is framed within other common AI/ML approaches and we conclude that HTM is a plausible and useful model for designing a direct brain-extension module and draft a design of a neuromorphic interface for processing asynchronous inputs. Outcome of this study is the practical evaluation of HTM’s capabilities on the designed synthetic anomaly dataset, a review of problems of the HTM theory and the current implementation, extended with suggested interesting research direction for the future.

12:30-14:00Lunch Break
15:00-15:30Coffee Break
15:30-17:30 Session 12A: From the brain to advanced BICA

17:15-17:30 - discussion

Location: UL102
Critical Branching Neural Computation

ABSTRACT. Complex brains use spikes for information processing on the timescales of milliseconds to seconds. For spikes to be computationally useful, their propagation must be regulated to avoid overly damped and overly amplified dynamical regimes. The homeostatic state of balanced propagation is known as critical branching. Kello and colleagues have developed a local, general algorithm to achieve global critical branching in virtually any kind of spiking network. The algorithm accounts for evidence of critical branching in the form of power laws, and it maximizes the computational capacity of recurrent spiking networks as measured using reservoir computing techniques. The algorithm is also general enough to be integrated with other mechanisms of learning and plasticity, such as Hebbian learning, spike timing dependent plasticity, and reinforcement learning. Critical branching and learning are demonstrated in a simple classification task, and the more complex task of learning to generate speech acoustics from recordings of conversations.

The Individuation of Social Systems: A Cognitive Framework

ABSTRACT. We present a socio-human cognitive framework that radically deemphasizes the role of individual human agents required for both the formation of social systems and their ongoing operation thereafter. Our point of departure is Simondon’s (1992) theory of individuation, which we integrate with the enactive theory of cognition (Di Paolo et al., 2010) and Luhmann’s (1996) theory of social systems. This forges a novel view of social systems as complex, individuating sequences of communicative interactions that together constitute distributed yet distinct cognitive agencies, acquiring a capacity to exert influence over their human-constituted environment. We conclude that the resulting framework suggests several different paths of integrating AI agents into human society. One path suggests the emulation of a largely simplified version of the human mind, reduced in its functions to a specific triple selection-making which is necessary for sustaining social systems. Another one conceives AI systems that follow the distributed, autonomous architecture of social systems, instead that of humans.

A Formal Model of Script Construction Based on Salience and Abstraction
SPEAKER: Carlos León

ABSTRACT. The role of schemas seems to be crucial in general human cognition. Scripts, as a kind of cognitive schema, seem to be a good model for describing human behavior, and have been applied as formal elements to create computational models. However, a formal model of how scripts are created frow external stimuli has not been proposed yet. This paper proposes a computational process of script acquisition based on activation of perceivable properties. The activation is used to find relations between the abstracted properties and to filter and group stimuli into specific schemas. The output of an implementation of the computational model has been tested against the result of experiments with humans. Overall results indicate that the model is plausible and can be used to describe a number of phenomena related to script acquisition.

Automatic Case Generation to Enable Advanced Behaviors in Agents

ABSTRACT. Intelligent organism evolved beyond hardwired reflexes to acquiring the ability to create soft wired reflexes and to weave them into increasingly complex behavior. The GPME is a generic agent that emulates this process. GPME enhances the function of host agents by enabling them to develop and apply advanced behaviors. In this paper, we provide the results of a simple prototype that demonstrates the subset of GPME algorithms that are used to identify host behaviors from a time-series of perceptions about host observations and host actions.

Learning Task Goals Interactively with Visual Demonstrations
SPEAKER: James Kirk

ABSTRACT. Humans are extremely good at quickly teaching and learning new tasks through situated instructions; tasks such as learning a novel game or household chore. From studying such instructional interactions, we have observed that humans excel at communicating information through multiple modalities, including visual,linguistic, and physical ones. Rosie is a tabletop robot implemented in the Soar architecture that learns new tasks from online interactive language instruction. In the past, the features of each task's goal were explicitly described by a human instructor through language. In this work, we develop and study additional techniques for learning representations of goals. For game tasks, the agent can be given visual demonstrations of goal states, refined by human instructions. For procedural tasks,the agent uses information derived from task execution to determine which state features must be included in its goal representations. Using both approaches, Rosie learns correct goal representations from a single goal example or task execution across multiple games, puzzles, and procedural tasks. As expected, in most cases, the number of words required to teach the task is reduced when visual goal demonstrations are used. We also identify shortcomings of our approach and outline future research.

Continuous Phone Recognition in the Sigma Cognitive Architecture

ABSTRACT. Spoken language processing is an important capability of human intelligence that has hitherto been unexplored by cognitive architectures. This reflects on both the symbolic and sub-symbolic nature of the speech problem, and the capabilities provided by cognitive architectures to model the latter and its rich interplay with the former. Sigma has been designed to leverage the state-of-the-art hybrid (discrete + continuous) mixed (symbolic + probabilistic) capability of graphical models to provide in a uniform non- modular fashion effective forms of, and integration across, both cognitive and sub-cognitive behavior. In this article, previous work on speaker dependent isolated word recognition has been extended to demonstrate Sigma’s feasibility to process a stream of fluent audio and recognize phones, in an online and incremental manner with speaker independence. Phone recognition is an important step in integrating spoken language processing into Sigma. This work also extends the acoustic front-end used in the previous work in service of speaker independence. All of the knowledge used in phone recognition was added supraarchitecturally – i.e. on top of the architecture – without requiring the addition of new mechanisms to the architecture.

Adaptive Modelling of Trauma: Development and Recovery of Patients

ABSTRACT. In this paper, a computational model is presented to simulate traumas, including their development, recovery, and the effect of group support. The model is built upon mechanisms known from cognitive and social neuroscience. Using the model, several scenarios were explored, considering both individual and multiple persons. The simulation results of the model were compared to a data-set on symptoms and recovery of traumatized patients. The obtained model enables simulation and analysis of group therapy and its effects on traumatized patients.

15:30-17:30 Session 12B: Motivations and evaluations

17:15-17:30 - discussion

Location: U311
A Cognitive-Affective Architecture for ECAs

ABSTRACT. The development of Embodied Conversational Agents (ECAs) involves a large number of challenges such as the modeling of cognitive and affective functions in order to achieve realism and believability in this type of intelligent agents. An approach to provide ECAs with capabilities for cognitive processing such as learning, decision making, planning, and perception has been the use of cognitive architectures.Moreover, the literature reports several affective models for the generation, classification, and management of emotions in ECAs. Nevertheless, there is a need of cognitive-affective architectures that address the problem of achieving natural interaction and realistic behavior in ECAs. In this paper, we discuss the state of the art on existing cognitive architectures, affective models, and ECAs, and propose a cognitive-affective architecture based on Soar and extended with an affective model inspired by ALMA. The proposed cognitive-affective architecture is designed to allow ECAs to include and take advantage of mechanisms such as reinforcement learning, episodic memory, and emotion management.

Modeling the Interaction of Emotion and Cognition in Autonomous Agents

ABSTRACT. A major goal in various fields has been the development of believable, intelligent, and social Autonomous Agents (AAs) whose behavior is influenced by affective signals. This endeavor has promoted the development of cognitive architectures for AAs that incorporate processes that imitate those of human cognition and emotions. However, there is still a need for appropriate environments in such agent architectures for the modeling of the interaction between emotional and cognitive components. In this paper, we address the following research question: “how to model the interaction of emotion and cognition in agent architectures so that AAs are able to generate consistent emotional states and display believable emotional behaviors”. We address this problem from the perspective of the development of Computational Models of Emotions (CMEs). In particular, we propose an integrative framework for constructing CMEs whose design is focused on two main aspects: (1) the modeling of the underlying mechanisms of emotions, and (2) the incorporation of input and output interfaces that facilitate the interaction between affective processes implemented in CMEs and cognitive processes implemented in agent architectures.

Sparse Holographic Graph Neuron
SPEAKER: Denis Kleyko

ABSTRACT. In this article, we present a modification of the recently proposed Holographic Graph Neuron architecture for memorizing patterns of generic sensor stimuli. The original approach represents patterns as dense binary vectors, where zeros and ones are equiprobable. The presented modification employs sparse binary distributed representations where the number of ones is significantly less than zeros. Sparse representations are more biologically plausible because activities of real neurons are sparse. Moreover, given the same vector dimensionality, sparse representations use memory more efficiently and allow faster operations. Performance in a set of tasks was studied comparing the original approach and the sparse version for three different sizes of dimensionality. The results show that the modified sparse Holographic Graph Neuron retains the accuracy of the original approach.

Recognizing permuted words with Vector Symbolic Architectures: "A Cmabirgde tset for mahcenis"
SPEAKER: Evgeny Osipov

ABSTRACT. This paper proposes a simple encoding scheme for words using principles of Vector Symbolic Architectures. The proposed encoding allows finding a valid word in the dictionary for a given permuted word (represented using the proposed approach) using only a single operation - calculation of Hamming distance to the distributed representations of valid words in the dictionary. The proposed encoding scheme can be used as an additional processing mechanism for models of word embedding, which also form vectors to represent the meanings of words, in order to match the distorted words in the text to the valid words in the dictionary.

Implementing and Growing a 'Seed' Safe/Moral Motivational System for 50PH14
SPEAKER: David Kelley

ABSTRACT. Will be entered shortly

An Analysis of the CHC model for Comparing Cognitive Architectures

ABSTRACT. There are many cognitive architectures available nowadays, and each architecture has its own different mechanisms. Therefore, we need to identify the advantages and disadvantages of these architectures in order to improve upon them. In this paper, we propose new metrics for comparing cognitive architectures based on the Cattell-Horn-Carroll (CHC) model, which is used in psychology to explain factors of intelligence. Here, we analyze factors of intelligence in the CHC model and interpret them as elements of a new cognitive architecture. Then, the CHC model is investigated with respect to \data" and \processing" to obtain a metric for each component. We present examples using Soar and LIDA to illustrate comparing different cognitive architectures and demonstrate the effectiveness of our approach.

Evaluation of Cognitive Architectures Inspired by Cognitive Biases

ABSTRACT. Cognitive architectures are frequently built to model naturally intelligent behavior. This aims on two primary goals: On one hand these architectures model human behavior in order to give a better understanding of the human thought process. On the other hand cognitive architectures are an approach of modeling artificial intelligence. Those two goals might be conflicting, as humans sometimes act irrationally e.g. because they were cognitively biased. In this work, we analyze on a theoretical level whether cognitive architectures are also biased. Therefore we first abstract more general behavior from cognitive fallacies. Then we evaluate for the architectures Clarion, LEABRA and LIDA to what extent they can be biased.