Incremental learning of gestures for human–robot interaction

For a robot to cohabit with people, it should be able to learn people’s nonverbal social behavior from experience. In this paper, we propose a novel machine learning method for recognizing gestures used in interaction and communication. Our method enables robots to learn gestures incrementally during human–robot interaction in an unsupervised manner. It allows the user to leave the number and types of gestures undefined prior to the learning. The proposed method (HB-SOINN) is based on a self-organizing incremental neural network and the hidden Markov model. We have added an interactive learning mechanism to HB-SOINN to prevent a single cluster from running into a failure as a result of polysemy of being assigned more than one meaning. For example, a sentence: “Keep on going left slowly” has three meanings such as, “Keep on (1)”, “going left (2)”, “slowly (3)”. We experimentally tested the clustering performance of the proposed method against data obtained from measuring gestures using a motion capture device. The results show that the classification performance of HB-SOINN exceeds that of conventional clustering approaches. In addition, we have found that the interactive learning function improves the learning performance of HB-SOINN.     

Simultaneous cognitive origin of life and information

Shannon’s information quantity I(E) = log(1/P(E)) is defined under an assumption of the existence of a “cognitive subjective entity” capable of judging yes/no or occurred/not-occurred of an event E (which occurs with a probability P(E)). The final acceptor/user of information is a living individual, although first and/or intermediate sender(s) and/or acceptor(s) of information may be either living individual(s) or nonliving element(s) or man-made machine(s). Therefore we can conclude that information is a most essential character of living individuals, and that information and life must have emerged simultaneously as a “minimum cognitive system” (MCS). Since then, living individuals/lives must have evolved as “self-revising learning neural network machines” capable of “active evolution”. How MCS could have emerged was discussed.     

Organizing the Web's Information Explosion to Discover Unknown Unknowns

This paper introduces the TORISHIKI-KAI project, which aims to construct a million-word-scale semantic network from the Web using state of the art knowledge acquisition methods. The resulting network can be browsed as a Web search directory, and we show that the directory is useful for finding “unknown unknowns” — in the infamous words of D.H. Rumsfeld: things “we don't know we don't know.” Because typically we have no way to look for information we don't even know is missing, a crucial characteristic of unknown unknowns is that they are very difficult to discover through keyword-based Web search. Some examples of the unknown unknowns we have found include unexpected troubles associated with commercial products, surprising new combinations of ingredients in new recipes, unexpected tools or methods for commiting suicide, and so on. We expect such information to be useful for risk management, innovation support, and the detection of harmful information on the Web.     

Mental states as multi-context systems

Although Belief-Desire-Intention (BDI) agents have been deeply investigated from both a theoretical and a pragmatic perspective, less attention has been paid to the inherent recursive structure of mental states, which plays an essential role when modelling high level interaction between intelligent agents. This paper tries to capture this property by introducing a multi-context approach to the representation of mental states. A semantics for multi-context formalisms is provided based on the definition of “mental structure”, which is a hierarchical lattice of triangular modules <x,B,D>, where the component x represents the agent x’s mental state as a whole, while B and D represent specifically x’s beliefs and x’s desires. If other mental attitudes, as intention and commitment, are to be considers as primitives, then they can be embodied in the basic module, otherwise they can be represented in terms of beliefs and desires. The old notion of clause is rediscovered in order to facilitate the heavy automated theorem-proving necessary to exploit the potentiality of the formalism for the intelligent interaction with the external environment. The main advantages of this approach are the support for “unconsciousness” and the fact that inferences themselves can be modelled as mental attitudes. Some advanced dynamics of mental states, as the abductive revision of mental states after the reception of a communication, will easily be applied over this formalism.      

Entertainment capture through heart rate activity in physical interactive playgrounds

An approach for capturing and modeling individual entertainment (“fun”) preferences is applied to users of the innovative Playware playground, an interactive physical playground inspired by computer games, in this study. The goal is to construct, using representative statistics computed from children’s physiological signals, an estimator of the degree to which games provided by the playground engage the players. For this purpose children’s heart rate (HR) signals, and their expressed preferences of how much “fun” particular game variants are, are obtained from experiments using games implemented on the Playware playground. A comprehensive statistical analysis shows that children’s reported entertainment preferences correlate well with specific features of the HR signal. Neuro-evolution techniques combined with feature set selection methods permit the construction of user models that predict reported entertainment preferences given HR features. These models are expressed as artificial neural networks and are demonstrated and evaluated on two Playware games and two control tasks requiring physical activity. The best network is able to correctly match expressed preferences in 64% of cases on previously unseen data (p−value 6 · 10⁻⁵). The generality of the methodology, its limitations, its usability as a real-time feedback mechanism for entertainment augmentation and as a validation tool are discussed.     

To hear—to say: the mediating presence of the healing witness

Illness and trauma challenge self-narratives. Traumatized individuals, unable to speak about their experiences, suffer in isolation. In this paper, I explore Kristeva’s theories of the speaking subject and signification, with its symbolic and semiotic modalities, to understand how a person comes to speak the unspeakable. In discussing the origin of the speaking subject, Kristeva employs Plato’s chora (related to choreo, “to make room for”). The chora reflects the mother’s preparation of the child’s entry into language and forms an interior darkroom, the reservoir of lived experience, from which self-narratives issue. Unable to speak of their suffering, traumatized individuals need someone to help them make room for a time of remembrance, someone who is a willing and capable listener. I call such a person a healing witness. Through the mediating presence of the healing witness, fragmented memories of trauma are recreated and incorporated into self-narratives that are sharable with others. Unfortunately, opportunities for witnessing are vanishing. In the last section, I examine the failure of modern media and communication technologies to bear (“hold,” “carry,” “transport”) acts of witnessing. I argue that they perturb the semiotic. According to Kristeva, meaning arises from the dialectical tension between the semiotic (drives and affects) and the symbolic (logic and rules) and is threatened by arid discourse, psychosomatic illnesses, and outbreaks of violence when the semiotic is not represented. Unless we open technology to the imaginary, we risk losing the capacity to bear witness to one another and to create narratives and connections that are meaningful.     

Why Computers Can’t Feel Pain

The most cursory examination of the history of artificial intelligence highlights numerous egregious claims of its researchers, especially in relation to a populist form of ‘strong’ computationalism which holds that any suitably programmed computer instantiates genuine conscious mental states purely in virtue of carrying out a specific series of computations. The argument presented herein is a simple development of that originally presented in Putnam’s (Representation & Reality, Bradford Books, Cambridge in 1988) monograph, “Representation & Reality”, which if correct, has important implications for turing machine functionalism and the prospect of ‘conscious’ machines. In the paper, instead of seeking to develop Putnam’s claim that, “everything implements every finite state automata”, I will try to establish the weaker result that, “everything implements the specific machine Q on a particular input set (x)”. Then, equating Q (x) to any putative AI program, I will show that conceding the ‘strong AI’ thesis for Q (crediting it with mental states and consciousness) opens the door to a vicious form of panpsychism whereby all open systems, (e.g. grass, rocks etc.), must instantiate conscious experience and hence that disembodied minds lurk everywhere.     

A Neural Network Scheme for Long-Term Forecasting of Chaotic Time Series

The accuracy of a model to forecast a time series diminishes as the prediction horizon increases, in particular when the prediction is carried out recursively. Such decay is faster when the model is built using data generated by highly dynamic or chaotic systems. This paper presents a topology and training scheme for a novel artificial neural network, named “Hybrid-connected Complex Neural Network” (HCNN), which is able to capture the dynamics embedded in chaotic time series and to predict long horizons of such series. HCNN is composed of small recurrent neural networks, inserted in a structure made of feed-forward and recurrent connections and trained in several stages using the algorithm back-propagation through time (BPTT). In experiments using a Mackey-Glass time series and an electrocardiogram (ECG) as training signals, HCNN was able to output stable chaotic signals, oscillating for periods as long as four times the size of the training signals. The largest local Lyapunov Exponent (LE) of predicted signals was positive (an evidence of chaos), and similar to the LE calculated over the training signals. The magnitudes of peaks in the ECG signal were not accurately predicted, but the predicted signal was similar to the ECG in the rest of its structure.     

An Algorithm for Constructing 3D Struts

A simple robust"strut algorithm"is presented which, when given a graph embedded in 3D space, thickens its edges into solid struts. Various applications, crystallographic and sculptural, are shown in which smooth high-genus forms are the output. A toolbox of algorithmic techniques allow for a variety of novel, visually engaging forms that express a mathematical aesthetic. In sculptural examples, hyperbolic tessellations in the Poincaréplane are transformed in several ways to three-dimensional networks of ...     

On the Stability of Dynamic Diffusion Load Balancing

We consider the problem of dynamic load balancing in arbitrary (connected) networks on n nodes. Our load generation model is such that during each round, n tasks are generated on arbitrary nodes, and then (possibly after some balancing) one task is deleted from every non-empty node. Notice that this model fully saturates the resources of the network in the sense that we generate just as many new tasks per round as the network is able to delete. We show that even in this situation the system is stable, in that the total load remains bounded (as a function of n alone) over time. Our proof only requires that the underlying “communication” graph be connected. (It of course also works if we generate less than n new tasks per round, but the major contribution of this paper is the fully saturated case.) We further show that the upper bound we obtain is asymptotically tight (up to a moderate multiplicative constant) by demonstrating a corresponding lower bound on the system load for the particular example of a linear array (or path). We also show some simple negative results (i.e., instability) for work-stealing based diffusion-type algorithms in this setting. A preliminary version of this paper entitled “Dynamic diffusion load balancing” was published in Proc. 32nd International Colloquium on Automata, Languages and Programming (ICALP’05), Lecture Notes in Computer Science 3580, Springer-Verlag, pp. 1386–1398. P. Berenbrink is supported by NSERC Discovery Grant 250284-2002. R. Martin is supported by EPSRC grant “Discontinuous Behaviour in the Complexity of Randomized Algorithms.”     

Giotto and Renaissance Perspective

A careful geometric examination of the blind arcades (coretti) depicted in Giotto’s fresco on the choir wall in the Arena Chapel in Padua shows that they were designed and painted according to the rules of what I term “progressive costruzione legittima” and thus represent simulations of visual images. Because no images of this type have come down from Classical Antiquity and because the literary references remain silent in this respect, the coretti must be considered, according to today’s knowledge, the oldest monuments manifesting the application of the costruzione legittima. This means the history of the central (linear) perspective must be rewritten. In any case it was not a Renaissance invention. I expressly agree with the researchers who see Giotto’s painting in conjunction with the findings of the Scholastic “optics specialists” (such as Grosseteste, Witelo, Bacon), who all stood with their feet firmly planted on the ground of Euclid’s rigidly geometrically conceived visual theory and its Arab commentators.     

Attractor switching by neural control of chaotic neurodynamics

Chaotic attractors of discrete-time neural networks include infinitely many unstable periodic orbits, which can be stabilized by small parameter changes in a feedback control. Here we explore the control of unstable periodic orbits in a chaotic neural network with only two neurons. Analytically, a local control algorithm is derived on the basis of least squares minimization of the future deviations between actual system states and the desired orbit. This delayed control allows a consistent neural implementation, i.e. the same types of neurons are used for chaotic and controlling modules. The control signal is realized with one layer of neurons, allowing selective switching between different stabilized periodic orbits. For chaotic modules with noise, random switching between different periodic orbits is observed.     

Network invariants for real-time systems

We extend the approach of model checking parameterized networks of processes by means of network invariants to the setting of real-time systems. We introduce timed transition structures (which are similar in spirit to timed automata) and define a notion of abstraction that is safe with respect to linear temporal properties. We strengthen the notion of abstraction to allow a finite system, then called network invariant, to be an abstraction of networks of real-time systems. In general the problem of checking abstraction of real-time systems is undecidable. Hence, we provide sufficient criteria, which can be checked automatically, to conclude that one system is an abstraction of a concrete one. Our method is based on timed superposition and discretization of timed systems. We exemplify our approach by proving mutual exclusion of a simple protocol inspired by Fischer’s protocol, using the model checker TLV. Part of this work was done during O. Grinchtein’s stay at Weizmann Institute. This author was supported by the European Research Training Network “Games”.     

Rudolph M. Schindler—Proportion,Scale and the “Row”

Jin-Ho Park interprets Schindler’s “reference frames in space” as set forth in his 1916 lecture note on mathematics, proportion, and architecture, in the context of Robinson’s1898–99 articles in the Architectural Record. Schindler’s unpublished, handwritten notes provide a source for his concern for “rhythmic” dimensioning in architecture. He uses a system in which rectangular dimensions are arranged in “rows.” Architectural examples of Schindler’s Shampay, Braxton-Shore and How Houses illustrate the principles.     

Dynamic Analysis of Stochastic Recurrent Neural Networks

This paper addresses the issue of pth moment exponential stability of stochastic recurrent neural networks (SRNN) with time-varying interconnections and delays. With the help of the Dini derivative of the expectation of V(t, X(t)) “along” the solution X(t) of the model and the technique of Halanay-type inequality, some novel sufficient conditions on pth moment exponential stability of the trivial solution has been established. Conclusions of the development as presented in this paper have gone beyond some published results and are helpful to design stability of networks when stochastic noise is taken into consideration. An example is also given to illustrate the effectiveness of our results.     

Charting the behavioural state of a person using a backpropagation neural network

This paper describes the application of a backpropagation artificial neural network (ANN) for charting the behavioural state of previously unseen persons. In a simulated theft scenario participants stole or did not steal some money and were interviewed about the location of the money. A video of each interview was presented to an automatic system, which collected vectors containing nonverbal behaviour data. Each vector represented a participant’s nonverbal behaviour related to “deception” or “truth” for a short period of time. These vectors were used for training and testing a backpropagation ANN which was subsequently used for charting the behavioural state of previously unseen participants. Although behaviour related to “deception” or “truth” is charted the same strategy can be used to chart different psychological states over time and can be tuned to particular situations, environments and applications. We thank those who kindly volunteered to participate in the study.     

Beyond Mind: How Brains Make up Artificial Cognitive Systems

What I call semiotic brains are brains that make up a series of signs and that are engaged in making or manifesting or reacting to a series of signs: through this semiotic activity they are at the same time engaged in “being minds” and so in thinking intelligently. An important effect of this semiotic activity of brains is a continuous process of disembodiment of mind that exhibits a new cognitive perspective on the mechanisms underling the semiotic emergence of meaning processes. Indeed at the roots of sophisticated thinking abilities there is a process of disembodiment of mind that presents a new cognitive perspective on the role of external models, representations, and various semiotic materials. Taking advantage of Turing’s comparison between “unorganized” brains and “logical” and “practical” machines” this paper illustrates the centrality to cognition of the disembodiment of mind from the point of view of the interplay between internal and external representations, both mimetic and creative. The last part of the paper describes the concept of mimetic mind I have introduced to shed new cognitive and philosophical light on the role of computational modeling and on the decline of the so-called Cartesian computationalism.     

Functional Topology Classification of Biological Computing Networks

Current analyses of complex biological networks focus either on their global statistical connectivity properties (e.g. topological path lengths and nodes connectivity ranks) or the statistics of specific local connectivity circuits (motifs). Here we present a different approach – Functional Topology, to enable identification of hidden topological and geometrical fingerprints of biological computing networks that afford their functioning – the form-function fingerprints. To do so we represent the network structure in terms of three matrices: 1. Topological connectivity matrix – each row (i) is the shortest topological path lengths of node i with all other nodes; 2. Topological correlation matrix – the element (i,j) is the correlation between the topological connectivity of nodes (i) and (j); and 3. Weighted graph matrix – in this case the links represent the conductance between nodes that can be simply one over the geometrical length, the synaptic strengths in case of neural networks or other quantity that represents the strengths of the connections. Various methods (e.g. clustering algorithms, random matrix theory, eigenvalues spectrum etc.), can be used to analyze these matrices, here we use the newly developed functional holography approach which is based on clustering of the matrices following their collective normalization. We illustrate the approach by analyzing networks of different topological and geometrical properties: 1. Artificial networks, including – random, regular 4-fold and 5-fold lattice and a tree-like structure; 2. Cultured neural networks: A single network and a network composed of three linked sub-networks; and 3. Model neural network composed of two overlapping sub-networks. Using these special networks, we demonstrate the method’s ability to reveal functional topology features of the networks.     

Database Integration Using Neural Networks: Implementation and Experiences

Applications in a wide variety of industries require access to multiple heterogeneous distributed databases. One step in heterogeneous database integration is semantic integration: identifying corresponding attributes in different databases that represent the same real world concept. The rules of semantic integration can not be ‘pre-programmed’ since the information to be accessed is heterogeneous and attribute correspondences could be fuzzy. Manually comparing all possible pairs of attributes is an unreasonably large task. We have applied artificial neural networks (ANNs) to this problem. Metadata describing attributes is automatically extracted from a database to represent their ‘signatures’. The metadata is used to train neural networks to find similar patterns of metadata describing corresponding attributes from other databases. In our system, the rules to determine corresponding attributes are discovered through machine learning. This paper describes how we applied neural network techniques in a database integration problem and how we represent an attribute with its metadata as discriminators. This paper focuses on our experiments on effectiveness of neural networks and each discriminator. We also discuss difficulties of using neural networks for this problem and our wish list for the Machine Learning community. Received 18 February 1999 / Revised 22 April 1999 / Accepted in revised form 20 November 1999     

The Method of Levels of Abstraction

The use of “levels of abstraction” in philosophical analysis (levelism) has recently come under attack. In this paper, I argue that a refined version of epistemological levelism should be retained as a fundamental method, called the method of levels of abstraction. After a brief introduction, in section “Some Definitions and Preliminary Examples” the nature and applicability of the epistemological method of levels of abstraction is clarified. In section “A Classic Application of the Method of Abstraction”, the philosophical fruitfulness of the new method is shown by using Kant’s classic discussion of the “antinomies of pure reason” as an example. In section “The Philosophy of the Method of Abstraction”, the method is further specified and supported by distinguishing it from three other forms of “levelism”: (i) levels of organisation; (ii) levels of explanation and (iii) conceptual schemes. In that context, the problems of relativism and antirealism are also briefly addressed. The conclusion discusses some of the work that lies ahead, two potential limitations of the method and some results that have already been obtained by applying the method to some long-standing philosophical problems.