J. Mark Bishop

NAPSAC: High Noise, High Dimensional Robust Estimation - it's in the Bag.

An umber of the most powerful robust estimation algorithms, such as RANSAC, MINPRAN and LMS ,h ave their basis in selecting random minimal sets of data to instantiate hypotheses. However, their performance degrades in higher dimensional spaces due to the exponentially decreasing probability of sampling a set that is composed entirely of inliers. In order to overcome this, rather than picking sets at random, a new strategy is proposed that alters the way samples are taken, under the assumption that inliers will tend to be closer to one another than outliers. Based on this premise, the NAPSAC (N Adjacent Points SAmple Consensus) algorithm is derived and its performance is shown to be superior to RANSAC in both high noise and high dimensional spaces.

Convergence analysis of stochastic diffusion search

Abstract In this paper we present a connectionist searching technique - the Stochastic Diffusion Search (SDS), capable of rapidly locating a specified pattern in a noisy search space. In operation SDS finds the position of the pre-specified pattern or if it does not exist - its best instantiation in the search space. This is achieved via parallel exploration of the whole search space by an ensemble of agents searching in a competitive cooperative manner. We prove mathematically the convergence of stochastic diffusion search. SDS converges to a statistical equilibrium when it locates the best instantiation of the object in the search space. Experiments presented in this paper indicate the high robustness of SDS and show good scalability with problem size. The convergence characteristic of SDS makes it a fully adaptive algorithm and suggests applications in dynamically changing environments.

Small-World Effects in Lattice Stochastic Diffusion Search

Stochastic Diffusion Search is an efficient probabilistic bestfit search technique, capable of transformation invariant pattern matching. Although inherently parallel in operation it is difficult to implement efficiently in hardware as it requires full inter-agent connectivity. This paper describes a lattice implementation, which, while qualitatively retaining the properties of the original algorithm, restricts connectivity, enabling simpler implementation on parallel hardware. Diffusion times are examined for different network topologies, ranging from ordered lattices, over small-world networks to random graphs.

Dynamic Knowledge Representation in Connectionist Systems

One of the most pervading concepts underlying computational models of information processing in the brain is linear input integration of rate coded uni-variate information by neurons. After a suitable learning process this results in neuronal structures that statically represent knowledge as a vector of real valued synaptic weights. Although this general framework has contributed to the many successes of connectionism, in this paper we argue that for all but the most basic of cognitive processes, a more complex, multi-variate dynamic neural coding mechanism is required - knowledge should not be spacially bound to a particular neuron or group of neurons. We conclude the paper with discussion of a simple experiment that illustrates dynamic knowledge representation in a spiking neuron connectionist system.

Emotion and Anticipation in an Enactive Framework for Cognition (Response to Andy Clark)

Undeniably, anticipation plays a crucial role in cognition. By what means, to what extent, and what it achieves remain open questions. In a recent BBS target article, Andy Clark depicts an integrative model of the brain that builds on hierarchical Bayesian models of neural processing (Rao and Ballard, 1999; Friston, 2005; Brown et al., 2011), and their most recent formulation using the free-energy principle borrowed from thermodynamics (Feldman and Friston, 2010; Friston, 2010; Friston et al., 2010). Hierarchical generative models of cognition, such as those described by Clark, presuppose the manipulation of representations and internal models of the world, in as much detail as is perceptually available. Perhaps surprisingly, Clark acknowledges the existence of a “virtual version of the sensory data” (p. 4), but with no reference to some of the historical debates that shaped cognitive science, related to the storage, manipulation, and retrieval of representations in a cognitive system (Shanahan, 1997), or accounting for the emergence of intentionality within such a system (Searle, 1980; Preston and Bishop, 2002). Instead of demonstrating how this Bayesian framework responds to these foundational questions, Clark describes the structure and the functional properties of an action-oriented, multi-level system that is meant to combine perception, learning, and experience (Niedenthal, 2007). As pointed out by Clark, extreme models within this framework reduce experience to a mere by-product of the relationship between neural anticipatory signal and motor commands. Rightfully, Clark is uncertain of the radical proposal that we might “do away with the need to appeal to goals and rewards” (p. 59), and attempts to reinstate some aspects of emotional experience in the form of a frame of reference against which is construed this action-oriented predictive framework. We submit that this argument falls short of momentum in accounting for a rich phenomenology. Emotional experience simply cannot be reduced to a frame of reference: embodiment and embeddedness are at the core of the organism’s identity in its lived world, and fundamental aspects of emotional experience (Niedenthal, 2007). These features of experience situate the organism in the environment, which perceives and interacts with its immediate surrounding. This situatedness can only arise from concentric cycles of operations that combine integrated levels of anticipation with enactive processes of interaction with the environment. Anticipation not only takes place in the brain, which is the preferred level of perspective in the models introduced by Clark, but also at the many levels contained within the body, and between the body and the environment (Kurthen, 2007). In such a framework, the brain is a necessary but not sufficient part of the enactive organism, and emotion takes a central place as the scaffolding to awareness, especially in action-oriented perspectives (Frijda et al., 1989). By enactive processes, we refer to the closed-loop operations that settle the organism in the environment, which are grounded in, and shaped by the interaction itself, and that lead to the organism acting in ways optimal for adaptation and survival, supporting perception, learning, and experience. Within this framework, anticipation evidently plays a critical role in mitigating the interface with the environment. This situation relies on both the existence of natural constraints in the environment, and on the organism’s sensitivity to these constraints, which can either be based on their explicit description (i.e., a model of the world), or on the implicit, natural relationships that exist within the unitary system composed of both the organism and the environment. In the former case, prediction of future events occurs via the explicit manipulation of the description of the world against some metric of time. In contrast to this “weak” form of anticipation, which requires an expensive degree of energy to be sustained over time, a systemic form of anticipation may arise from the natural dispositions of both components of the system: a so-called “strong” anticipation based on the delayed feedback between the physical elements of the system, the properties of their synchronization and the strength of the coupling between them (Stepp and Turvey, 2010). The example of the oil drop in the saline solution that successfully exits a maze by following an appropriate Ph gradient illustrates this latter situation (Lagzi et al., 2010; http://www.youtube.com/watch?v=RXgP8rq_wfA). Arguably the oil drop does not possess or manipulate a model of the world to achieve this feat, but the interaction over time of the Ph of the oil drop with that of the saline solution leads the oil drop to move in the appropriate direction. In this context, enactive processes contrast sharply with the models described by Clark in that they assume an implicit, cheaper (energy-wise), representation-lean reference to the future, as the natural, bidirectional relation between the organism and the environment unfolds over time. We thus submit that these models may be more immune to most informational bottlenecks evident in light of the requirements for surviving in the world, as exemplified by the historical debates we referred to earlier, and may therefore be more adequate than other models to account for cognition. At the level of the (embodied) brain, enactive processes may interface with the environment through several mechanisms including the synchronization of neural assemblies and large-scale integration of information (Engel et al., 2001; Varela et al., 2001). Much is needed to characterize these mechanisms. In conclusion, we would like to formulate a word of caution. It is a mistake to conclude that, because model x can account for data y (epistemic concerns), therefore it must provide an accurate description of the inner workings of the brain (ontological description). Unless the model describes all the complexities of the embodied brain, embedded in the environment, one is almost always going to be making a conflation mistake. An extreme example might be representing “emotion” using a real number and, because this model can account for some data, wrongfully conclude that there must be an equivalent to the real number in the brain. Arguably, Clark’s review of a wide-range of data to justify hierarchical generative models falls into this category.

Contemporary Sensorimotor Theory: A Brief Introduction

‘Sensorimotor Theory’ offers a new enactive approach to perception that emphasises the role of motor actions and their effect on sensory stimuli. The seminal publication that launched the field is the target paper co-authored by J. Kevin O’Regan and Alva Noe and published in Behavioral and Brain Sciences (BBS) for open peer commentary in 2001 [27].

Swarmic autopoiesis and computational creativity

ABSTRACT In this paper two swarm intelligence algorithms are used, the first leading the “attention” of the swarm and the latter responsible for the tracing mechanism. The attention mechanism is coordinated by agents of Stochastic Diffusion Search where they selectively attend to areas of a digital canvas (with line drawings) which contains (sharper) corners. Once the swarms attention is drawn to the line of interest with a sharp corner, the corresponding line segment is fed into the tracing algorithm, Dispersive Flies Optimisation which “consumes” the input in order to generate a “swarmic sketch” of the input line. The sketching process is the result of the “flies” leaving traces of their movements on the digital canvas which are then revisited repeatedly in an attempt to re-sketch the traces they left. This cyclic process is then introduced in the context of autopoiesis, where the philosophical aspects of the autopoietic artist are discussed. The autopoetic artist is described in two modalities: gluttonous and contented. In the Gluttonous Autopoietic Artist mode, by iteratively focussing on areas-of-rich-complexity, as the decoding process of the input sketch unfolds, it leads to a less complex structure which ultimately results in an empty canvas; therein reifying the artworks “death”. In the Contented Autopoietic Artist mode, by refocussing the autopoietic artists reflections on “meaning” onto different constitutive elements, and modifying her reconstitution, different behaviours of autopoietic creativity can be induced and therefore, the autopoietic processes become less likely to fade away and more open-ended in their creative endeavour.

Autopoiesis, creativity and dance

ABSTRACT For many years three key aspects of creative processes have been glossed over by theorists eager to avoid the mystery of consciousness and instead embrace an implicitly more formal, computational vision: autonomy, phenomenality and the temporally embedded and bounded nature of creative processes. In this paper we will discuss autopoiesis and creativity; an alternative metaphor which we suggest offers new insight into these long overlooked aspects of the creative processes in humans and the machine, and examine the metaphor in the context of dance choreography.

Rethinking Construction: On Luciano Floridi’s ‘Against Digital Ontology’

In the fourteenth chapter of The Philosophy of Information, Luciano Floridi puts forth a criticism of ‘digital ontology’ as a step toward the articulation of an ‘informational structural realism’. Based on the claims made in the chapter, the present paper seeks to evaluate the distinctly Kantian scope of the chapter from a rather unconventional viewpoint: while in sympathy with the author’s doubts ‘against’ digital philosophy, we follow a different route. We turn our attention to the concept of construction as used in the book with the hope of raising some additional questions that might contribute to a better understanding of what is at stake in Floridi’s experimental epistemological response to digital ontology.

Extending Naive Bayes Classifier with Hierarchy Feature Level Information for Record Linkage

Probabilistic record linkage has been well investigated in recent years. The Fellegi-Sunter probabilistic record linkage and its enhanced version are commonly used methods, which calculate match and non-match weights for each pair of corresponding fields of record-pairs. Bayesian network classifiers --- naive Bayes classifier and TAN have also been successfully used here. Very recently, an extended version of TAN called ETAN has been developed and proved superior in classification accuracy to conventional TAN. However, no previous work has applied ETAN in record linkage and investigated the benefits of using a naturally existing hierarchy feature level information. In this work, we extend the naive Bayes classifier with such information. Finally we apply all the methods to four datasets and estimate the