A novel approach to uncertain temporal inference is presented. Allen Interval Algebra is extended to fuzzy time-intervals via representing the latter as trapeziums with distinct beginning, middle and end. An uncertain version of the Interval Algebra composition table is developed via running a computer simulation in which a large number of fuzzy time-intervals are drawn from an assumed probability distribution.

A novel category of theories is proposed, providing a potential explanation for the representation of complex knowledge in the human (and, more generally, mammalian) brain. Firstly, a "glocal" representation for concepts is suggested, involving localized representations in a sparse network of "concept neurons" in the Medial Temporal Lobe, coupled with a complex dynamical attractor representation in other parts of cortex. Secondly, it is hypothesized that a combi-natory logic like representation is used to encode abstract relationships without explicit use of variable bindings, perhaps using systematic asynchronization among concept neurons to indicate an analogue of the combinatory-logic operation of function application. While unraveling the specifics of the brain's knowledge representation mechanisms will require data beyond what is currently available, the approach presented here provides a class of possibilities that is neurally plausible and bridges the gap between neurophysiological realities and mathematical and computer science concepts.

A practical roadmap to human-level artificial general intelligence is outlined, leading from the current situation in which complex AGI-oriented cognitive architectures remain partially implemented and inadequately tested, to a future in which AGI systems are deployed to carry out a variety of practical tasks currently only achievable via humans. The roadmap involves simultaneous development of proto-AGI applications in a number of different areas: videogame agent control, mobile robotics, genomics, natural language dialogue, theorem proving and program learning. The simultaneous pursuit of diverse application areas using a single AGI architecture is proposed as critical to maintaining generality of intelligence throughout the development process. While acknowledging the profound difficulty of time estimation for a project of this complexity, it is projected that the roadmap outlined could potentially be completed in a decade or less.

The Deep Spatio-Temporal Inference Network (DeSTIN) is a deep learning architecture which combines un-supervised learning and Bayesian inference. The original version of DeSTIN incorporates k-means clustering inside each processing node. Here we propose to replace k-means with a more sophisticated algorithm, online EM (Expectation Maximization), and show that this improves DeSTIN's performance on image classification and restoration tasks.

A high-level artificial general intelligence (AGI) architecture called goal-oriented learning meta-architecture (GOLEM) is presented, along with an informal but careful argument that GOLEM may be capable of preserving its initial goals while radically improving its general intelligence. As a meta-architecture, GOLEM can be wrapped around a variety of different base-level AGI systems, and also has a role for a powerful narrow-AI subcomponent as a probability estimator. The motivation underlying these ideas is the desire to create AGI systems fulfilling the multiple criteria of being: massively and self-improvingly intelligent, probably beneficial and almost surely not destructive.

The hypothesis is presented that "cognitive synergy" -- proactive and mutually-assistive feedback between different cognitive processes associated with different types of memory -- may serve as a foundation for advanced artificial general intelligence. A specific AI architecture founded on this idea, OpenCogPrime, is described, in the context of its application to control virtual agents and robots. The manifestations of cognitive synergy in OpenCogPrime's procedural and declarative learning algorithms are discussed in some detail.

Current work is described wherein simplified versions of the Novamente cognition engine (NCE) are being used to control virtual agents in virtual worlds such as game engines and second life. In this context, an IRC (imitation-reinforcement-correction) methodology is being used to teach the agents various behaviors, including simple tricks and communicative acts. Here we describe how this work may potentially be exploited and extended to yield a pathway toward giving the NCE robust, ultimately human level natural language conversation capability. The pathway starts via using the current system to instruct NCE-controlled agents in semiosis and gestural communication; and then continues via integration of a particular sort of hybrid rule-based/statistical NLP system (which is currently partially complete) into the NCE-based virtual agent system, in such a way as to allow experiential adaptation of the rules underlying the NLP system, in a manner that builds on the agentpsilas knowledge of semiosis and gesture.

A novel approach, model-based clustering, is described foridentifying complex interactions between genes or gene-categories based on static gene expression data. The approach deals with categorical data, which consists of a set of gene expressionprofiles belonging to one category, and a set belonging to anothercategory. An evolutionary algorithm (Meta-Optimizing Semantic Evolutionary Search, or MOSES) is used to learn an ensemble of classification models distinguishing the two categories, based on inputs that are features corresponding to gene expression values. Each feature is associated with a model-based vector, which encodes quantitative information regarding the utilization of the feature across the ensembles of models. Two different ways of constructing these vectors are explored. These model-based vectors are then clustered using a variant of hierarchical clustering called Omniclust. The result is a set of model-based clusters, in which features are gathered together if they are often considered together by classification models -- which may be because they're co-expressed, or may be for subtler reasons involving multi-gene interactions. The method is illustrated by applying it to two datasets regarding human gene expression, one drawn from brain cells and pertinent to the neurogenetics of aging, and the other drawn from blood cells and relating to differentiating between types of lymphoma. We find that, compared to traditional expression-based clustering, the new method often yields clusters that have higher mathematical quality (in the sense of homogeneity and separation) and also yield novel and meaningful insights into the underlying biological processes.

No abstract available.

A novel method, MUTIC (model utilization-based clustering), is described for identifying complex interactions between genes or gene-categories based on gene expression data. The method deals with binary categorical data, which consists of a set of gene expression profiles divided into two biologically meaningful categories. It does not require data from multiple time points. Gene expression profiles are represented by feature vectors whose component features are either gene expression values, or averaged expression values corresponding to gene ontology or protein information resource categories. A supervised learning algorithm (genetic programming) is used to learn an ensemble of classification models distinguishing the two categories based on the feature vectors corresponding to their members. Each feature is associated with a "model utilization vector," which has an entry for each high-quality classification model found, indicating whether or not the feature was used in that model. These utilization vectors are then clustered using a variant of hierarchical clustering called Omniclust. The result is a set of model-utilization-based clusters, in which features are gathered together if they are often considered together by classification models - which may be because they're co-expressed, or may be for subtler reasons involving multi-gene interactions. The MUTIC method is illustrated via applying it to a dataset regarding gene expression in human brains of various ages. Compared to traditional expression-based clustering, MUTIC yields clusters that have higher mathematical quality (in the sense of homogeneity and separation) and also yield novel insights into the underlying biological processes.

Support Vector Machines are used to combine the outputs of multiple entity extractors, thus creating a composite entity extraction system. The composite system has a significantly higher f-measure than any of the component systems. Compared to a standard voting technique for combining the results of multiple entity extractors, the SVM approach produces comparable precision and recall statistics but tends to utilize fewer of the component entity extractors, thus providing superior computational efficiency, which is critical in practical applications. In this paper, we present our experimental results of comparing a standard voting technique with SVM that each aggregate four entity extractors. We also describe our future plans of integrating agent-based technology into our experimental testbed where we examine the evolution of composite techniques as part of the analysis stream. Given that much of the improvement comes from tuning the algorithms to the data stream with a human-in-the-loop, we are considering the merits of employing cognitive agents that are strategically embedded in the workflow for processing data. As we tune the algorithms for better performance on the data streams, we envision agents learning the patterns of data streams and apply the appropriate tuning to ensure optimality.

It is proposed that the creation of artificial general intelligence (AGI) at the human level and ultimately beyond is a problem addressable via integrating computer science algorithms and data structures within a cognitive architecture oriented toward experiential learning. A general conceptual framework for AGI is presented, beginning with a philosophy of mind based on the concept of pattern, then moving to a general mathematical and conceptual framework for modeling intelligent systems, self-modifying evolving probabilistic hypergraphs (SMEPH), and finally to an overview of a specific design for AGI, the Novamente AI engine. The problem of teaching an AGI system is discussed, in the context of Novamente's embodiment in the AGI-SIM simulation world. An educational program based loosely on Piaget's developmental stages is outlined, followed by more detailed consideration of the learning by Novamente in AGI-SIM of the Piagetan infant-level capability of "object permanence".

A novel technique is described, wherein Support Vector Machines are used to perform relatively effective text categorization based on small numbers of positive examples (fewer than 10 in some cases). It is assumed that in addition to the positive examples a query describing the positive category is given (in the form of a set of key phrases or a sentence). The technique combines two innovations: a special way of altering the SVM score threshold based on looking at the distribution of scores across the training set; and, a method of feature selection that involves retaining only features that display semantic association to the content words in the query (according to a word-association database produced by statistical analysis of a parsed corpus). Examples are given on a number of test cases drawn from the Reuters and FBIS news archives.

We describe an extension of the Bayesian Optimization Algorithm (BOA), a probabilistic model building genetic algorithm, to the domain of program tree evolution. The new system, BOA programming (BOAP), improves significantly on previous probabilistic model building genetic programming (PMBGP) systems in terms of the articulacy and open-ended flexibility of the models learned, and hence control over the distribution of instances generated. Innovations include a novel tree representation and a generalized program evaluation scheme.