A computational knowledge representation model for cognitive computers

The accumulating data are easy to store but the ability of understanding and using it does not keep track with its growth. So researches focus on the nature of knowledge processing in the mind. This paper proposes a semantic model (CKRMCC) based on cognitive aspects that enables cognitive computer to process the knowledge as the human mind and find a suitable representation of that knowledge. In cognitive computer, knowledge processing passes through three major stages: knowledge acquisition and encoding, knowledge representation, and knowledge inference and validation. The core of CKRMCC is knowledge representation, which in turn proceeds through four phases: prototype formation phase, discrimination phase, generalization phase, and algorithm development phase. Each of those phases is mathematically formulated using the notions of real-time process algebra. The performance efficiency of CKRMCC is evaluated using some datasets from the well-known UCI repository of machine learning datasets. The acquired datasets are divided into training and testing data that are encoded using concept matrix. Consequently, in the knowledge representation stage, a set of symbolic rule is derived to establish a suitable representation for the training datasets. This representation will be available in a usable form when it is needed in the future. The inference stage uses the rule set to obtain the classes of the encoded testing datasets. Finally, knowledge validation phase is validating and verifying the results of applying the rule set on testing datasets. The performances are compared with classification and regression tree and support vector machine and prove that CKRMCC has an efficient performance in representing the knowledge using symbolic rules.     

Incremental learning of context free grammars based on bottom-up parsing and search

This paper describes approaches for machine learning of context free grammars (CFGs) from positive and negative sample strings, which are implemented in Synapse system. The grammatical inference consists of a rule generation by “inductive CYK algorithm,” mechanisms for incremental learning, and search. Inductive CYK algorithm generates minimum production rules required for parsing positive samples, when the bottom-up parsing by CYK algorithm does not succeed. The incremental learning is used not only for synthesizing grammars by giving the system positive strings in the order of their length but also for learning grammars from other similar grammars. Synapse can synthesize fundamental ambiguous and unambiguous CFGs including nontrivial grammars such as the set of strings not of the form ww with w∈{a,b}⁺.     

A meta-cognitive sequential learning algorithm for neuro-fuzzy inference system

In this paper, we present a meta-cognitive sequential learning algorithm for a neuro-fuzzy inference system, referred to as, ‘Meta-Cognitive Neuro-Fuzzy Inference System’ (McFIS). McFIS has two components, viz., a cognitive component and a meta-cognitive component. The cognitive component employed is a Takagi–Sugeno–Kang type-0 neuro-fuzzy inference system. A self-regulatory learning mechanism that controls the learning process of the cognitive component, by deciding what-to-learn, when-to-learn and how-to-learn from sequential training data, forms the meta-cognitive component. McFIS realizes the above decision by employing sample deletion, sample reserve and sample learning strategy, respectively. The meta-cognitive component use the instantaneous error of the sample and spherical potential of the rule antecedents to select the best training strategy for the current sample. Also, in sample learning strategy, when a new rule is added the rule consequent is assigned such that the localization property of Gaussian rule is fully exploited. The performance of McFIS is evaluated on four regression and eight classification problems. The performance comparison shows the superior generalization performance of McFIS compared to existing algorithms.     

A unified perspective and new results on RHT computing, mixture based learning, and multi-learner based problem solving

On one hand, multiple object detection approaches of Hough transform (HT) type and randomized HT type have been extended into an evidence accumulation featured general framework for problem solving, with five key mechanisms elaborated and several extensions of HT and RHT presented. On the other hand, another framework is proposed to integrate typical multi-learner based approaches for problem solving, particularly on Gaussian mixture based data clustering and local subspace learning, multi-sets mixture based object detection and motion estimation, and multi-agent coordinated problem solving. Typical learning algorithms, especially those that base on rival penalized competitive learning (RPCL) and Bayesian Ying-Yang (BYY) learning, are summarized from a unified perspective with new extensions. Furthermore, the two different frameworks are not only examined with one viewed crossly from a perspective of the other, with new insights and extensions, but also further unified into a general problem solving paradigm that consists of five basic mechanisms in terms of acquisition, allocation, amalgamation, admission, and affirmation, or shortly A5 paradigm.     

Towards an ensemble based system for predicting the number of software faults

Software fault prediction using different techniques has been done by various researchers previously. It is observed that the performance of these techniques varied from dataset to dataset, which make them inconsistent for fault prediction in the unknown software project. On the other hand, use of ensemble method for software fault prediction can be very effective, as it takes the advantage of different techniques for the given dataset to come up with better prediction results compared to individual technique. Many works are available on binary class software fault prediction (faulty or non-faulty prediction) using ensemble methods, but the use of ensemble methods for the prediction of number of faults has not been explored so far. The objective of this work is to present a system using the ensemble of various learning techniques for predicting the number of faults in given software modules. We present a heterogeneous ensemble method for the prediction of number of faults and use a linear combination rule and a non-linear combination rule based approaches for the ensemble. The study is designed and conducted for different software fault datasets accumulated from the publicly available data repositories. The results indicate that the presented system predicted number of faults with higher accuracy. The results are consistent across all the datasets. We also use prediction at level l (Pred(l)), and measure of completeness to evaluate the results. Pred(l) shows the number of modules in a dataset for which average relative error value is less than or equal to a threshold value l. The results of prediction at level l analysis and measure of completeness analysis have also confirmed the effectiveness of the presented system for the prediction of number of faults. Compared to the single fault prediction technique, ensemble methods produced improved performance for the prediction of number of software faults. Main impact of this work is to allow better utilization of testing resources helping in early and quick identification of most of the faults in the software system.     

The identification,implementation, and evaluation of critical user interface design features of computer-assisted instruction programs in mathematics for students with learning disabilities

Critical user interface design features of computer-assisted instruction programs in mathematics for students with learning disabilities and corresponding implementation guidelines were identified in this study. Based on the identified features and guidelines, a multimedia computer-assisted instruction program, ‘Math Explorer’, which delivers addition and subtraction word problem-solving instruction for students with learning disabilities at the early elementary level, was designed and developed. Lastly, usability testing was conducted to assess whether Math Explorer was well-designed in terms of the interface for students with learning disabilities. Given the results of the usability testing, this study corroborated the fact that the critical user interface design features and guidelines in mathematics computer-assisted instruction programs would be essential for facilitating the mathematical learning of students with learning disabilities. Implications for practice and future research were discussed.     

Improving active Mealy machine learning for protocol conformance testing

Using a well-known industrial case study from the verification literature, the bounded retransmission protocol, we show how active learning can be used to establish the correctness of protocol implementation I relative to a given reference implementation R. Using active learning, we learn a model M _R of reference implementation R, which serves as input for a model-based testing tool that checks conformance of implementation I to M _R . In addition, we also explore an alternative approach in which we learn a model M _I of implementation I, which is compared to model M _R using an equivalence checker. Our work uses a unique combination of software tools for model construction (Uppaal), active learning (LearnLib, Tomte), model-based testing (JTorX, TorXakis) and verification (CADP, MRMC). We show how these tools can be used for learning models of and revealing errors in implementations, present the new notion of a conformance oracle, and demonstrate how conformance oracles can be used to speed up conformance checking.     

Comparison of methods for logic-query implementation

A logic query Q is a triple < G, LP, D, where G is the query goal, LP is a logic program without function symbols, and D is a set of facts, possibly stored as tuples of a relational database. The answers of Q are all facts that can be inferred from LP ∪ D and unify with G. A logic query is bound if some argument of the query goal is a constant; it is canonical strongly linear (a CSL query) if LP contains exactly one recursive rule and this rule is linear, i.e., only one recursive predicate occurs in its body. In this paper, the problem of finding the answers of a bound CSL query is studied with the aim of comparing for efficiency some well-known methods for implementing logic queries: the eager method, the counting method, and the magic-set method. It is shown that the above methods can be expressed as algorithms for finding particular paths in a directed graph associated to the query. Within this graphical formalism, a worst-case complexity analysis of the three methods is performed. It turns out that the counting method has the best upper bound for noncyclic queries. On the other hand, since the counting method is not safe if queries are cyclic, the method is extended to safely implement this kind of queries as well.     

Fuzzy conditional inference under max-⊙ composition

This paper shows that the majority of fuzzy inference methods for a fuzzy conditional proposition “If x is A then y is B,” with A and B fuzzy concepts, can infer very reasonable consequences which fit our intuition with respect to several criteria such as modus ponens and modus tollens, if a new composition called “max-⊙ composition” is used in the compositional rule of inference, though reasonable consequences cannot always be obtained when using the max-min composition, which is used usually in the compositional rule of inference. Furthermore, it is shown that a syllogism holds for the majority of the methods under the max-⊙ composition, though they do not always satisfy the syllogism under the max-min composition.     

U-shaped, iterative, and iterative-with-counter learning

This paper solves an important problem left open in the literature by showing that U-shapes are unnecessary in iterative learning from positive data. A U-shape occurs when a learner first learns, then unlearns, and, finally, relearns, some target concept. Iterative learning is a Gold-style learning model in which each of a learner’s output conjectures depends only upon the learner’s most recent conjecture and input element. Previous results had shown, for example, that U-shapes are unnecessary for explanatory learning, but are necessary for behaviorally correct learning. Work on the aforementioned problem led to the consideration of an iterative-like learning model, in which each of a learner’s conjectures may, in addition, depend upon the number of elements so far presented to the learner. Learners in this new model are strictly more powerful than traditional iterative learners, yet not as powerful as full explanatory learners. Can any class of languages learnable in this new model be learned without U-shapes? For now, this problem is left open.     

Heuristic and probabilistic methods for taking efficient readings in diagnostic systems

This paper is devoted to the application of the following three approaches in diagnostic problems: heuristic, based on justifications of predictions of the results of operation of device components; probabilistic, using the structural dependence between components; and combined, with application of the probabilistic method and a heuristics for testing the workability of components. The capabilities of assumption-based truth maintenance systems (the so-called ATMS), which allow one to maintain consistency in DBs, are employed. The methods for choosing the point of the recurrent measurement were tested for a nine-bit device for parity checking. The results of testing have confirmed that the methods using the probabilistic and combined approaches provide the maximum efficiency of the choice of places for taking readings.     

Good edit similarity learning by loss minimization

Similarity functions are a fundamental component of many learning algorithms. When dealing with string or tree-structured data, measures based on the edit distance are widely used, and there exist a few methods for learning them from data. However, these methods offer no theoretical guarantee as to the generalization ability and discriminative power of the learned similarities. In this paper, we propose an approach to edit similarity learning based on loss minimization, called GESL. It is driven by the notion of (?,??,??)-goodness, a theory that bridges the gap between the properties of a similarity function and its performance in classification. Using the notion of uniform stability, we derive generalization guarantees that hold for a large class of loss functions. We also provide experimental results on two real-world datasets which show that edit similarities learned with GESL induce more accurate and sparser classifiers than other (standard or learned) edit similarities.     

Characterizing reinforcement learning methods through parameterized learning problems

The field of reinforcement learning (RL) has been energized in the past few decades by elegant theoretical results indicating under what conditions, and how quickly, certain algorithms are guaranteed to converge to optimal policies. However, in practical problems, these conditions are seldom met. When we cannot achieve optimality, the performance of RL algorithms must be measured empirically. Consequently, in order to meaningfully differentiate learning methods, it becomes necessary to characterize their performance on different problems, taking into account factors such as state estimation, exploration, function approximation, and constraints on computation and memory. To this end, we propose parameterized learning problems, in which such factors can be controlled systematically and their effects on learning methods characterized through targeted studies. Apart from providing very precise control of the parameters that affect learning, our parameterized learning problems enable benchmarking against optimal behavior; their relatively small sizes facilitate extensive experimentation. Based on a survey of existing RL applications, in this article, we focus our attention on two predominant, ??first order?? factors: partial observability and function approximation. We design an appropriate parameterized learning problem, through which we compare two qualitatively distinct classes of algorithms: on-line value function-based methods and policy search methods. Empirical comparisons among various methods within each of these classes project Sarsa(??) and Q-learning(??) as winners among the former, and CMA-ES as the winner in the latter. Comparing Sarsa(??) and CMA-ES further on relevant problem instances, our study highlights regions of the problem space favoring their contrasting approaches. Short run-times for our experiments allow for an extensive search procedure that provides additional insights on relationships between method-specific parameters??such as eligibility traces, initial weights, and population sizes??and problem instances.     

A novel semi-supervised learning framework with simultaneous text representing

Text representation has received extensive attention in text mining tasks. There are various text representation models. Among them, vector space model is the most commonly used one. For vector space model, the core technique is term weighting. To date, a great deal of different term-weighting methods have been proposed, which can be divided into supervised group and unsupervised group. However, it is not advisable to use these two groups of methods directly in semi-supervised applications. In semi-supervised applications, the majority of the supervised term-weighting methods are not applicable as the label information is insufficient; meanwhile, the unsupervised term-weighting methods cannot make use of the provided category labels. Thus, a semi-supervised learning framework for iteratively revising the text representation by an EM-like strategy is proposed in this paper. Furthermore, a new supervised term-weighting method t f.sd f is proposed. T f.sd f has the ability to emphasize the importance of terms that are unevenly distributed among all the classes and weaken the importance of terms that are uniformly distributed. Experimental results on real text data show that the proposed semi-supervised learning framework with the aid of t f.sd f performs well. Also, t f.sd f is shown to be efficient for supervised learning.     

Information theory in property testing and monotonicity testing in higher dimension

In property testing, we are given oracle access to a function f, and we wish to test if the function satisfies a given property P, or it is ϵ-far from having that property. In a more general setting, the domain on which the function is defined is equipped with a probability distribution, which assigns different weight to different elements in the domain. This paper relates the complexity of testing the monotonicity of a function over the d-dimensional cube to the Shannon entropy of the underlying distribution. We provide an improved upper bound on the query complexity of the property tester.     

Two New Perspectives on Multi-Stage Group Testing

The group testing problem asks to find d?n defective elements out of n elements, by testing subsets (pools) for the presence of defectives. In the strict model of group testing, the goal is to identify all defectives if at most d defectives exist, and otherwise to report that more than d defectives are present. If tests are time-consuming, they should be performed in a small constant number s of stages of parallel tests. It is known that a test number O(dlogn), which is optimal up to a constant factor, can be achieved already in s=2 stages. Here we study two aspects of group testing that have not found major attention before. (1) Asymptotic bounds on the test number do not yet lead to optimal strategies for specific n,d,s. Especially for small n we show that randomized strategies significantly save tests on average, compared to worst-case deterministic results. Moreover, the only type of randomness needed is a random permutation of the elements. We solve the problem of constructing optimal randomized strategies for strict group testing completely for the case when d=1 and s≤2. A byproduct of our analysis is that optimal deterministic strategies for strict group testing for d=1 need at most 2 stages. (2) Usually, an element may participate in several pools within a stage. However, when the elements are indivisible objects, every element can belong to at most one pool at the same time. For group testing with disjoint simultaneous pools we show that Θ(sd(n/d)^1/s ) tests are sufficient and necessary. While the strategy is simple, the challenge is to derive tight lower bounds for different s and different ranges of d versus n.     

Some variants of adaptive sampling procedures and their applications

Sequential analysis as a sampling technique facilitates efficient statistical inference by considering less number of observations in comparison to the fixed sampling method. The optimal stopping rule dictates the sample size and also the statistical inference deduced thereafter. In this research we propose three variants of the already existing multistage sampling procedures and name them as (i) Jump and Crawl (JC), (ii) Batch Crawl and Jump (BCJ) and (iii) Batch Jump and Crawl (BJC) sequential sampling methods. We use the (i) normal, (ii) exponential, (iii) gamma and (iv) extreme value distributions for the point estimation problems under bounded risk conditions. We highlight the efficacy of using the right adaptive sampling plan for the bounded risk problems for these four distributions, considering two different loss functions, namely (i) squared error loss (SEL) and (ii) linear exponential (LINEX) loss functions. Comparison and analysis of our proposed methods with existing sequential sampling techniques is undertaken and the importance of this study is highlighted using extensive theoretical simulation runs.     

Using schedulers to test probabilistic distributed systems

Formal methods are one of the most important approaches to increasing the confidence in the correctness of software systems. A formal specification can be used as an oracle in testing since one can determine whether an observed behaviour is allowed by the specification. This is an important feature of formal testing: behaviours of the system observed in testing are compared with the specification and ideally this comparison is automated. In this paper we study a formal testing framework to deal with systems that interact with their environment at physically distributed interfaces, called ports, and where choices between different possibilities are probabilistically quantified. Building on previous work, we introduce two families of schedulers to resolve nondeterministic choices among different actions of the system. The first type of schedulers, which we call global schedulers, resolves nondeterministic choices by representing the environment as a single global scheduler. The second type, which we call localised schedulers, models the environment as a set of schedulers with there being one scheduler for each port. We formally define the application of schedulers to systems and provide and study different implementation relations in this setting.