Tony C. Smith

Introducing Machine Learning Concepts with WEKA.

This chapter presents an introduction to data mining with machine learning. It gives an overview of various types of machine learning, along with some examples. It explains how to download, install, and run the WEKA data mining toolkit on a simple data set, then proceeds to explain how one might approach a bioinformatics problem. Finally, it includes a brief summary of machine learning algorithms for other types of data mining problems, and provides suggestions about where to find additional information.

Determination of ribonuclease sequence-specificity using Pentaprobes and mass spectrometry

The VapBC toxin-antitoxin (TA) family is the largest of nine identified TA families. The toxin, VapC, is a metal-dependent ribonuclease that is inhibited by its cognate antitoxin, VapB. Although the VapBCs are the largest TA family, little is known about their biological roles. Here we describe a new general method for the overexpression and purification of toxic VapC proteins and subsequent determination of their RNase sequence-specificity. Functional VapC was isolated by expression of the nontoxic VapBC complex, followed by removal of the labile antitoxin (VapB) using limited trypsin digestion. We have then developed a sensitive and robust method for determining VapC ribonuclease sequence-specificity. This technique employs the use of Pentaprobes as substrates for VapC. These are RNA sequences encoding every combination of five bases. We combine the RNase reaction with MALDI-TOF MS to detect and analyze the cleavage products and thus determine the RNA cut sites. Successful MALDI-TOF MS analysis of RNA fragments is acutely dependent on sample preparation methods. The sequence-specificity of four VapC proteins from two different organisms (VapC(PAE0151) and VapC(PAE2754) from Pyrobaculum aerophilum, and VapC(Rv0065) and VapC(Rv0617) from Mycobacterium tuberculosis) was successfully determined using the described strategy. This rapid and sensitive method can be applied to determine the sequence-specificity of VapC ribonucleases along with other RNA interferases (such as MazF) from a range of organisms.

Learning language using genetic algorithms

Strict pattern-based methods of grammar induction are often frustrated by the apparently inexhaustible variety of novel word combinations in large corpora. Statistical methods offer a possible solution by allowing frequent well-formed expressions to overwhelm the infrequent ungrammatical ones. They also have the desirable property of being able to construct robust grammars from positive instances alone. Unfortunately, the “zero-frequency” problem entails assigning a small probability to all possible word patterns, thus ungrammatical n-grams become as probable as unseen grammatical ones. Further, such grammars are unable to take advantage of inherent lexical properties that should allow infrequent words to inherit the syntactic properties of the class to which they belong.

Semantic and generative models for lossy text compression

The complementary paradigms of text compression and image compression suggest that there may be potential for applying methods developed for one domain to the other. In image coding, lossy techniques yield compression factors that are vastly superior to those of the best lossless schemes, and we show that this is also the case for text. This paper investigates the resulting tradeoff between subjective quality of the transmission and its compression factor. Two different methods are described, which can be combined into an extremely effective technique that provides far better compression than the present state of the art and yet preserves a reasonable degree of perceived match between the original and received text. The major challenge for lossy text compression is the quantitative evaluation of the quality of this match.

Adding some spice to CS1 curricula

CS1 curricula that focus primarily on teaching programming skills may fail to give students a sense of the breadth of topics to which a knowledge of computability might apply. We attempt to correct for this by adding a survey of computing concepts into our introductory course. Sophisticated topics usually reserved for senior level courses are presented in a popular science manner, and given equal time alongside the essential introductory programming material. By exposing students to a broad range of specific computational problems we endeavour to make their first course more interesting and enjoyable, and instil in them a sense of vision for areas they might specialise in as computing majors.

Investigating the usability of social networking sites for teenagers with autism

Teenagers with Autism Spectrum Disorder (ASD) tend to be socially isolated due to the interpersonal challenges of autism, yet they enjoy using computers. This study investigated the hypothesis that social networking sites provide the appropriate communication tools for teenagers with ASD. It concluded that although social networking sites remove extraneous stimuli which results in social anxiety among people with ASD, their design functionality does not provide the motivation required to initiate/conduct communication/social interaction among teenagers with ASD. Rather, people with ASD are motivated to communicate with others if the communication is part of an activity. Therefore, a multiplayer networking game has the potential of motivating teenagers with ASD to interact with others through entertainment.

Derivation of Classes from Use Cases Automatically Generated by a Three-Level Sentence Processing Algorithm

Identification of classes from a requirements specification is one of the most important and difficult tasks during analysis and design of new object- oriented systems. Recent developments in CASE tool research have begun incorporating quite sophisticated natural language processing techniques to help generate class diagrams; however, considerable problems remain in the form of overlooked and/or excess classes. The difficulty stems largely from inherent ambiguity in written language; and a general lack of conciseness and completeness in requirements specifications. We argue that a more effective approach is to obtain classes from use cases, rather than directly from the specification. Our justification is that use cases describe specific functionalities of the system, and therefore the candidate objects involved in effecting those functionalities. To that end, we propose an algorithm to i) extract use case sentences from requirements, ii) validate functional specifications in each sentence of the use cases so that classes can be identified, and iii) reuse previously validated, domain- dependent use cases to supply missing functional specifications that may contain participating classes. Thereafter, classes are automatically obtained from these generated use cases. Problems arising from ambiguity in the written requirements are mitigated by imposing a small number of restrictions on the language that can be used for writing requirements, and the sentence structure of use cases is formalized.

Semantic Role Labeling via Consensus in Pattern-Matching

A cathode for use in the electrolytic refining of copper which comprises a carrying bar and a flat starting plate, which is secured to the carrying bar and made of special steel and provided with electrically insulating edge strips at least at its vertical longitudinal edges, which are formed in their edge faces with a groove, which contains an edge strip, that is made of a natural or synthetic polymer and fits into said groove and protrudes from the edge throughout its length. To prevent a deposition of metal in the critical region, the insulating edge strip which has been fitted into the dovetail groove has been formed by a longitudinal folding of a polymer film and contains at its fold line a wire of special steel, and the side portions of said film are coextensive and lie one over the other and are liquid-tightly joined to each other. The invention relates also to a process of manufacturing the cathode.

Objective evaluation of inferred context-free grammars

An infinite number of context-free grammars may be inferred from a given training set. The defensibility of any single grammar hinges on the ability to compare that grammar against others in a meaningful way. In keeping with the minimum description length principle, smaller grammars are preferred over larger ones, but only insofar as the small grammar does not over-generalise the language being studied. Furthermore, measures of size must incorporate the grammars ability to cover sentences of the source language not included in the training set. This paper describes a method for evaluating the quality of context-free grammars according to (i) the complexity of each grammar and (ii) the amount of disambiguation information necessary for much grammar to reproduce the training set. The sum of the two evaluations is used as an objective measure of a grammars information content. Three grammars are used as examples of this process.<<ETX>>

Using the online cross-entropy method to learn relational policies for playing different games

By defining a video-game environment as a collection of objects, relations, actions and rewards, the relational reinforcement learning algorithm presented in this paper generates and optimises a set of concise, human-readable relational rules for achieving maximal reward. Rule learning is achieved using a combination of incremental specialisation of rules and a modified online cross-entropy method, which dynamically adjusts the rate of learning as the agent progresses. The algorithm is tested on the Ms. Pac-Man and Mario environments, with results indicating the agent learns an effective policy for acting within each environment.