Sin Wee Lee

Performance-guided neural network for rapidly self-organising active network management

We present a neural network for real-time learning and mapping of patterns using an external performance indicator. In a non-stationary environment where new patterns are introduced over time, the learning process utilises a novel snap-drift algorithm that performs fast, convergent, minimalist learning (snap) when the overall network performance is poor and slower, more cautious learning (drift) when the performance is good. Snap is based on a modified form of Adaptive Resonance Theory (CGIP 37(1987)54); and drift is based on Learning Vector Quantization (LVQ) (Proc. IJCNN 1(1990a)545). The two are combined within a semi-supervised learning system that shifts its learning style whenever it receives a significant change in performance feedback. The learning is capable of rapid re-learning and re-stabilisation, according to changes in external feedback or input patterns. We have incorporated this algorithm into the design of a modular neural network system, Performance-guided Adaptive Resonance Theory (P-ART) (Proc. IJCNN 2(2003)1412; Soft computing systems: Design, Management and application, IOS Press, Netherland, 2002; pp. 21-31). Simulation results show that the system discovers alternative solutions in response to significant changes in the input patterns and/or in the environment, which may require similar patterns to be treated differently over time. The simulations involve attempting to optimise the selection of network services in a non-stationary, real-time active computer network environment, in which the factors influencing the required selections are subject to change.

Phonetic feature discovery in speech using snap-drift learning

This paper presents a new application of the snap-drift algorithm [1]: feature discovery and clustering of speech waveforms from non-stammering and stammering speakers. The learning algorithm is an unsupervised version of snap-drift which employs the complementary concepts of fast, minimalist learning (snap) & slow drift (towards the input pattern) learning. The Snap-Drift Neural Network (SDNN) is toggled between snap and drift modes on successive epochs. The speech waveforms are drawn from a phonetically annotated corpus, which facilitates phonetic interpretation of the classes of patterns discovered by the SDNN.

Snap-drift neural network for selecting student feedback

This paper investigates the application of the snap-drift neural network (SDNN) to the provision of guided student learning in formative assessments. SDNN is able to adapt rapidly by performing a combination of fast, convergent, minimal intersection learning (snap) and Learning Vector Quantization (drift) to capture both precise sub-features in the data and more general holistic features. Snap and drift are combined within a modal learning system that toggles its learning style between the two modes. In this particular application the SDNN is trained with responses from past students to Multiple Choice Questions (MCQs). The neural network is able to categorise the learners responses as having a significant level of similarity with a subset of the students it has previously categorised. Each category is associated with feedback composed by the lecturer on the basis of the level of understanding and prevalent misconceptions of that category-group of students. The feedback addresses the level of knowledge of the individual and guides them towards a greater understanding of particular concepts. The trained snap-drift neural network is integrated into an on-line Multiple Choice Questions (MCQs) system. This approach has been implemented and trialled with two cohorts of students using data sets of student answers related to a topic from an Introduction to Computer System module. Results indicate that significant learning support is provided for the students.

Snap-drift learning for phrase recognition

This paper presents a new application of the snap-drift algorithm by S. W. Lee, et al. (2004): phrase recognition using a set of phrases from the Lancaster parsed corpus (LPC) by R. Garside, et al. (1987). The learning algorithm is the classifier version of snap-drift. In this version, along with the complementary concepts of fast minimalist learning (snap) and slow drift towards the input pattern, each node of the snap-drift neural network (SDNN) swaps between snap and drift modes when declining performance is indicated on that particular node. This method enables the SDNN to learn at node level, in the sense that each node has its learning mode toggled independently of the other nodes. Learning on each node is also reinforced by enabling learning with a probability that decreases with increasing performance. The simulations demonstrate that learning is stable, and the results have consistently shown similar classification performance and advantages in terms of speed in comparison with a multilayer perceptron (MLP) and back-propagation by J. Topper, et al. (2002), D. E. Rumelhart, et al. (1986) applied to the same problem.

Activities of daily life recognition using process representation modelling to support intention analysis

Purpose – This paper aims to focus on applying a range of traditional classification- and semantic reasoning-based techniques to recognise activities of daily life (ADLs). ADL recognition plays an important role in tracking functional decline among elderly people who suffer from Alzheimer’s disease. Accurate recognition enables smart environments to support and assist the elderly to lead an independent life for as long as possible. However, the ability to represent the complex structure of an ADL in a flexible manner remains a challenge. Design/methodology/approach – This paper presents an ADL recognition approach, which uses a hierarchical structure for the representation and modelling of the activities, its associated tasks and their relationships. This study describes an approach in constructing ADLs based on a task-specific and intention-oriented plan representation language called Asbru. The proposed method is particularly flexible and adaptable for caregivers to be able to model daily schedules for Alzheimer’s patients. Findings – A proof of concept prototype evaluation has been conducted for the validation of the proposed ADL recognition engine, which has comparable recognition results with existing ADL recognition approaches. Originality/value – The work presented in this paper is novel, as the developed ADL recognition approach takes into account all relationships and dependencies within the modelled ADLs. This is very useful when conducting activity recognition with very limited features.

Effective Diagnostic Feedback for Online Multiple-Choice Questions

When students attempt MCQs (Multiple-Choice Questions) they generate invaluable information which can form the basis for understanding their learning behaviours. In this research, the information is collected and automatically analysed to provide customized, diagnostic feedback to support students’ learning. This is achieved within a web-based system, incorporating the SDNN (Snap-drift neural network) based analysis of students’ responses to MCQs. This paper presents the results of a large trial of the method and the system which demonstrates the effectiveness of the feedback in guiding students towards a better understanding of particular concepts.