Nooraini Yusoff

Emotion classification using neural network

Neural networks have found profound success in the area of pattern recognition. By repeatedly showing a neural network inputs classified into groups, the network can be trained to discern the criteria used to classify, and it can do so in a generalized manner allowing successful classification of new inputs not used during training. With the explosion of research in emotion in recent year, the application of pattern recognition technology to emotion detection has become increasingly interesting. Since emotion has become an important interface for the communication between human and machine, it plays a basic role in rational decision-making, learning, perception, and various cognitive tasks. Humans emotion can be detected based on the physiological measurements, facial expression and vocal recognition. Since human shows the same facial muscles when expressing a particular emotion, therefore the emotion can be quantified. In this study, six primary emotions such as anger, disgust, fear, happiness, sadness and surprise were classified using Neural Network. Real dataset of facial expression images were captured and processed to prepare for Neural Network training and testing. The dataset was tested on Multilayer Layer Perceptron with Backpropagation learning algorithm and Regression analysis. The experimental results reveal that Neural Network has a misclassification rate of 2.5% while Regression analysis yields a misclassification rate of 33.33%.

Classifying breast cancer types based on fine needle aspiration biopsy data using random forest classifier

Breast cancer is a complex and heterogeneous disease due to its diverse morphological features, as well as different clinical outcome. As a result, breast cancer patients may response to different therapeutic options. Currently, difficulties in recognizing the breast cancer types lead to inefficient treatments. Generally, there are two types of breast cancer, known as malignant and benign. Therefore it is necessary to devise a clinically meaningful classification of the disease that can accurately classify breast cancer tissues into relevant classes. This study aims to classify breast cancer lesions which have been obtained from fine needle aspiration (FNA) procedure using random forest. Random forest is a classifier built based on the combination of decision trees and has been identified to perform well in comparison to other machine learning techniques. This method has been tested on approximately 700 data, which consists of 458 instances from benign cases and 241 instances belong to malignant cases. The performance of proposed method is measured based on sensitivity, specificity and accuracy. The experimental results show that, random forest achieved sensitivity of 75%, specificity of 70% and accuracy about 72%. Thus, it can be concluded that random forest can accurately classify breast cancer types given a small number of features and it works as a promising tool to differentiate malignant from benign tumor at early stage.

Learning anticipation through priming in spatio-temporal neural networks

In this paper, we propose a reward-based learning model inspired by the findings from a behavioural study and biologically realistic properties of spatio-temporal neural networks. The model simulates the cognitive priming effect in stimulus-stimulus-response association. Synaptic plasticity is dependent on a global reward signal that enhances the synaptic changes derived from spike-timing dependent plasticity (STDP) process. We show that by priming a network with a cue stimulus can facilitate the response to a later stimulus. The network can be trained to associate a stimulus pair (with an inter-stimulus interval) to a response, as well as to recognise the temporal sequence of the stimulus presentation.

Supervised associative learning in spiking neural network

In this paper, we propose a simple supervised associative learning approach for spiking neural networks. In an excitatory-inhibitory network paradigm with Izhikevich spiking neurons, synaptic plasticity is implemented on excitatory to excitatory synapses dependent on both spike emission rates and spike timings. As results of learning, the network is able to associate not just familiar stimuli but also novel stimuli observed through synchronised activity within the same subpopulation and between two associated subpopulations.

Pair-Associate learning with modulated spike-time dependent plasticity

We propose an associative learning model using reward modulated spike-time dependent plasticity in reinforcement learning paradigm. The task of learning is to associate a stimulus pair, known as the predictor−choice pair, to a target response. In our model, a generic architecture of neural network has been used, with minimal assumption about the network dynamics. We demonstrate that stimulus-stimulus-response association can be implemented in a stochastic way within a noisy setting. The network has rich dynamics resulting from its recurrent connectivity and background activity. The algorithm can learn temporal sequence detection and solve temporal XOR problem.

Modelling the Stroop Effect: Dynamics in Inhibition of Automatic Stimuli Processing

In this study, we simulate the dynamics of suppressing an automatic stimulus processing which interferes with a different non-automatic target task. The dynamics can be observed in terms of interference and facilitation effects that influence target response processing time. For this purpose, we use Hopfield neural network with varying attention modulation in a colour-word Stroop stimuli processing paradigm. With the biologically realistic features of the network, our model is able to model the Stroop effect in comparison to the human performance.

Filter-Based Gene Selection Method for Tissues Classification on Large Scale Gene Expression Data

DNA microarray technology is a current innovative tool that has offers a new perspective to look sight into cellular systems and measure a large scale of gene expressions at once. Regardless the novel invention of DNA microarray, most of its results relies on the computational intelligence power, which is used to interpret the large number of data. At present, interpreting large scale of gene expression data remain a thought-provoking issue due to their innate nature of “high dimensional low sample size”. Microarray data mainly involved thousands of genes, n in a very small size sample, p. In addition, this data are often overwhelmed, over fitting and confused by the complexity of data analysis. Due to the nature of this microarray data, it is also common that a large number of genes may not be informative for classification purposes. For such a reason, many studies have used feature selection methods to select significant genes that present the maximum discriminative power between cancerous and normal tissues. In this study, we aim to investigate and compare the effectiveness of these four popular filter gene selection methods namely Signal-to-Noise ratio (SNR), Fisher Criterion (FC), Information Gain (IG) and t-Test in selecting informative genes that can distinguish cancer and normal tissues. Two common classifiers, Support Vector Machine (SVM) and Decision Tree (C4.5) are used to train the selected genes. These gene selection methods are tested on three large scales of gene expression datasets, namely breast cancer dataset, colon dataset, and lung dataset. This study has discovered that IG and SNR are more suitable to be used with SVM while IG fit for C4.5. In a colon dataset, SVM has achieved a specificity of 86% with SNR while and 80% for IG. In contract, C4.5 has obtained a specificity of 78% for IG on the identical dataset. These results indicate that SVM performed slightly better with IG pre-processed data compare to C4.5 on the same dataset.

Spatio-temporal event association using reward-modulated spike-time-dependent plasticity

Abstract For goal-directed learning in spiking neural networks, target spike templates are usually required. Optimal performance is achieved by minimising the error between the desired and output spike timings. However, in some dynamic environments, a set of learning targets with precise encoding is not always available. For this study, we associate a pair of spatio-temporal events with a target response using a reinforcement learning approach. The learning is implemented in a recurrent spiking neural network using reward-modulated spike-time-dependent plasticity. The learning protocol is simple and inspired by a behavioural experiment from a neuropsychology study. For a goal-directed application, learning does not require a target spike template. In this study, convergence is measured by synchronicity of activities in associated neuronal groups. As a result of learning, a network is able to associate a pair of events with a temporal delay in a dynamic setting. The results demonstrate that the algorithm can also learn temporal sequence detection. Learning has also been tested in face-voice association using real biometric data. The loose dependency between the models anatomical properties and functionalities could offer a wide range of applications, especially in complex learning environments.

Copy-move forgery detection in digital image

Copy-move is considered as one of the most popular kind of digital image tempering, in which one or more parts of a digital image are copied and pasted into different locations. Geometric transformation is among the major challenges in detecting copy-move forgery of a digital image. In such forgery, the copied and moved parts of a forged image are either rotated or/and re-scaled. Hence, in this study we propose a combination of Discrete Wavelet Transform (DWT) and Speeded Up Robust Features (SURF) to detect a copy-move activity. The experiments results prove that the proposed method is superior with overall accuracy 95%. The copy-move attacks in digital image has been successfully detected and the method is also can detect the fraud parts exposed to rotation and scaling issue.

Spiking Neural Network with Lateral Inhibition for Reward-Based Associative Learning

In this paper we propose a lateral inhibitory spiking neural network for reward-based associative learning with correlation in spike patterns for conflicting responses. The network has random and sparse connectivity, and we introduce a lateral inhibition via an anatomical constraint and synapse reinforcement. The spiking dynamic follows the properties of Izhikevich spiking model. The learning involves association of a delayed stimulus pair to a response using reward modulated spike-time dependent plasticity (STDP). The proposed learning scheme has improved our initial work by allowing learning in a more dynamic and competitive environment.