scholarly journals A Simple Convolutional Neural Network with Rule Extraction

2019 ◽  
Vol 9 (12) ◽  
pp. 2411 ◽  
Author(s):  
Guido Bologna
Keyword(s):  
Extraction Method ◽  
Cross Validation ◽  
Rule Extraction ◽  
High Fidelity ◽  
Multi Layer Perceptron ◽  
Extraction Techniques ◽  
Textual Data ◽  
Input Layer ◽  
Axis Parallel ◽  
Fully Connected

Classification responses provided by Multi Layer Perceptrons (MLPs) can be explained by means of propositional rules. So far, many rule extraction techniques have been proposed for shallow MLPs, but not for Convolutional Neural Networks (CNNs). To fill this gap, this work presents a new rule extraction method applied to a typical CNN architecture used in Sentiment Analysis (SA). We focus on the textual data on which the CNN is trained with “tweets” of movie reviews. Its architecture includes an input layer representing words by “word embeddings”, a convolutional layer, a max-pooling layer, followed by a fully connected layer. Rule extraction is performed on the fully connected layer, with the help of the Discretized Interpretable Multi Layer Perceptron (DIMLP). This transparent MLP architecture allows us to generate symbolic rules, by precisely locating axis-parallel hyperplanes. Experiments based on cross-validation emphasize that our approach is more accurate than that based on SVMs and decision trees that substitute DIMLPs. Overall, rules reach high fidelity and the discriminative n-grams represented in the antecedents explain the classifications adequately. With several test examples we illustrate the n-grams represented in the activated rules. They present the particularity to contribute to the final classification with a certain intensity.

scholarly journals A Two-Step Rule-Extraction Technique for a CNN

Electronics ◽  
2020 ◽  
Vol 9 (6) ◽  
pp. 990
Author(s):  
Guido Bologna ◽  
Silvio Fossati
Keyword(s):  
Neural Networks ◽  
Deep Learning ◽  
Skin Cancer ◽  
Convolutional Neural Networks ◽  
Cancer Diagnosis ◽  
Medical Diagnosis ◽  
Rule Extraction ◽  
High Fidelity ◽  
Max Pooling ◽  
Fully Connected

The explanation of the decisions provided by a model are crucial in a domain such as medical diagnosis. With the advent of deep learning, it is very important to explain why a classification is reached by a model. This work tackles the transparency problem of convolutional neural networks(CNNs). We propose to generate propositional rules from CNNs, because they are intuitive to the way humans reason. Our method considers that a CNN is the union of two subnetworks: a multi-layer erceptron (MLP) in the fully connected layers; and a subnetwork including several 2D convolutional layers and max-pooling layers. Rule extraction exhibits two main steps, with each step generating rules from each subnetwork of the CNN. In practice, we approximate the two subnetworks by two particular MLP models that makes it possible to generate propositional rules. We performed the experiments with two datasets involving images: MNISTdigit recognition; and skin-cancer diagnosis. With high fidelity, the extracted rules designated the location of discriminant pixels, as well as the conditions that had to be met to achieve the classification. We illustrated several examples of rules by their centroids and their discriminant pixels.

KLASIFIKASI CITRA GENUS PANTHERA MENGGUNAKAN METODE CONVOLUTIONAL NEURAL NETWORK (CNN)

2019 ◽  
Vol 24 (3) ◽  
pp. 220-228
Author(s):  
Gusti Alfahmi Anwar ◽  
Desti Riminarsih
Keyword(s):  
Neural Network ◽  
Convolutional Neural Network ◽  
Network Model ◽  
Neural Network Model ◽  
Input Layer ◽  
Fully Connected

Panthera merupakan genus dari keluarga kucing yang memiliki empat spesies popular yaitu, harimau, jaguar, macan tutul, singa. Singa memiliki warna keemasan dan tidak memilki motif, harimau memiliki motif loreng dengan garis-garis panjang, jaguar memiliki tubuh yang lebih besar dari pada macan tutul serta memiliki motif tutul yang lebih lebar, sedangkan macan tutul memiliki tubuh yang sedikit lebih ramping dari pada jaguar dan memiliki tutul yang tidak terlalu lebar. Pada penelitian ini dilakukan klasifikasi genus panther yaitu harimau, jaguar, macan tutul, dan singa menggunakan metode Convolutional Neural Network. Model Convolutional Neural Network yang digunakan memiliki 1 input layer, 5 convolution layer, dan 2 fully connected layer. Dataset yang digunakan berupa citra harimau, jaguar, macan tutul, dan singa. Data training terdiri dari 3840 citra, data validasi sebanyak 960 citra, dan data testing sebanyak 800 citra. Hasil akurasi dari pelatihan model untuk training yaitu 92,31% dan validasi yaitu 81,88%, pengujian model menggunakan dataset testing mendapatan hasil 68%. Hasil akurasi prediksi didapatkan dari nilai F1-Score pada pengujian didapatkan sebesar 78% untuk harimau, 70% untuk jaguar, 37% untuk macan tutul, 74% untuk singa. Macan tutul mendapatkan akurasi terendah dibandingkan 3 hewan lainnya tetapi lebih baik dibandingkan hasil penelitian sebelumnya.

scholarly journals Integrated Metabolomics and Transcriptomics Using an Optimised Dual Extraction Process to Study Human Brain Cancer Cells and Tissues

Metabolites ◽  
2021 ◽  
Vol 11 (4) ◽  
pp. 240
Author(s):  
Alison Woodward ◽  
Alina Pandele ◽  
Salah Abdelrazig ◽  
Catherine A. Ortori ◽  
Iqbal Khan ◽  
...  
Keyword(s):  
Metabolic Pathways ◽  
Extraction Method ◽  
Limit Of Detection ◽  
Rna Extraction ◽  
Extraction Process ◽  
Extraction Methods ◽  
Serum Starvation ◽  
Extraction Techniques ◽  
Metabolic Genes ◽  
Dual Extraction

The integration of untargeted metabolomics and transcriptomics from the same population of cells or tissue enhances the confidence in the identified metabolic pathways and understanding of the enzyme–metabolite relationship. Here, we optimised a simultaneous extraction method of metabolites/lipids and RNA from ependymoma cells (BXD-1425). Relative to established RNA (mirVana kit) or metabolite (sequential solvent addition and shaking) single extraction methods, four dual-extraction techniques were evaluated and compared (methanol:water:chloroform ratios): cryomill/mirVana (1:1:2); cryomill-wash/Econospin (5:1:2); rotation/phenol-chloroform (9:10:1); Sequential/mirVana (1:1:3). All methods extracted the same metabolites, yet rotation/phenol-chloroform did not extract lipids. Cryomill/mirVana and sequential/mirVana recovered the highest amounts of RNA, at 70 and 68% of that recovered with mirVana kit alone. sequential/mirVana, involving RNA extraction from the interphase of our established sequential solvent addition and shaking metabolomics-lipidomics extraction method, was the most efficient approach overall. Sequential/mirVana was applied to study a) the biological effect caused by acute serum starvation in BXD-1425 cells and b) primary ependymoma tumour tissue. We found (a) 64 differentially abundant metabolites and 28 differentially expressed metabolic genes, discovering four gene-metabolite interactions, and (b) all metabolites and 62% lipids were above the limit of detection, and RNA yield was sufficient for transcriptomics, in just 10 mg of tissue.

scholarly journals Deep Convolution Neural Networks Learned Image Classification for Early Cancer Detection Using Lightweight

2021 ◽  
Author(s):  
Seshadri Ramana K ◽  
Bala Chowdappa K ◽  
Obulesu ooruchintala ◽  
Deena Babu Mandru ◽  
kallam suresh
Keyword(s):  
Cancer Detection ◽  
Early Cancer ◽  
The Body ◽  
Early Cancer Detection ◽  
Detection Accuracy ◽  
Convolutional Network ◽  
Max Pooling ◽  
Input Layer ◽  
Deep Convolutional Network ◽  
Fully Connected

Abstract Cancer is uncontrolled cell growth in any part of the body. Early cancer detection aims to identify patients who exhibit symptoms early on in order to maximise their chances of a successful treatment. Cancer disease mortality is decreased through early detection and treatment. Numerous researchers proposed a variety of image processing and machine learning approaches for cancer detection. However, existing systems did not improve detection accuracy or efficiency. A Deep Convolutional Neural Learning Classifier Model based on the Least Mean Square Filterative Ricker Wavelet Transform (L-DCNLC) is proposed to address the aforementioned issues. The L-DCNLC Model's primary objective is to detect cancer earlier by utilising a fully connected max pooling deep convolutional network with increased accuracy and reduced time consumption. The fully connected max pooling deep convolutional network is composed of one input layer, three hidden layers, and one output layer. Initially, the input layer of the L-DCNLC Model considers the number of patient images in the database as input.

Sign in / Sign up

Export Citation Format

Share Document