Transformation Across Deep Learning Frameworks

Since the presented approach uses MatConvNet of Matlab as a preliminary training platform, the pre-trained CNN model of MatConvNet cannot be directly integrated into the Xcode platform currently. Therefore, developers need a third-party platform as a bridge, so that developers can transfer the model of Matlab to the Xcode environment and finally mount the model to an app for executing and testing on the iOS device. Apple provides developers with Core ML Tools to support the Caffe framework. Therefore, developers can convert the Caffe model into the ML model through Core ML Tools. Moreover, the Caffe provides MatCaffe for connecting Matlab and Caffe. It is apparent that developers can achieve the goal through these two bridges.

Handwriting 99 Multiplication on App Store

The Modified NIST (MNIST) database, consisting of 70,000 handwritten digit images, in partition to 60,000 training patterns and 10,000 testing patterns, serves as a typical benchmark of evaluating performance of handwritten digit classification. After the LeNet CNNs model proposed by LeCun, researchers regarded this example as “Hello, World” in the field of deep learning. This chapter compares traditional approaches with the CNN model. The dataset of training and testing CNN models here is expanded to the Extension-MNIST (EMNIST) database. It will be employed to pre-train a CNN model for recognizing the handwritten digit image and installation on the iOS device. The user of the presented App can directly write digits on the touchscreen, and the smartphone instantly recognizes what were written. The pre-trained model subject to EMNIST database with a test accuracy of 99.4% has been integrated to an iOS App, termed as handwriting 99 multiplication, which has been successfully published on Apple's App Store.

iOS App and Architecture of Convolutional Neural Networks

Deep convolutional neural networks (CNN) have attracted many attentions of researchers in the field of artificial intelligence. Based on several well-known architectures, more researchers and designers have joined the field of applying deep learning and devising a large number of CNNs for processing datasets of interesting. Equipped with modern audio, video, screen-touching components, and other sensors for online pattern recognition, the iOS mobile devices provide developers and users friendly testing and powerful computing environments. This chapter introduces the trend of developing pattern recognition CNN Apps on iOS devices and the neural organization of convolutional neural networks. Deep learning in Matlab and executing CNN models on iOS devices are introduced following the motivation of combining mathematical modelling and computation with neural architectures for developing pattern recognition iOS apps. This chapter also gives contexts of discussing typical hidden layers in the CNN architecture.

Mobile-Aided Breast Cancer Diagnosis by Deep Convolutional Neural Networks

After verifying the capability of deep learning for basic image recognition, this chapter further extends image recognition to App-aided breast cancer diagnosis. Human cancer has been considered as the most important health problem. For medical image recognition of breast cancer, the presented approach is no longer the same as the traditional. It needs no axioms for distinguishing malignant and benign tumors, and no hand-crafted textural descriptors for feature extraction. The goal is to develop a mobile-aided diagnosis system of directly processing raw medical images. It automatically extracts features and learn filters of a deep CNN subject to labelled medical images in advance. This chapter presents a CNN architecture for diagnosing breast cancer images, illustrating effectiveness of problem solving by designing classifiers, respectively diagnosing lobular carcinoma breast cancer against phyllodes tumor and papillary carcinoma against adenosis. The performances of two classifiers for breast cancers diagnosis are separately summarized by the testing accuracy rates of 94.9% and 87.3%.

Pattern Recognition of Handwritten English Characters

After success of a total solution to handwriting 99 multiplication by deep learning, this chapter further addresses on the problem with increased complexity. In addition to handwritten digital dataset, the EMNIST database provides multiple balanced or unbalanced datasets. These datasets contain different combinations of handwritten digit and letter images. It is believed that well trained deep CNNs can handle unbalanced datasets, so this chapter chose By_Class of EMNIST database as a dataset to increase the difficulty of problem solving and extend the application of iOS Apps. This chapter discusses classification of handwritten English character, including uppercase and lowercase, data audition due to requirement of further improvement, and online tests on iOS devices. After a long time of training, the developer got the pre-trained CNN model. For 58,405 testing images, the recognition accuracy rate was as high as 97.0%.

Deep Learning Theory and Software

In the past decade, deep learning has achieved a significant breakthrough in development. In addition to the emergence of convolution, the most important is self-learning of deep neural networks. By self-learning methods, adaptive weights of kernels and built-in parameters or interconnections are automatically modified such that the error rate is reduced along the learning process, and the recognition rate is improved. Emulating mechanism of the brain, it can have accurate recognition ability after learning. One of the most important self-learning methods is back-propagation (BP). The current BP method is indeed a systematic way of calculating the gradient of the loss with respect to adaptive interconnections. The main core of the gradient descent method addresses on modifying the weights negatively proportional to the determined gradient of the loss function, subsequently reducing the error of the network response in comparison with the standard answer. The basic assumption for this type of the gradient-based self-learning is that the loss function is the first-order differential.