Product Details

AI610 AI620 AI625 ABB Analog Input

This course has a two-fold purpose: first it presents the fundamental aspects of Artificial Neural Networks (ANN) and second it introduces to more advanced topics such as Deep Learning (DP) Networks. After a short review of conventional neural networks and learning processes, the course will introduce modern practices for deep networks, including training, optimization, convolutional networks, recurrent and recursive nets. Furthermore, the course will focus on practical methodologies concerning the design, data preprocessing, hyperparameter selection, implementation and performance evaluation of a Deep Learning system as well as applications of deep learning techniques to real world problems, such big data mining, image processing and natural language processing.

Upon succesful completion of this course students should be able to:  Recall the basic algorithms and methods of Artificial Neural Networks basics as well as their training algorithms.  Discuss, explain and report various deep learning algorithms for a specific problem.  Choose and interpret a suitable algorithm/method in order to meet problem’s specifications.  Analyze and combine known technics in order to face real world problems.  Analyze and preprocessing the given data to fit in ANN and DL algorithms/methods.  Put the different system parts (preprocessed data, implemented, algorithms, user interface) together in order to create a new operational learning system.  Evaluate the performance of the developed learning system.  Discuss the fundamental concepts of several types of deep neural networks.  Apply Deep Learning approaches to a variety of tasks.

Our Email: 2235954483@qq.com

contact number:13313705507

contacts：HE

Machine Learning Basics: Presentation of some perspectives on traditional machine learning techniques, such as Neural Networks (NNs) that have strongly influenced the development of deep learning algorithms. After a short introduction to NNs, we will discuss the model of a neuron, and network architectures. Then, the different types of learning processes are presented. Finally, the several aspects concerning the training of a single layer perceptron are discussed. Deep Feedforward Networks: Presentation of deep learning neural network modes for function approximation. A simple learning example and the gradient-based learning are discussed, as well as other aspects like hidden units and architecture design. Then follows the foundation of the Back Propagation algorithm for deep learning and its variations. The relative algorithms are analyzed extensively, and implementation aspects are discussed. Regularization for Deep Learning: Presentation of selected advanced techniques for regularization and optimization of deep network models, such as parameter’s norm penalization, norm penalties as constrained optimization and dataset augmentation. Furthermore, the semisupervised learning paradigm and feature extraction techniques, bagging and ensemble methods are discussed. Optimization for Training Deep Models: Several challenges of training optimization, such as parameter optimization and adaptive learning rates will be discussed. Furthermore, the relative algorithms are presented and analyzed as well as optimization strategies and metaalgorithms. Convolutional Neural Networks (CNNs): Introduction to convolutional networks for scaling to large data sets. Presentation of the main building blocks of CNNs such as convolutional filters and their characteristics (stride, depth, width), activation functions, pooling operator. Several aspects of convolution operation are discussed and efficient algorithms for random or unsupervised features are presented as well as the neuroscientific basis of convolution networks. Sequence Modeling: Recurrent and Recursive Neural Nets (RNNs): Deep recurrent and recursive neural networks for temporal sequences processing are presented. The challenge of long-term dependencies, the description of long sort-term memory and other gating mechanisms as well as optimization aspects are discussed. Practical Methodology: General guidelines for the practical methodology involved in designing, building and configuring an application involving deep learning are discussed. These aspects include performance metrics, baseline models, gathering more data, hyperparameters’ selection and debugging strategies. An example shows how these aspects are faced.