Intro and Overview Machine Learning Lecture
Contents
Intro and Overview Machine Learning Lecture#
Author: Prof. Dr. Johannes Maucher
Email: maucher@hdm-stuttgart.de
Last Update: October, 4th 2021
Goals of this Lecture#
Goals of this lecture are:
Understand the basic concepts, procedures and categories of Machine Learning
For any task or problem: Know, which algorithm-type is suitable for this task
Understand conventional ML-algorithms
Understand Deep Learning
Understand the currently most important types of Deep Neural Networks
Implementations are provided in order to support understanding
Contents#
Intro#
Basic Concepts of Data Mining and Machine Learning
Definition
Categories
Validation
Conventional ML Algorithms#
Neural Networks and Deep Neural Networks#
-
Natural Neuron
General Notions for Artificial Neural Networks
Single Layer Perceptron (SLP)
Architectures for Regression and Classification
Gradient Descent- and Stochastic Gradient Descent Learning
Gradient Descent- and Stochastic Gradient Descent Learning
Multilayer Perceptron (MLP) Architectures for Regression and Classification
Backpropagation-Algorithm for Learning
Recurrent Neural Networks (RNN)
Simple Recurrent Neural Networks (RNNs)
Long short-term Memory Networks (LSTMs)
Gated Recurrent Units (GRUs)
Application Categories of Recurrent Networks
Deep Neural Networks: Convolutional Neural Networks (CNN)
Overall Architecture of CNNs
General concept of convolution filtering
Layer-types of CNNs:
Convolution,
Pooling,
Fully-Connected
MLP and CNN for Object Classification
Example Data: Cifar-10 Image Dataset
Image Representation in numpy
Define, train and evaluate MLP in Keras
Define, train and evaluate CNN in Keras
Apply pretrained CNNs for object classification - original task
Access image from local file system
Download and apply pretrained CNNs for object recognition in arbitrary images
-
Download pretrained feature-extractor (CNN without the classifier part)
Define new classifier architecture and concatenate it with pretrained classifier
Fine-tune network with task-specific data
Apply the fine-tuned network for object-recognition
Autoencoder#
GAN#
Reinforcement Learning#
Modelling of Text#
Modelling of Words and Texts / Word Embeddings
Concept of Word-Embeddings
Skip-Gram and CBOW
Working with pretrained word-embeddings
Text Classification with CNNs and LSTMs
Example Data: IMDB-Movie Reviews for Sentiment Classification
Text preprocessing and representation with Keras
Load and apply pretrained word-embedding
News classification with CNN
News classification with LSTM
Graph Neural Networks#
-
Concepts of GNNs
Implementation of GNN with Keras
Document Classification with GNNs