Journey to learn Machine Learning & Neural Network - Basics
0 likes135 views
This document provides an overview of artificial intelligence, machine learning, and deep learning. It defines artificial intelligence as software that can seem human-like and machine learning as a set of algorithms and methods that allow software to learn from data. It discusses typical machine learning approaches like data preprocessing, feature extraction, model training, and visualization. It also covers different types of neural networks, commonly used frameworks like TensorFlow and PyTorch, and considerations for choosing an activation function and model architecture.
Journey to learn Machine Learning & Neural Network - Basics
- 1. AI, Machine / Deep Learning
- 2. Artificial Intelligence “a broad set of methods, algorithms and technologies that make software 'smart' in a way that may seem human-like to an outside observer. Lynne Parker, NSF ” Machine Learning Deep LearningNLP Concept Univariate Linear Regression Neural Network Named Entity Recognition Ecosystem
- 3. Machine Learning Data Analysis Mathematics & Statistics Data Graphs Tools, Frameworks & Libraries Ecosystem
- 5. Typical ML Approach Feature Extraction Data processing Modules Prepared Data Data Pre- processing Machine Learning Algorithms Training Best Model Structured Data Unstructured Data f(x) ▷ Training • Supervised • Un-Supervised • Semi-Supervised • Reinforced
- 6. ▷Where to get the data from? ▷Type of data ▷Type of features ▷Operations intended Data Visualization
- 7. ▷Supervised ▷Unsupervised Training ▷Semi-Supervised Training ▷Reinforcement Training Types of Training
- 8. Tools Feature extraction / finetuning existing models: Use Caffe Complex uses of pretrained models: Use Torch Write your own layers: Use Torch RNNs: Use Theano or TensorFlow Huge model, need model parallelism: Use TensorFlow Frameworks Caffe Torch Theano Tensorflow Language C++, Python Lua Python Python Pre-trained Models Yes Yes Yes Inception GPU Yes Yes Yes Yes GPU (Parallel Models) No Yes ?? Yes Source Code C++ Lua ?? ?? RNN No Average Yes Yes (Best)
- 9. What is Neural Network ?
- 11. NN Use Case Tools, Frameworks & Libraries -1.0 -.75 0.0 1.0.75.5
- 12. NN Use Case using Linear Functions 0.5 0.0 - 0.75 0.75 + + + + -.2 0.0 0.8 -.5 S S S S S S S S Solid Vertical Diagonal Horizontal Hidden Neurons Input Neurons Output Neurons 1.0 -1.0
- 13. NN Use Case using Linear Functions - 0.75 0.75 -.2 0.0 0.8 -.5 S S S S S S S S Solid Vertical Diagonal Horizontal Hidden Neurons Input Neurons Output Neurons 1.0 -1.0 0.0 0.0 0.0 0.0 0.0 -1.0 -1.0 -1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 RELU
- 14. Is that so straight forward for me to learn?
- 15. NN Use Case ▷What activation function to choose? ▷What NN Architecture should I choose? ▷Activities once model is created
- 16. Any questions ? Thank You!
Editor's Notes
- #4: Journey to this point Data Analysis – Tools: Excel, Access, RDBMS, Orange, R Mathematics – Calculus, Functions, Venn Diagrams Data graphs – Sigmoid, tanh, hyperbole, linear regression, Logistic Regression Tools – Python, Tensorflow, Tensorboard, AWS, Azure, Jupyter Notebooks, Anaconda, Spyder
- #5: Input Data Output / Class / Label Training / Learning rate / Hyperparameters Dataset Epochs, batch size, iterations Neurons Neural Network Activation networks Multilayer Perceptron – Non linear activation networks, Gradient Descent for backpropagation Cross Entropy
- #6: Example of data preparation like DNA, Housing, Flight delays Data pre-processing example on NLP Feature extraction from image and text perspective Type of training – Supervised, Un/Semi Supervised Function or Model Algorithm Model type – Also talk about ADABoost
- #13: Four Features – Solid Vertical Diagonal Horizontal Weights are 1.0, -1.0 and 0.0
- #14: Four Features – Solid Vertical Diagonal Horizontal Weights are 1.0, -1.0 and 0.0 Explain Rectified Linear Unit
- #15: Error Function & Back propagation