際際滷

際際滷Share a Scribd company logo

Introduction to machine learning november 25, 2017

?Download as PPTX, PDF?
?
Manish Panchmatia
Manish Panchmatia

introductory talk about AI/ML along with Saurabh Bansal at ＾Applied Singularity￣ meetup. https://www.meetup.com/AppliedSingularity/events/244992138/

1 of 23
Machine Learning Introduction by Manish Panchmatia & Saurabh Bansal
? AI = New Electricity By Andrew NG ? What was meaning of CEO? ? Chief Engineering Officer ? Chief Electrical Officer ? New Product = Any Product + Electricity ? Now New Product = Any Product + AI
Some more Quotes about AI ? AI + Product means People should be able to interact with computing in a natural and seamless way. C Sundar Pichai ? The world's first trillionaire will be an AI entrepreneur C Mark Cuban ? Predicting the future isn¨t magic, it¨s artificial intelligence. - Dave Waters
AI Examples Auto Tagging Self Driving Car Translation E-commerce Recommendation Gmail Smart Reply q Windows Cortana , Apple Siri : Virtual Assistant Prisma photo editor : Image to painting https://www.youtube.com/watch?v=Y2VF8tmLFHw
Types of Data
AI ? ML ? DL ? Artificial Intelligence, deep learning, machine learning? ！?whatever you¨re doing if you don¨t understand it ！?learn it. Because otherwise you¨re going to be a dinosaur within 3 years. - Mark Cuban
AI C ML Techniques
Algorithm Cheat Sheet
ML WorkFlow
Data Analysis Tools & Languages
Machine Learning Basics
Neural Network Basics
CNN for Image Classifications
Introduction to machine learning november 25, 2017
Convolutional Neural Network 1. CNN is a feed-forward network that can extract topological properties from an image. 2. Like almost every other neural networks they are trained with a version of the back-propagation algorithm. 3. Convolutional Neural Networks are designed to recognize visual patterns directly from pixel images with minimal preprocessing. 4. They can recognize patterns with extreme variability (such as handwritten characters).
Deep Learning 1. Deep Learning can contain a number of layers of Neural Networks. 2. The input to each layer (two-dimensional arrays) looks a lot like the output (two-dimensional arrays) from previous layer. 3. Each layer have a number of steps like Pooling, Convolution, Rectification, Normalization, etc.
Convolution Suppose we have a problem, we need to select whether input image is a ｀X¨ or ｀O¨
Convolution So we will get maximum outputs at the pixels, where the pattern matches with filter pixels.
Pooling Pooling involves stepping a small window across an image and taking the maximum / average value from the window at each step called Strides. So 2 types of pooling: - Average Pooling - Max Pooling
Rectified Linear Units (ReLU) 1. ReLU is most used Activation function in Neural Networks. 2. It¨s math is very simple！ wherever a negative number occurs, swap it out for a 0. 3. This helps the CNN stay mathematically healthy.
Fully connected layers 1. Fully connected layers take the high-level filtered images and translate them into categories.
CNN with MNIST Dataset https://github.com/Hvass-Labs/TensorFlow- Tutorials/blob/master/02_Convolutional_Neural_Network.ipynb
Thank You

More Related Content

Introduction to machine learning november 25, 2017

  • 1. Machine Learning Introduction by Manish Panchmatia & Saurabh Bansal
  • 2. ? AI = New Electricity By Andrew NG ? What was meaning of CEO? ? Chief Engineering Officer ? Chief Electrical Officer ? New Product = Any Product + Electricity ? Now New Product = Any Product + AI
  • 3. Some more Quotes about AI ? AI + Product means People should be able to interact with computing in a natural and seamless way. C Sundar Pichai ? The world's first trillionaire will be an AI entrepreneur C Mark Cuban ? Predicting the future isn¨t magic, it¨s artificial intelligence. - Dave Waters
  • 4. AI Examples Auto Tagging Self Driving Car Translation E-commerce Recommendation Gmail Smart Reply q Windows Cortana , Apple Siri : Virtual Assistant Prisma photo editor : Image to painting https://www.youtube.com/watch?v=Y2VF8tmLFHw
  • 6. AI ? ML ? DL ? Artificial Intelligence, deep learning, machine learning? ！?whatever you¨re doing if you don¨t understand it ！?learn it. Because otherwise you¨re going to be a dinosaur within 3 years. - Mark Cuban
  • 7. AI C ML Techniques
  • 10. Data Analysis Tools & Languages
  • 15. Convolutional Neural Network 1. CNN is a feed-forward network that can extract topological properties from an image. 2. Like almost every other neural networks they are trained with a version of the back-propagation algorithm. 3. Convolutional Neural Networks are designed to recognize visual patterns directly from pixel images with minimal preprocessing. 4. They can recognize patterns with extreme variability (such as handwritten characters).
  • 16. Deep Learning 1. Deep Learning can contain a number of layers of Neural Networks. 2. The input to each layer (two-dimensional arrays) looks a lot like the output (two-dimensional arrays) from previous layer. 3. Each layer have a number of steps like Pooling, Convolution, Rectification, Normalization, etc.
  • 17. Convolution Suppose we have a problem, we need to select whether input image is a ｀X¨ or ｀O¨
  • 18. Convolution So we will get maximum outputs at the pixels, where the pattern matches with filter pixels.
  • 19. Pooling Pooling involves stepping a small window across an image and taking the maximum / average value from the window at each step called Strides. So 2 types of pooling: - Average Pooling - Max Pooling
  • 20. Rectified Linear Units (ReLU) 1. ReLU is most used Activation function in Neural Networks. 2. It¨s math is very simple！ wherever a negative number occurs, swap it out for a 0. 3. This helps the CNN stay mathematically healthy.
  • 21. Fully connected layers 1. Fully connected layers take the high-level filtered images and translate them into categories.
  • 22. CNN with MNIST Dataset https://github.com/Hvass-Labs/TensorFlow- Tutorials/blob/master/02_Convolutional_Neural_Network.ipynb