Learn the basics of Deep Learning with hands on exercises using the Caffe deep learning framework and the DIGITS visual interface. Build your own model and start classifying images.
Begin with a visual understanding of machine learning and deep learning concepts with this quick dive tutorial for beginners.
A hot new topic with lots of opportunities
Artificial intelligence, machine learning and deep learning are in the news and all around us. They give us the promise of computers solving tasks that until recently were very hard for computers: speech recognition, translation, object recognition, image classification, autonomous driving cars.
Caffe framework is free, open sourced, continuously improved, has good documentation and even has an entire zoo of pre trained deep neural network models for image classification and other computer vision tasks. It is very fast and extensible and has most layers and utilities one could hope for (convolutions, pooling, relu, softmax, accuracy) so all you have to do is understand how to control this powerful tool.
DIGITS is NVIDIA's tool to help improve the process of designing, debugging and visualizing the inner workings of a deep neural network and works perfectly with Caffe.
The underling idea is very simple: instead of explicitly programming one should give lots of labeled examples and allow the computer to learn.
Content and Overview
Suitable for beginning deep learning engineers.
Thanks to DIGITS and Caffe there is a little programming and a lot of visual steps but a good mathematical and programming background is recommended.
You should already have Ubuntu (recommended) and DIGITS installed (a fork of Caffe will be included with the DIGITS installation).
The course will take you through the natural steps of getting training and testing data, designing a model, training the model and evaluating it.
Students completing the course will have the knowledge and courage to experiment and create amazing, useful and functional Convolutional Deep Learning Networks.
download mnist, uncompress, read labels
Deep Learning GPU Training System (DIGITS) by NVIDIA
Lightning Memory-Mapped Database (LMDB)
supervized vs unsupervized learning
training vs validation vs testing
web server, jobs
25% for validation
pooling, sub sampling
full connection, inner product
automatic check reader
generalize the lesson to real problem
gradient of the error, stochastic gradient descent
base learning rate
classify unseen handwritten digits
single, multiple images
top N predictions per category
fully connected feedforward neural network
number of outputs
solver mode: GPU
solver type: SGC
caffe log output
variable learn rate
channels == 1
I am a Machine Learning Engineer, Deep Learning Engineer and even an Indie Game Developer with a Major in Compilers and a Master's degree in Artificial Intelligence from University Politehnica of Bucharest.
I am passionate about Games and Artificial Intelligence. I love to give life to A.I. agents in my project or my friend's projects and I want to teach you too.