Deep learning is becoming increasingly powerful in its applications, from speech recognition to dog recognition, from AlphaGo to self-driving cars. It's also becoming easier and easier to jump in and use deep neural networks, with frameworks like Keras designed to facilitate fast and easy prototyping whilst remaining user friendly.
This workshop will be a (very) brief introduction to deep neural networks with a focus on using convolutional neural networks for image classification and object recognition. No prior knowledge is assumed, and we'll be starting our journey in the mid-1950s with the concept of artificial neurons, moving swiftly onto perceptrons, multi-layer perceptrons, and finally into the recent advances in deep learning. The workshop will be hands-on throughout with code examples and exercises provided to apply what you've just learned.
Pre-requisites & preparation
--Students should have familiarity with Python. Background in linear algebra is preferred but not required. For those wishing to dig a little deeper, brilliant.org has a great free course on Artificial Neural Networks, and the "Math for Neural Networks" module is especially recommended.
- Sign up for IBM Watson Studio. Here you can create a project and run
the workshop notebooks.
Sign up at https://dataplatform.cloud.ibm.com/
- Or run the workshop locally by installing Anaconda. We will be using
the Anaconda distribution of Python 3.6. The download comes pre-
installed with a lot of the libraries required for machine learning. You
can get this from https://www.anaconda.com/download/
-- Install TensorFlow. Installing TensorFlow can be as simple as conda install tensorflow, although Google recommends installing the officially supported/tested version from the Python Package Index (PyPI) using pip install tensorflow. Regardless, there are full and up-to-date installation instructions here: https://www.tensorflow.org/install/. If you have GPUs on your machine, then feel free to install the TensorFlow with GPU support (e.g. pip install tensorflow-gpu). Note that there are a number of pre-requisites, such as CUDA Toolkit 9.0, the associated NVIDIA drivers, cuDNN v7.0 and a GPU card with CUDA Compute Capability 3.5 or higher (this is all detailed on TensorFlow's installation guide). You do not need to enable GPU support for this workshop.
-- Install Keras. This should be as simple as pip install keras from within your Anaconda environment. For more information please see: https://keras.io/#installation
Speaker/Instructor -John Sandall, Founder and Principal Data Scientist at Coefficient, a data consultancy offering data science, engineering, machine learning and other AI-related services as well as bespoke training courses. In April 2017 he created SixFifty.org.uk in order to predict the UK General Election using open data and advanced modelling techniques and is currently a Fellow of Newspeak House. For more information please visit artofinference.com or follow him on Twitter @john_sandall
Agenda - 18.30 - 19.00 : Registration, Food, Drinks and Networking
19.00 - 21.00 : Workshop