With the rise of Big Data, the diversity of data has exploded well beyond samples of a fixed number of interpretable variables. Now data includes free text, time series (financial transactions, power usage), audio (speech), images and video. These "signals" usually must be processed so that meaningful variables can be extracted and structured before they can be used in data analyses and machine learning applications.
This training teaches participants how to make powerful data representations from signals for machine learning applications. It covers feature engineering and feature learning in two consecutive sessions and provides Python code for both. Through instructor-led discussion and interactive, hands-on exercises, you will learn how to apply signal processing techniques to unstructured data.
Q: Is Signal Processing for Data Science training right for me?
- Yes - if you are a data scientist with knowledge and/or experience in applying machine learning with Python (e.g. Numpy, Scipy, Scikit-learn, Pandas).
Q: What will I achieve by completing this training?
You will learn:
- How to create data representations from unstructured data with signal processing techniques such as convolution and Fourier analysis
- How to model your unstructured data with the bag-of-words model
- How to perform feature learning for dimensionality reduction
- How to perform end-to-end training of deep convolutional networks
You will gain hands-on experience in:
- Processing unstructured data using feature engineering and machine learning techniques in Python with NumPy, Pandas, Sklearn and Keras
- Performing image, speech, and time series classification using a variety of machine learning techniques
Q: What else should I know?
You will need your own laptop with the following requirements:
- A Python3 distribution with the numerical processing stack (Numpy, etc.) should be installed, for example with Anaconda. You should also have permission to install more packages.
- Keras is used for deep learning, so please install it in advance (preferably with a TensorFlow backend).