This article is featured in the free magazine "Data Science in Production – Download here
During Dutch Data Science Week (June 6 – 9), Ivo Everts will present a Signal Processing workshop. This workshop takes place in Amsterdam on Thursday, June 8th and costs €95,- excluding VAT. Click here to register
The workshop is suitable for basically any data scientist that is dealing with diverse data, because not only the amount of data is ever-increasing, but also the diversity.
So, you can think about speech data for speech recognition purposes, or time series in general. Signals from households and image data. We will see how you can do speech recognition, for example how Siri works. And we will answer the question "How you can do image recognition or image manipulation with deep convolutional networks?"
What Are Signals?
A signal is a function that takes a certain value at a certain time or a certain location in the case of images. That means that any mathematical operator can be applied to the function in any of the domains.
Nowadays you have deep learning and convolutional neural networks, with which a lot of problems have been solved and the general performance has increased quite a lot.
But still it is important to understand exactly what is happening inside the models and in the convolutional layers, so that you can apply it in the right way in your setting.
Also, if you don’t have a lot of data, then deep learning usually doesn’t work so well. It is important to understand signal processing techniques, so that if you don’t have a lot of data, you can engineer the features yourself.
The workshop starts with an overview of possible applications, in the domains of time series, speech recognition and image recognition.
Then we will cover some fundamental concepts of signal processing. Most notably we will cover the convolution so you will understand exactly what convolution is and how you can use it. Also, we will talk about Fourier analysis so that we can decompose a signal into its frequencies, and harmonics. Different sine and cosine components of a signal.
Then we will apply those techniques to classify time series data. In this case, we have some time series of decease outbreaks in the US. We will also apply the techniques to extract features from speech data, so that we can identify a speaker, his age or gender.
Finally, we will also apply the signal processing techniques and deep convolutional neural networks to image data for visual recognition.
I will cover quite some theory of signal processing in the first part. In the second part, I will show a lot of practical examples in Python on how you can use those methods and hand it out afterwards so you can apply it yourself.
Click here to register