Signal Processing Techniques for Deep Learning on Sensor Data

How can you use sharp time-frequency techniques to enhance the information present in the signals and subsequently use deep Convolutional Neural Networks to enable you to build predictive models which can be used for tasks like signal classification
_________________________________________________________________________
Avinash Nehemiah, Product Manager for Deep Learning Computer Vision and Automated Driving, MathWorks
_________________________________________________________________________
לאתר שלנו: www.systematic...
לאירועים נוספים שלנו: www.systematic...
לבלוג שלנו: www.systematic...
לפייסבוק שלנו: / matlabisrael
ללינקדין שלנו: / 1774805

[@ShantoShanto]

Very good presentation. I was looking for a lecture on application of deep learning in the signal processing space.

[@SumaSuma-gn6rz]

E

[@rajdeepkumarnath8944]

Very nice presentation. However, I have some concerns about the second use-case. What about the test accuracy? Because any deep learning model can get fairly good accuracy in the training phase especially if it has to distinguish between only three classes. For example, in one of my work, I have seen that using just one preprocessing layer, one dense layer, and one classification layer, the training accuracy was above 90% in detecting three levels of stress from ECG signal. In that work, I did not do any preprocessing manually. I just took the noisy raw signal, and dumped them in the neural network. So my point is that, for medical data such as physiological signals, analyzing the training accuracy for evaluating performance is not a good idea because usually they mean nothing. For example, you can have 90% accuracy on training data, but on test data, your model might as well just be performing random guess.

[@merveozdas1193]

yes İ agree with you, I need to distinguish eeg channels, I dont know how can I do so that in th real time seperating