Here is the brief of last night's event:
Jakob Aungiers talked about Long Short-Term Memory (LSTM) Neural Networks for Time Series Prediction expanding on his article LSTM NEURAL NETWORK FOR TIME SERIES PREDICTION.
He ran his Python code based on Keras package and demonstrated trends prediction of sin wave and stock data.
Ajit played video which shows work of Sandro Bovelli of Sentetic where robot learns moves by "observing" human's moves via attached sensors.
I drew a parallel between LSTM NN and low-pass digital filters - both are removing high-frequency noise and have periodical signal of low frequency as the output. Digital filters are applied only to electrical signals but LSTM NN can be applied on any type of discrete data and also they are multidimensional - NN have multiple data channels (features) as inputs.
Jakob Aungiers talked about Long Short-Term Memory (LSTM) Neural Networks for Time Series Prediction expanding on his article LSTM NEURAL NETWORK FOR TIME SERIES PREDICTION.
He ran his Python code based on Keras package and demonstrated trends prediction of sin wave and stock data.
Ajit played video which shows work of Sandro Bovelli of Sentetic where robot learns moves by "observing" human's moves via attached sensors.
I drew a parallel between LSTM NN and low-pass digital filters - both are removing high-frequency noise and have periodical signal of low frequency as the output. Digital filters are applied only to electrical signals but LSTM NN can be applied on any type of discrete data and also they are multidimensional - NN have multiple data channels (features) as inputs.
No comments:
Post a Comment