Lecture 10 | Recurrent Neural Networks
Why Data Science Stack Exchange recommends it
Data Science Stack Exchange answerers point people to it when asking how to model sequences, time series or video with RNNs and LSTMs.
Recommended by 3 people across 3 different months, 2018 to 2018.
What it is
Lecture 10 of Stanford's CS231n (2017) on recurrent neural networks for sequence data. It covers language modeling, image captioning with soft spatial attention, and the LSTM and GRU architectures, taught by Fei-Fei Li, Justin Johnson and Serena Yeung.
What people said
Recommends Gaurav Kumar · 2018-02 · Data Science Stack Exchange · on “Basic Time Series Classification Examples”
What you need to learn are RNNs and LSTMs. Here are some links I used to learn these: CS231n RNN+LSTM lecture [this video] Understanding LSTMs [LINK] Also I would suggest you to use Keras [LINK] , a Tensorflow API. In my experience, it makes working with RNNs and LSTMs [LINK] way easier, if you're a beginner. I found these articles which seemed related to your problem: Time series classification project by naveen sai on github [LINK] [LINK] [LINK] Keywords to search: Time series classification tensorflow keras
Recommends Gal Avineri · 2018-11 · Data Science Stack Exchange · on “Using classification of previous sample in neural networks”
The approach you describe is predicting base on the features of the current step and the previous states. If you believe that there is information on the previous states, than i would try to approach this problem in the Time Series approach. The algorithms in this approach basically do what you request for, they predict based on the features of the current step and based on the previous states. You can learn about the approach to understand it better in this [this video] cs231n video. You can also look for tutorials on implementing models that fit the time-series problem. The most popular model i know is caleld LSTM . Here [LINK] is a tutorial on implementing LSTM in keras. I hope you will find i
Recommends oezguensi · 2018-12 · Data Science Stack Exchange · on “Analyzing Videos using Deep Learning”
Yes there is. Actually using NNs on sequences is a big part of Deep Learning and one of the most powerful NNs come from this field of research. They are called Recurrent Neural Networks or RNNs . I would recommend you to read this [LINK] article or watch this [this video] Stanford lecture, to learn more about them. Especially a variant of RNNs, the so called LSTM that stands for Long Short Term Memory enables networks to memorize earlier inputs. Based on your explanation I would suggest you have a many-to-one task where you input many frames and output one class. These types of tasks where you input images is most often solved using a combination of CNN and RNN . You could read this [LINK] articl
Cites 1mike12 · 2018-08 · Data Science Stack Exchange · on “In RNNs why do networks always use the last output vs the last input?”
That totally makes sense. Thanks a lot! Also one more thing to add that I just realized from watching this video [this video] , the whole premise of my questions kinda falls apart when there's only one input and the network is trying to generate multiple outputs.