As far as I can tell, neural networks have a fixed number of neurons in the input layer.
If neural networks are used in a context like NLP, sentences or blocks of text of varying sizes are fed to a network. How is the varying input size reconciled with the fixed size of the input layer of the network? In other words, how is such a network made flexible enough to deal with an input that might be anywhere from one word to multiple pages of text?
If my assumption of a fixed number of input neurons is wrong and new input neurons are added to/removed from the network to match the input size I don't see how these can ever be trained.
I give the example of NLP, but lots of problems have an inherently unpredictable input size. I'm interested in the general approach for dealing with this.
For images, it's clear you can up/downsample to a fixed size, but, for text, this seems to be an impossible approach since adding/removing text changes the meaning of the original input.
The term 'neural networks' encompasses many different architectures. The one that you are describing is also known as Multilayer perceptron (MLP), and yes, it has a fixed input size, so you cannot use that configuration with inputs of different sizes. In fields like NLP, or others that deal with sequences of data, they use Recurrent Neural Networks. This is a different configuration that establishes relationships between elements in a sequence (for example words, time sequences, etc...). They allow varying input sizes. Finally, the most common solution for images is Convolutional Neural Networks. As you mention, you can up/downsample to a fixed size, but you can also have varying input sizes in these networks (without resampling the original image). The Deep Learning Book explains all these different architectures, and why you can use them with varying input sizes https://www.deeplearningbook.org/
The term 'neural networks' encompasses many different architectures. The one that you are describing is also known as Multilayer perceptron (MLP), and yes, it has a fixed input size, so you cannot use that configuration with inputs of different sizes. In fields like NLP, or others that deal with sequences of data, they use Recurrent Neural Networks. This is a different configuration that establishes relationships between elements in a sequence (for example words, time sequences, etc...). They allow varying input sizes. Finally, the most common solution for images is Convolutional Neural Networks. As you mention, you can up/downsample to a fixed size, but you can also have varying input sizes in these networks (without resampling the original image). The Deep Learning Book explains all these different architectures, and why you can use them with varying input sizes https://www.deeplearningbook.org/
In addition to Roberto Vega 's response, as an example, in keras for constructing a Recurrent Neural Network pipeline, you can specify variable-length inputs. Empirically, you are able to pad (masking) your shorter sequences with zeros since all inputs to Keras models must be tensors.
The number of neurons in the output of the neural network (NN) or convolutional neural network is fixed. They cannot be altered once is network is designed. To deal with the varying number of outputs different approach has to be followed. One basic idea will be to use multiple networks each with a different number of output nodes.
I have worked on a fingertips detection and the output has a different number of visual fingers depending upon hand gesture. This problem was somewhat similar to your problem. To deal with the varying number of finger case, I designed a NN with neuron having maximum possible outputs and used a second output to identify how many fingers are visual. Combining these two output I obtained my final result.
The detail of which can be found in my paper titled " Fingertips Detection in Egocentric Video Frames using Deep Neural Networks ".
You may get some ideas for your problem by following the above paper.
I hope this answer helps you.
Thank you Purnendu Mishra Javad PourMostafa Roshan Sharami Roberto Vega . This will be helpful
Have a look at
https://stats.stackexchange.com/questions/388859/is-it-possible-to-give-variable-sized-images-as-input-to-a-convolutional-neural
- Badges
- Science topic