Pre-training of Recurrent Neural Networks via Linear Autoencoders

Authors: Luca Pasa, Alessandro Sperduti

NeurIPS 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Using four well known datasets of sequences of polyphonic music, we show that the proposed pre-training approach is highly effective, since it allows to largely improve the state of the art results on all the considered datasets.
Researcher Affiliation Academia Luca Pasa, Alessandro Sperduti Department of Mathematics University of Padova, Italy {pasa,sperduti}@math.unipd.it
Pseudocode Yes Algorithm 1 shows in pseudo-code the main steps of our procedure.
Open Source Code No The paper does not state that the code for their proposed linear autoencoder pre-training methodology is open-source or provide a link to it. It only mentions using a third-party Theano-based software for RNN training.
Open Datasets Yes In order to evaluate our pre-training approach, we decided to use the four polyphonic music sequences datasets used in [21] for assessing the prediction abilities of the RNN-RBM model.
Dataset Splits Yes Each dataset is split in training set, validation set, and test set. Statistics on the datasets, including largest sequence length, are given in columns 2-4 of Table 1.
Hardware Specification Yes Time in seconds needed to compute pre-training matrices (Pre-) (on Intel c Xeon c CPU E5-2670 @2.60GHz with 128 GB) and to perform training of a RNN with p = 50 for 5000 epochs (on GPU NVidia K20).
Software Dependencies No The paper mentions 'Theano-based stochastic gradient descent software' but does not provide a specific version number for Theano or any other software dependency.
Experiment Setup Yes Our pre-training approach (Pre T-RNN) has been assessed by using a different number of hidden units (i.e., p is set in turn to 50, 100, 150, 200, 250) and 5000 epochs of RNN training