I was successful in replicating the results using the provided MNIST dataset; however, I faced challenges in reproducing the outcomes with other datasets using default parameters. Specifically, I used the default structure of AutoEncoder as the code given on https://github.com/shahsohil/DCC. I followed the same training process as the tutorial on github shows. I used the same commands for pretrainning as those used for MNIST for other datasets, but the results were not as expected. Here is the ami results I got, the first is our result, the second it the reuslt on the paper: YTF: 0.69 | 0.88; Yale: 0.11 | 0.96; reuters: 0.02 | 0.57; RCV1: 0.04: 0.50.
What is the the training parameters used during your experiments?
I was successful in replicating the results using the provided MNIST dataset; however, I faced challenges in reproducing the outcomes with other datasets using default parameters. Specifically, I used the default structure of AutoEncoder as the code given on https://github.com/shahsohil/DCC. I followed the same training process as the tutorial on github shows. I used the same commands for pretrainning as those used for MNIST for other datasets, but the results were not as expected. Here is the ami results I got, the first is our result, the second it the reuslt on the paper: YTF: 0.69 | 0.88; Yale: 0.11 | 0.96; reuters: 0.02 | 0.57; RCV1: 0.04: 0.50.
What is the the training parameters used during your experiments?