Questions from machine learning. I need detailed answers...

1. Once a dimensionality reduction method has been trained, can this operation also be reversed? 2. Does it make sense to run two different dimension reduction methods one after the other? 3. Suppose we reduce the dimension of a 1000-dimensional data set using PCA so that 95% of the variance is explained. To which dimension will the data set be reduced? 4. How can you measure the performance of a dimension reduction method? 5. What is the difference between hard and soft voting