Experimental results # 2

* The Effect of Incorporating Synthetic Data:

Due to resource limitations, it is challenging to systematically evaluate the effect of training with synthetic data. In our experiments, we trained the model for only one epoch and used a small training corpus, which restricts the scope of our analysis.

 

The primary goal of incorporating synthetically generated time series data into the training corpus is to enhance both the diversity and volume of the dataset. However, for this strategy to be effective, the overall size of the training dataset must be sufficiently large. In our case, the training dataset is relatively small. Thus, when synthetic data is added, it reduces the proportion of real data, which may lead to a loss of diversity. We hypothesize that the observed degradation in performance when adding synthetic data - contrary to the results of previous studies - stems from the small size of our training set and the limited number of training steps.