R-N/ml-utility-loss

Synthesizer Tuning

Opened this issue · 0 comments

R-N commented
  • LCT-GAN performs rather poorly even after 1k tuning trials. Indeed, the author was selling its convergence speed rather than overall performance.
  • RealTabFormer training takes a very long time, so does the tuning. Maybe we'll just use default parameters as the author did in the evaluation.