Hi everyone, I'm new to research and would like to start writing my first paper. I have some ideas, but I'm unsure how to make them a publishable work.
Some specific questions I struggle with:
Some specific questions I struggle with:
- Novelty: I've been told my work may not be novel enough. For example, I try combining existing designs and replacing one thing with another (for example: replacing the UNet in a VAE with a vision transformer)
- Math: Many papers I read modify loss functions and then provide long appendices with proofs. To be honest, I'm not that good with math, I even struggle understand those proofs, should I focus on improving this before attempting to write, or is it okay to start with more empirical work?
- Results: Published papers often outperform existing benchmarks or reduce the training/sampling time significantly with only a small trade off in quality. I sometimes struggle just to implement my own ideas, so I'm not sure how to realistically compare against strong baselines. How do beginners approach this?
