The ICML rebuttal process is brutal, and the "novelty" accusation is the most frustrating weapon in a reviewer's arsenal. You have done the hard work. You have outperformed baselines, including ones your method should not have beaten. You have identified the exact reasons for your success, and your field is calling the results groundbreaking. Yet the reviewer keeps circling back to a claim that is not just wrong, but a strawman. The instinct to argue semantics will fail you. The better move is to refuse the frame entirely.
The reviewer is not asking you to prove novelty. They are asking you to justify your existence in a system that rewards flashy new components over rigorous, surprising results. But your evidence is the story. You combined existing components in a way that was never attempted in your domain, and you introduced new components that the reviewers are dismissing out of hand. That is not a lack of novelty. That is a failure of the reviewer to engage with the substance. Your rebuttal should not beg for their approval. It should lay out the evidence as a direct challenge: here is what we did, here is why it works, and here is why no one else has done it.
Practical terms matter here. When you respond, do not write a paragraph defending the word "novel." Write a table or a bulleted list that contrasts what existed before, what you combined, and what the unexpected outcome was. Point to the baselines you outperformed that you should not have. That is not a side note. That is the core of your contribution. The reviewers are not asking you to prove that your method is entirely without precedent. They are asking you to show that your method is not obvious in hindsight. Your results are surprising. The fact that you can explain them precisely makes the contribution stronger, not weaker. Unexplained magic is suspect. Explained magic is science.
So here is your task. In the rebuttal, stop using the word "novel" altogether. Replace it with "unexpected," "previously uncombined," and "empirically demonstrated." Force the reviewers to argue with your results, not your vocabulary. If they still insist on the novelty strawman, then you have done your job. The evidence is on your side. The rest is just a reviewer who has already made up their mind, and no amount of adjectives will change that. Your time is better spent preparing for the next submission, where the work will speak for itself.