From Physics to ML: Turning Revised Scores Into a Confident Resubmit

Navigating the transition from theoretical physics to machine learning can be daunting, especially when preparing for conferences like ICML 2026.

3 min readMachine Learning

The math here is better than most people realize, and that is exactly why this resubmission deserves more confidence than the poster is allowing themselves. Moving from a 4333 to a 4433 after the first review cycle is not a coin flip or a hopeful guess. It is a direct response to specific, actionable feedback. Two reviewers said, in effect, "do this and we will raise our score." The poster did that. That is not a gamble; that is a completed transaction waiting for the paperwork.

What stands out is the clarity of the remaining risk. The two weak rejects did not say "this work is fundamentally flawed." They said "add a parameter sweep" and "answer these questions properly." Both have been addressed. The 4433 already reflects that the other reviewers saw the revisions and moved up. So the realistic outcome is not a 30 to 40 percent chance of a 4444. It is a near certainty of at least a 4443, with the 4444 sitting exactly where the poster placed it, as a plausible upside rather than a desperate hope. The mistake would be treating a 4443 as a failure when it is a clear acceptance in most deep learning theory venues.

The deeper point here is about how the poster is framing the transition from physics to ML. In physics, the review process often feels like a gauntlet where the goal is to survive. In ML conferences, the reviewers are telling you what they need to be convinced. That is a gift. The poster has already learned to read that signal, which is more than many seasoned ML researchers manage. The funding concern is real, but it is a separate problem from the paper's fate. The paper is in good shape. The funding application should be built on that confidence, not held hostage to a worst-case scenario that the evidence does not support.

The practical takeaway is simple: when reviewers tell you exactly what they want and you give it to them, you have done the hard part. The remaining uncertainty is not about the quality of the work. It is about whether the other reviewers will honor their word. That is not a 30 percent bet. That is a near sure thing with a bonus round attached. So the poster should submit, secure the funding, and stop treating a 4443 as a consolation prize. It is the result of listening, revising, and moving the needle. That is how the game is won.

From Machine Learning

Hi, I am currently making the jump to ML from theoretical physics. I just got done with the review period, went from 4333 to 4433, but the remaining two weak rejects said 1) that if I add a parameter sweep and a small section (which I did) they’d raise, and the other reviewer said that if some of their questions were addressed properly they’d also raise the score. I think the most likely outcome is hopefully 4443, but with maybe a 30-40% chance of 4444. The area is deep learning theory. I have never been through the process of applying…

Read the original at Machine Learning