arXiv:2611.23440v1 Announce Type: new Abstract: Selecting less computation in a neural operator need not reduce inference time. We study this gap using conformal calibration to choose Fourier modes or channel width per input subject to a fixed error budget. A triangle-inequality envelope bounds all candidates simultaneously; a direct-score rule handles non-monotone candidate errors and, with an input-independent score, calibrates a static baseline under the same marginal guarantee. Under frozen-model and exchangeability assumptions, both rules bound the probability of issuing a prediction outside the budget and abstain when calibration is infeasible. On Burgers and Darcy flow across three training seeds, mode selection reduces active spectral support without a measurable latency reduction, because full-grid transforms and channel mixing dominate the cost. Narrowing channels does reduce Darcy forward cost, but reference-based selection must first compute the dense prediction. An input-only binary width selector avoids that pass and runs at 0.52-0.95 times dense latency; yet static widths calibrated under the same rule are often as fast or faster, and on a separately generated 1,000-example test set they generally achieve higher coverage, while the selector falls below its 90% target in more than half of the settings. Adaptive computation in neural operators should therefore be judged by achieved coverage and complete runtime against a matched static comparator.
Conformally Calibrated Adaptive Computation in Neural Operators
A study posted as arXiv:2611.23440v1 finds that adaptive computation in neural operators often fails to cut inference time, because full-grid transforms and channel mixing dominate cost. On Burgers and Darcy flow across three training seeds, Fourier mode selection reduced active spectral support without a measurable latency reduction, and an input-only binary width selector ran at 0.52-0.95 times dense latency but fell below its 90% coverage target in more than half of settings, while static widths calibrated under the same conformal rule were often as fast or faster and generally achieved higher coverage on a separately generated 1,000-example test set. The authors conclude adaptive computation in neural operators should be judged by achieved coverage and complete runtime against a matched static comparator.
Benchmark adaptive neural-operator computation against a static width calibrated under the same conformal rule, and report achieved coverage plus full runtime, since mode selection gave no measurable latency gain.
more in playbook: Inference and self-hostingRun your AI side-project on zahid.host
EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.