# Conformal Inference

3 related articles

Subscribe to updates

arXiv Nov 12, 2026
Fractional-grid forecasts split in two: learned frequency prediction, separately reconstructed memory for the error bound
A fractional grid's observation window misses earlier motion, and that history reaches the future through a memory force. The study separates frequency point prediction from memory reconstruction as two learning objectives: across six-node, ten-cycle, ten-mesh and twenty-mesh cases, a response-trained linear predictor refined by a neural residual lowers frequency RMSE relative to that linear predictor by 0.074 to 0.104 at the primary noise level. A separately reconstructed force serves as the reference for an energy inequality plus an angle test, issuing frequency envelopes in cases where the point predictor alone cannot support them.
A fractional grid's observation window misses earlier motion, and that history reaches the future through a memory force. The study separates frequency point prediction from memory reconstruction as two learning objectives: across six-node, ten-cycle, ten-mesh and twenty-mesh cases, a response-trained linear predictor refined by a neural residual lowers frequency RMSE relative to that linear predictor by 0.074 to 0.104 at the primary noise level. A separately reconstructed force serves as the reference for an energy inequality plus an angle test, issuing frequency envelopes in cases where the point predictor alone cannot support them.
이환서（First author）
Fractional Kuramoto Grid Frequency Prediction Memory Force Learning Nonlinear Energy Bound Conformal Inference
arXiv Oct 5, 2026
Online conformal calibration makes coverage steerable in RL-driven hardware-aware NAS, pruning 25-50% of evaluations at no measured accuracy cost
The work replaces one-shot quantile estimation with online feedback control (Adaptive Conformal Inference with tuning-free, locally-adaptive, and group-conditional variants) to address non-exchangeability of candidates during reinforcement-learning search, tracking requested coverage to within about 10^-3 across three architecture families and both single-step and sequential search (three seeds) while pruning 25-50% of evaluations at no measured accuracy cost, whereas static calibration loses coverage control and a Gaussian-process baseline stays conservative.
The work replaces one-shot quantile estimation with online feedback control (Adaptive Conformal Inference with tuning-free, locally-adaptive, and group-conditional variants) to address non-exchangeability of candidates during reinforcement-learning search, tracking requested coverage to within about 10^-3 across three architecture families and both single-step and sequential search (three seeds) while pruning 25-50% of evaluations at no measured accuracy cost, whereas static calibration loses coverage control and a Gaussian-process baseline stays conservative.
Pedro Brandimarte（First author）
Adaptive Conformal Inference Online Calibration Neural Architecture Search Conformal Inference
arXiv Oct 5, 2026
CISE constrains self-evolving search with conformal interval rewards, returning only true positives under high-fidelity evaluation across three materials-science tasks
The work proposes Conformal Interval-Driven Self-Evolution (CISE), which builds candidate-specific reward intervals via conditional conformal inference and iteration-wise online density-ratio estimation, uses conservative interval-based rewards for evolutionary feedback, and returns candidates only when all required property intervals lie entirely within their feasible regions; across three self-evolving search tasks in materials science, all candidates returned by CISE are true positives under high-fidelity evaluation, whereas baselines return more candidates but include false positives.
The work proposes Conformal Interval-Driven Self-Evolution (CISE), which builds candidate-specific reward intervals via conditional conformal inference and iteration-wise online density-ratio estimation, uses conservative interval-based rewards for evolutionary feedback, and returns candidates only when all required property intervals lie entirely within their feasible regions; across three self-evolving search tasks in materials science, all candidates returned by CISE are true positives under high-fidelity evaluation, whereas baselines return more candidates but include false positives.
Yonsei University
Kangjun Noh（First author）
Conformal Interval-Driven Self-Evolution Self-Evolving Search Conformal Inference Large Language Model

You’re all caught up

Fractional-grid forecasts split in two: learned frequency prediction, separately reconstructed memory for the error bound

Online conformal calibration makes coverage steerable in RL-driven hardware-aware NAS, pruning 25-50% of evaluations at no measured accuracy cost

CISE constrains self-evolving search with conformal interval rewards, returning only true positives under high-fidelity evaluation across three materials-science tasks