← Back to all news
arXiv Machine Learning By Calvin Jary, Nafiseh Kahani

Selective Next-Location Prediction with MetaModelSelect: A Lightweight Post-Hoc Confidence Metamodel

Read the original on arXiv Machine Learning →

The paper introduces MetaModelSelect, a lightweight post‑hoc confidence metamodel that improves selective next‑location prediction for mobile networks. It uses only the outputs of a frozen recurrent neural network—specifically a learned embedding of the predicted location, the top four softmax probabilities, and Softmax Concentration—to determine whether the top‑1 prediction is correct, without retraining the backbone. Experiments on Changchun cellular mobility trajectories show that MetaModelSelect reduces risk–coverage metrics (AURC, E‑AURC, AUGRC) by up to 4.2% compared to maximum softmax probability and achieves the lowest mean risk across all partial coverages.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Machine Learning.

Related stories

arXiv Machine Learning
Sep 29

Metacognitive Selective Ensemble for Mobile Systems

MetaSE is an active ensemble framework that selects a small set of reliable models for mobile sensing tasks, reducing computational cost while maintaining accuracy. It leverages short-term persistence in model reliability, uses post-execution evidence to prune unreliable members, and only triggers lightweight routing when replacements are needed. Experiments on four human activity recognition datasets and four model architectures show MetaSE outperforms a fixed three-model ensemble and matches the accuracy of more expensive adaptive and full-ensemble approaches, achieving 2.7× speedup and 69% memory savings on a Raspberry Pi 4B.

By Sungmin Lee, Kichang Lee, Joonhee Lee, JaeYeon Park, Songkuk Kim, JeongGil Ko
More like this →
arXiv Machine Learning
Sep 7

BER-PEF: Unified Human Mobility Predictability Evaluation via Bayes Error Rate Estimation

BER-PEF is a Bayes‑error‑rate‑based framework that transforms BER estimation into human mobility predictability estimation, enabling comparison of different estimators even when ground truth predictability is not observable. It maps various data types—symbolic sequences, numeric trajectories, contextual features, and learned representations—into a shared feature–label space and evaluates estimator outputs along controlled perturbation curves against a common predictability reference interval. Experiments on datasets such as Foursquare NYC/TKY, GeoLife, and T‑Drive show that several BER‑based estimators outperform existing methods on symbolic sequences and numeric trajectories, and that aggregating evidence across multiple perturbation levels yields a more reliable basis for selecting estimators.

By En Xu, Jingtao Ding, Zhiwen Yu, Yong Li
More like this →