---
title: "GARLIC: Graph Attention-based Relational Learning of Multivariate Time Series in Intensive Care"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2608.10969v1"
permalink: "https://feed.astralor.com/items/itm_549f96cefcdad4bd3b0b2589"
published: "2026-08-11T14:29:48.000Z"
score: 71
channel: "tech"
tags: ["GARLIC","Explainable Deep Learning","Graph Attention Network","Missing Data Imputation","Intensive Care"]
---

# GARLIC: Graph Attention-based Relational Learning of Multivariate Time Series in Intensive Care

## AI commentary

> What stands out here is that interpretability is baked into the architecture itself rather than treated as a post-hoc add-on. For teams building clinical prediction systems—especially where clinicians need clear reasons behind alerts—this direction is more valuable than chasing metrics alone.

## At a glance

ICU medical data often takes the form of multivariate time series with irregular sampling and pervasive missingness, while clinical applications demand both accuracy and interpretability. GARLIC introduces a learnable exponential-decay encoder for imputation, captures inter-sensor dependencies via time-lagged summary graphs, and fuses global patterns with cross-dimensional sequential attention. All attention weights and graph edges are learned end-to-end, yielding observation-, signal-, and edge-level explanations. An alternating decoupled optimization scheme stabilizes training. GARLIC achieves state-of-the-art results on PhysioNet 2012/2019 and MIMIC-III, and demonstrates generalizability beyond the ICU domain.

## Key points

- Proposes a learnable exponential-decay encoder and time-lagged summary graphs to jointly learn imputation and inter-sensor dependencies end-to-end.
- All attention weights and graph edges serve as built-in observation-, signal-, and edge-level explanations without post-hoc tools.
- Achieves new state-of-the-art results on PhysioNet 2012/2019 and MIMIC-III, significantly improving AUROC and AUPRC.
- Feature-removal trials and real-time case studies validate attribution fidelity and actionable risk warning capability.

## Full text

Healthcare data, such as Intensive Care Unit (ICU) records, comprise heterogeneous multivariate time series sampled at irregular intervals with pervasive missingness. However, clinical applications demand predictive models that are both accurate and interpretable. We present our Graph Attention-based Relational Learning for Intensive Care (GARLIC) model, a novel neural network architecture that imputes missing data through a learnable exponential-decay encoder, captures inter-sensor dependencies via time-lagged summary graphs, and fuses global patterns with cross-dimensional sequential attention. All attention weights and graph edges are learned end-to-end to serve as built-in observation-, signal-, and edge-level explanations. To reconcile auxiliary reconstruction and primary classification objectives, we developed an alternating decoupled optimization scheme that stabilizes training. On three ICU benchmarks (PhysioNet 2012 & 2019, MIMIC-III), GARLIC sets the new state of the art in outcome prediction, significantly improving AUROC and AUPRC over best-performing baselines at comparable computational cost. Ablation studies confirm the contribution of each module, and feature-removal trials validate the fidelity of importance attribution through a monotonic performance drop (full > top 50% > random 50% > bottom 50%). Real-time case studies demonstrate actionable risk warnings with transparent explanations, marking a significant advance toward accurate, explainable deep learning for irregularly sampled ICU time series data. Moreover, we demonstrated \proposed{}'s superiority in data imputation and classification on various time-series datasets beyond the ICU domain, showing its generalizability and applicability to broader tasks.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_549f96cefcdad4bd3b0b2589