acceptodds
Under review as a conference paper at ICLR 2027

ARISE: Retrieval-Augmented Forecasting for Irregular Time Series

Abstract

Retrieval-augmented generation (RAG) supports time series forecasting by reusing historical windows and their subsequent observations. However, most existing methods are designed for regularly sampled data and face two challenges when extended to irregular multivariate time series (IMTS). Uneven observation density makes fixed windows prone to insufficient information or redundancy, while irregular sampling and asynchronous observations complicate the retrieval of relevant historical cases. To address these challenges, we propose ARISE, a retrieval-augmented forecasting framework for IMTS. First, ARISE introduces a dynamic window construction mechanism that learns a sample-specific left boundary for each historical window under multiple training objectives while keeping its right boundary fixed, yielding compact history–future pairs with predictive value. Second, a multi-evidence knowledge retrieval mechanism assesses historical relevance from three perspectives: observation patterns, temporal dynamics, and value information. By combining multi-view candidate retrieval with maximal marginal relevance (MMR) selection, ARISE balances relevance and diversity to retrieve complementary historical evidence. A retrieval-based memory fusion module then uses cross-attention to incorporate the retrieved histories and their observed continuations into the predictor's hidden representations. Finally, a forecasting decoder produces the final prediction. Experimental results on multiple real-world datasets show that ARISE outperforms existing state-of-the-art methods in forecasting accuracy.

Primary area:
learning on time series and dynamical systems
Keywords:
Irregular Time Series Forecasting, Retrieval-Augmented Forecasting, Multivariate Time Series

Reviews

Among:
Shape:
1234this paper, Top 4%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 3.33 over 3 reviews, above 96% of the others.
Reviewer_bHemOpenReview ↗
Rating
3(+1)
Confidence
4
Informativeness
2
Presentation
2
Soundness
1
Read review
Paper summary

This paper proposes ARISE, a retrieval-augmented forecasting framework for irregular multivariate time series (IMTS). ARISE operates in two stages on top of a frozen pretrained IMTS forecasting backbone (EB, DB). In the offline stage, Dynamic Window Construction evaluates C candidate left boundaries spaced proportionally across each sample's observed time span and learns a boundary selector and three view encoders (obs, dyn, val) under a composite objective (ℒcon = ℒforecast + λb ℒcompleteness + λl ℒlength) to populate a historical retrieval memory. In the online stage, Multi-Evidence Knowledge Retrieval retrieves top-K candidates per view, applies Maximal Marginal Relevance (MMR) to select R diverse histories, aligns each candidate's irregular future observations to the query's target timestamps via a Gaussian kernel (bandwidth σt), and applies multi-head cross-attention (with a learned null key K_∅) to produce a residual hidden-state correction Δh before the frozen backbone decoder DB.

The authors evaluate ARISE on 11 real-world IMTS datasets (PhysioNet, HumanActivity, USHCN, CESNET, Pamap2, EPA-Air, ClusterTrace, APTC, FNSPID, GDELT, Seabirds) across 3 horizons each against 9 non-retrieval IMTS baselines (TFMixer, ASTGI, APN, KAFNet, Hi-Patch, HyperIMTS, GNeuralFlow, GraFITi, Warpformer), reporting first-place average MAE on all 11 datasets and first-place average MSE on 9 of 11 datasets, along with component ablations and hyperparameter sensitivity plots on 4 datasets.

Critical strengths

Well-Motivated Problem Formulation and Target-Time Alignment for Irregular Retrieval: Extending retrieval-augmented forecasting from regular grids to irregular, asynchronous multivariate time series addresses a genuine practical gap in domains like clinical monitoring (PhysioNet) and environmental sensing (EPA-Air). In particular, using a Gaussian kernel over relative future offsets (τju vs. τqh) together with an explicit support mask Sqjhv and a learned null attention key K_∅ (Section 3.4.3) is a sensible way to handle asynchronous future timestamps and missing target support without interpolating the raw series onto a dense regular grid.

Critical weaknesses

Severe Technical Incompleteness and Unreproducible Core Formulations (Missing Equations and Unspecified Backbone): The paper omits the mathematical definitions of its core training losses and feature extractors, making the method impossible to reproduce from the manuscript: Identity of the Frozen Pretrained Backbone (EB, DB) Never Specified: Sections 3.4.2 and 3.4.4 state that ARISE encodes histories with a "frozen history encoder EB" and decodes hfused = hbase + Δh with a "pretrained backbone [that] remains fixed" (DB). Yet neither the main text nor the appendix ever identifies what architecture or pretrained model is used as (EB, DB)! Undefined Construction Losses in Eq. (5): Dynamic Window Construction—claimed as the first major contribution—is trained by ℒcon = ℒforecast + λb ℒcompleteness + λl ℒlength (Eq. 5). None of these three loss terms is ever defined mathematically anywhere in the main text or appendix; they are described only in three sentences of high-level prose (lines 254–260). Undefined View Descriptors and Encoders: The exact definitions of the boundary features ψim, the eligibility mask χselim, the three view feature maps ϕobs, ϕdyn, ϕval, the view encoder architecture Er, the content encoder uqjhv, and key hyperparameters (λb, λl, τcut, R, D, DA) are all missing. Internal Numerical Inconsistencies, Missing Retrieval Baselines, and Limited Novelty / Foundation-Model Alignment: Contradictions Between Tables 1–2 and Figures 3–4: In Table 2, w/o Attention Reader sets Δh = 0 (lines 427–428), which by Eq. (14) reduces ARISE to the standalone frozen pretrained backbone DB(EB(·)). However, its MSE on EPA-Air (1.57), CESNET (3.22), GDELT (2.97), and ClusterTrace (6.34) is dramatically worse than nearly all 2024–2026 baselines in Table 1 (e.g., TFMixer achieves 0.791, 2.327, 1.859, 4.199). Conversely, in the sensitivity plots (Figures 3–4), varying C, K, λMMR, σt across wide ranges barely changes MSE past the 3rd or 4th significant digit (e.g., CESNET stays within 2.1682–2.1692, a <0.05% variation, and λMMR = 0 is actually optimal on 2 of 4 datasets), whereas w/o Dynamic Window in Table 2 jumps from 2.16 to 2.71 (+25%). Furthermore, on GDELT, every sensitivity curve in Figures 3–4 lies between 1.910 and 1.923 (× 10-2), which contradicts the 1.819 (× 10-2) reported for ARISE in Tables 1, 2, and 3. No Retrieval-Augmented Baselines and No Variance / Latency Reporting: None of the 9 baselines in Table 1 is a retrieval-augmented method (e.g., RAFT, TS-RAG, Cross-RAG, SARAF, or RATL applied to the same IMTS backbone embeddings hbase). All results are single point estimates without standard deviations across random seeds, and there is no evaluation of memory or inference latency overhead despite running EB twice per query (hbase, hsel) and performing per-(horizon, variable) cross-attention. Novelty and Alignment with the LLM / Foundation Model Roadmap: Methodologically, ARISE combines heuristic summary descriptors with standard MMR (Carbonell & Goldstein, 1998), Gaussian kernel time alignment (mTAN, Shukla & Marlin, 2021), and history-key / future-value cross-attention (Cross-RAG, Lee et al., 2026). Rather than advancing unified continuous-time or event-tokenized foundation models that generalize zero-shot across irregular time series domains, ARISE requires multi-stage, dataset-specific training on handcrafted per-view statistics.

Questions for authors

Clarification / Core Assessment — What is the exact pretrained backbone (EB, DB) and why is w/o Attention Reader (Δh = 0) so weak in Table 2? Please specify the exact architecture and training recipe of the frozen backbone (EB, DB). Why does setting Δh = 0 (w/o Attention Reader in Table 2, which outputs the unaugmented frozen backbone prediction) yield MSEs of 1.57, 3.22, 2.97, 6.34—worse than almost every baseline in Table 1—and how can adding a retrieval residual Δh on top of such a weak frozen encoder beat state-of-the-art backbones like TFMixer and ASTGI? Clarification / Core Assessment — Exact mathematical formulations of ℒforecast, ℒcompleteness, ℒlength and ϕr: Please provide the exact mathematical equations for all three loss terms in Eq. (5), the eligibility mask χselim, and the three view feature maps ϕobs, ϕdyn, ϕval. Clarification / Core Assessment — Discrepancy on GDELT between Tables 1–3 (1.819) and Figures 3–4 (1.910–1.923), and comparison to RAG baselines: Why do all four sensitivity curves for GDELT in Figures 3 and 4 range between 1.9128 and 1.9203 (× 10-2), whereas Tables 1, 2, and 3 report 1.819 (× 10-2) for ARISE? In addition, how does ARISE compare against existing time-series RAG methods (such as RAFT, TS-RAG, or Cross-RAG) when applied directly in the latent space hbase of the same IMTS backbone?

Reviewer_QJjpOpenReview ↗
Rating
4(+1)
Confidence
3
Informativeness
2
Presentation
3
Soundness
3
Read review
Paper summary

The paper focuses on the irregular time-series forecasting task and approach via retrieval-based method. The authors point out two challenges in utilising the retrieval-based method to tackle this problem and proposed several techniques to circumvent. To build better database of irregular time-series that can be used for later forecasting, the paper proposed to use dynamic window and three-view features. To improve retrieval mechanism, the future values is proposed to be used. Experiments show that the approach can improve the performance on different datasets in irregular time-series forecasting.

Critical strengths
  • The experiments are extensively conducted in several benchmarks. Empirically, the authors showed that the ARISE approach can outperform other approaches in irregular time-series forecasting.
  • Ablation study also showed the importance of different components (Dynamic Window, Three-View Retrieval, Future Values Retrieval, and reading attention)
Critical weaknesses
  • The papers can explain clearer about how is the completeness loss computed (in the Equation 5, Line 253); and what are the features ψim\psi_{im}?
Questions for authors
  • Could the authors discuss about how the loss completeness is computed?
  • In line 271-272, could the authors discuss what is “value alignment” referring to?
Reviewer_Ej4KOpenReview ↗
Rating
3
Confidence
3
Informativeness
2
Presentation
2
Soundness
2
Read review
Paper summary

The paper proposes ARISE, a retrieval-augmented forecasting framework tailored to irregular multivariate time series (IMTS). ARISE tackles two core issues that impede direct application of RAG to IMTS: (i) constructing a prediction-oriented, non-redundant memory from irregularly sampled histories via a dynamic window construction mechanism that learns sample-specific left boundaries; and (ii) retrieving relevant historical cases under irregular sampling using multi-evidence (observation-pattern, dynamics, value) matching with MMR-based diversification and a reader that aligns and fuses paired futures through cross-attention. Extensive experiments across 11 real-world IMTS datasets show consistent gains over strong non-retrieval IMTS baselines, with informative ablations and sensitivity analyses.

Critical strengths
  1. Advances retrieval-augmented forecasting to a regime (irregular sampling with asynchronous variables) where prior RAG methods struggle, addressing a widely relevant and hard problem in healthcare, environmental sensing, and systems.

  2. Clearly articulates the two core challenges for RAG in IMTS and aligns the three proposed modules to those challenges.

  3. Evaluates on 11 diverse IMTS datasets covering multiple domains, with three horizons and comprehensive averages; includes ablations isolating the effect of each major component.

Critical weaknesses
  1. The overall pipeline is complex (selectors, three view encoders, content/score encoders, multiple normalizations); a concise complexity summary or module-wise schematic with parameter counts would help.

  2. Computational complexity and scalability are not thoroughly analyzed (e.g., memory size, retrieval latency, and fusion overhead for large M and many targets), which are crucial for RAG systems.

  3. No direct comparison to adapted RAG baselines (e.g., RAFT, TS-RAG, Cross-RAG) under irregular settings, even with reasonable adaptations (e.g., non-uniform time encodings or simple irregular-aware resampling), limits claims about RAG advances per se.

Questions for authors
  1. Can you report retrieval and fusion overhead (latency and memory) relative to the backbones, including memory size, and per-query computation cost? How does performance scale with larger memories?

  2. Could you include a comparison to adapted RAG baselines (e.g., RAFT/TS-RAG/Cross-RAG) under irregular sampling, even with simple irregular-aware encoders or time-offset features, to isolate where ARISE’s IMTS-specific design helps most?

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Loading the map…

Discussion (0)

Sign in to comment.