arxiv: 2605.13197 · v1 · submitted 2026-05-13 · 💻 cs.LG · cs.AI

Recognition: no theorem link

McCast: Memory-Guided Latent Drift Correction for Long-Horizon Precipitation Nowcasting

Penghui Wen , Yu Luo , Lintao Wang , Mengwei He , Patrick Filippi , Thomas Francis Bishop , Zhiyong Wang

Authors on Pith no claims yet

Pith reviewed 2026-05-14 20:32 UTC · model grok-4.3

classification 💻 cs.LG cs.AI

keywords precipitation nowcastinglatent drift correctionmemory bankautoregressive forecastinglong-horizon predictiondrift correctionmeteorological forecasting

0 comments

The pith

McCast corrects latent drift in autoregressive precipitation models using a memory bank to produce coherent long-horizon forecasts.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Autoregressive precipitation nowcasting models accumulate errors over successive steps, causing predictions to drift away from physically plausible trajectories. McCast counters this by maintaining a temporally organized memory bank that actively estimates and applies corrections to the evolving latent states during rollout. Rather than focusing only on improving accuracy at each individual step, the method calibrates the global sequence to stay consistent with observed meteorological evolution. This produces more reliable forecasts over extended time horizons on established benchmarks.

Core claim

By introducing a Drift-Corrective Memory Bank that extracts initial corrections from the current prediction and a reference state then refines them via temporally organized historical memory, McCast explicitly corrects divergent latent trajectories in autoregressive rollouts instead of relying solely on step-wise prediction improvements, thereby generating more temporally coherent and reliable long-horizon precipitation forecasts.

What carries the argument

The Drift-Corrective Memory Bank (DCBank) consisting of a Corrective Latent Extractor and Correction-Aware Memory Retrieval module that estimates and refines drift corrections from current latent predictions and historical states.

If this is right

Reduces cumulative error in multi-step rollouts by actively realigning latent evolution.
Yields state-of-the-art performance on SEVIR and MeteoNet benchmarks especially at longer horizons.
Shifts emphasis from local step accuracy to global temporal consistency in autoregressive forecasting.
Enables memory to serve as an active corrective mechanism rather than passive conditioning.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same memory-guided correction principle could apply to other autoregressive domains such as video frame prediction where drift similarly degrades long sequences.
Operational nowcasting pipelines might achieve longer useful forecast ranges without increasing model size if the memory bank generalizes across weather regimes.
Combining the latent correction with lightweight physical constraints could further reduce inconsistencies in regions with sparse observations.

Load-bearing premise

A temporally organized memory bank can reliably estimate and apply drift corrections to latent states without introducing new inconsistencies.

What would settle it

Ablation experiments on SEVIR or MeteoNet showing that long-horizon forecast skill scores remain unchanged or degrade when the drift-correction modules are removed or replaced with unordered memory retrieval.

Figures

Figures reproduced from arXiv: 2605.13197 by Lintao Wang, Mengwei He, Patrick Filippi, Penghui Wen, Thomas Francis Bishop, Yu Luo, Zhiyong Wang.

**Figure 1.** Figure 1: Visualization of DiffCast [3] with a vanilla memory bank, including the prediction and retrieved memories. Background color denotes intensity, red arrows indicate motion, and the dotted box highlights a zoomed-in region. The prediction shows drift and the retrieved memory lacks temporal consistency in intensity and motion. Existing methods mainly focus on mitigating the precipitation drift by improving per… view at source ↗

**Figure 2.** Figure 2: Overview of McCast. A backbone encoder maps the input sequence to a prior latent [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Figure 3: Details of the Drift-Corrective Memory Bank. An initial drift correction is estimated by [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

**Figure 4.** Figure 4: Performance comparison on SEVIR across different lead time and intensity thresholds. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗

**Figure 5.** Figure 5: Qualitative comparison on a SEVIR event. McCast produces forecasts with finer precipita [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗

**Figure 6.** Figure 6: Qualitative comparison on a SEVIR event with and without the correction-aware memory [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗

**Figure 7.** Figure 7: Qualitative comparison on a SEVIR event with and without active drift correction. Active [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗

**Figure 8.** Figure 8: Qualitative comparison on a SEVIR event with and without DCBank. DCBank better [PITH_FULL_IMAGE:figures/full_fig_p022_8.png] view at source ↗

**Figure 9.** Figure 9: Qualitative comparison of SimVP with and without DCBank on representative SEVIR [PITH_FULL_IMAGE:figures/full_fig_p022_9.png] view at source ↗

**Figure 10.** Figure 10: Prediction examples on the SEVIR dataset. [PITH_FULL_IMAGE:figures/full_fig_p024_10.png] view at source ↗

**Figure 11.** Figure 11: Prediction examples on the MeteoNet dataset. [PITH_FULL_IMAGE:figures/full_fig_p025_11.png] view at source ↗

read the original abstract

Existing precipitation nowcasting methods typically adopt an autoregressive formulation, where future states are predicted from previous outputs. However, such an approach accumulates errors over long rollouts, causing forecasts to drift away from physically plausible evolution trajectories. Although various studies have attempted to alleviate this problem by improving step-wise prediction accuracy, they largely neglect the global temporal evolution of meteorological systems and lack mechanisms to actively correct drift during rollouts. To address this issue, we propose McCast, a memory-guided latent drift correction method for precipitation nowcasting. Rather than treating memory as an unordered dictionary of latent states for passive conditioning, McCast leverages temporally organized memory to actively correct autoregressive latent evolution. Specifically, McCast introduces a Drift-Corrective Memory Bank (DCBank) that explicitly estimates the temporally consistent drift corrections to calibrate the divergent trajectory. DCBank performs drift correction in two stages: a Corrective Latent Extractor first predicts an initial correction from the current prediction and a reference latent state, and a Correction-Aware Memory Retrieval module then refines the initial correction using temporally organized historical memory. By explicitly correcting latent evolution, instead of improving step-wise prediction accuracy only, McCast produces more temporally coherent and reliable long-horizon forecasts. Experiments on two widely used benchmarks, SEVIR and MeteoNet, show that McCast achieves state-of-the-art performance, particularly in challenging long-horizon forecasting scenarios.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

McCast adds explicit latent drift correction via a time-structured memory bank and reports stronger long-horizon results on standard nowcasting benchmarks.

read the letter

McCast's main contribution is a memory-guided way to correct drift in latent space for better long-horizon nowcasts, and the paper shows it works on the usual benchmarks. The new part is the Drift-Corrective Memory Bank that does correction in two stages: the Corrective Latent Extractor pulls an initial fix from the current state and a reference, then the Correction-Aware Memory Retrieval uses temporally sorted historical latents to refine it. This is different from just conditioning on memory or improving single-step accuracy. It directly targets the accumulation of errors over many steps by adjusting the trajectory. The paper does a good job laying out the problem with autoregressive methods and explaining how the memory is organized by time to make corrections. The evaluation on SEVIR and MeteoNet with SOTA claims for long horizons is standard and relevant for the field. One soft spot is the reliance on the memory bank providing accurate drift estimates without new issues; the description is consistent, but without seeing detailed ablations or how much the correction actually reduces physical implausibility, it's hard to gauge the full impact. The assumption that latent corrections stay plausible seems to hold in their setup, but real-world tuning might be needed. This paper is for researchers in ML applied to meteorology who want to improve long-term forecast coherence. A reader looking for concrete architectural ideas to handle rollout drift would find it useful. I would send it to peer review because the method is well-motivated, the architecture is described clearly, and the benchmarks are appropriate for judging the claims.

Referee Report

2 major / 2 minor

Summary. The paper proposes McCast, a memory-guided latent drift correction framework for precipitation nowcasting. It augments standard autoregressive models with a Drift-Corrective Memory Bank (DCBank) comprising a Corrective Latent Extractor that predicts an initial correction from the current prediction and a reference latent state, followed by a Correction-Aware Memory Retrieval module that refines the correction using temporally organized historical memory. The central claim is that explicitly correcting latent evolution trajectories during rollout yields more temporally coherent long-horizon forecasts than methods focused solely on step-wise accuracy. Experiments on SEVIR and MeteoNet are reported to achieve state-of-the-art performance, particularly at extended horizons.

Significance. If the empirical results and ablations hold, the work demonstrates that structured, time-organized memory can actively calibrate divergent autoregressive trajectories in latent space, offering a concrete mechanism beyond incremental per-step improvements. This has potential value for operational nowcasting systems where forecast coherence over 30-60 minutes directly impacts decision-making. The explicit two-stage correction architecture is a clear contribution relative to passive memory conditioning approaches.

major comments (2)

[§4] §4 (Experiments): the central claim that drift correction produces more coherent forecasts than step-wise accuracy improvements alone requires explicit ablation isolating the DCBank contribution versus a strong autoregressive baseline with equivalent per-step accuracy; without this, the distinction remains unproven.
[§3.2] §3.2 (DCBank description): the selection criterion for the reference latent state in the Corrective Latent Extractor is not fully specified; if it depends on learned parameters rather than fixed temporal indexing, the method may require domain-specific tuning that undermines the 'active correction' advantage.

minor comments (2)

[Figure 2] Figure 2 (architecture diagram): the flow from Correction-Aware Memory Retrieval back to the latent state update should include explicit notation for the correction vector to improve readability.
[§4.1] §4.1 (dataset details): report the exact long-horizon intervals evaluated (e.g., 30 min, 60 min) and any preprocessing steps for SEVIR/MeteoNet to ensure reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the positive evaluation and constructive feedback. We address the two major comments below and will incorporate revisions to strengthen the manuscript.

read point-by-point responses

Referee: [§4] §4 (Experiments): the central claim that drift correction produces more coherent forecasts than step-wise accuracy improvements alone requires explicit ablation isolating the DCBank contribution versus a strong autoregressive baseline with equivalent per-step accuracy; without this, the distinction remains unproven.

Authors: We agree that an explicit ablation isolating the DCBank's contribution from per-step accuracy gains is necessary to substantiate the central claim. In the revised manuscript, we will add a new ablation experiment comparing McCast to a strengthened autoregressive baseline (e.g., a larger-capacity model or one trained with additional iterations to match per-step accuracy metrics on short horizons). This will demonstrate that the long-horizon coherence improvements arise specifically from the active drift correction rather than incremental accuracy alone. revision: yes
Referee: [§3.2] §3.2 (DCBank description): the selection criterion for the reference latent state in the Corrective Latent Extractor is not fully specified; if it depends on learned parameters rather than fixed temporal indexing, the method may require domain-specific tuning that undermines the 'active correction' advantage.

Authors: We appreciate this clarification request. The reference latent state is selected via fixed temporal indexing from the memory bank (i.e., the historical latent state at the matching relative time step in the organized sequence). This selection is deterministic and independent of learned parameters. We will explicitly document this criterion in the revised §3.2 to confirm that no domain-specific tuning is required for the active correction mechanism. revision: yes

Circularity Check

0 steps flagged

Minor self-citation not load-bearing; derivation self-contained

full rationale

The paper's derivation introduces McCast via the Drift-Corrective Memory Bank (DCBank) with its two explicit stages (Corrective Latent Extractor predicting initial correction from current prediction and reference latent, followed by Correction-Aware Memory Retrieval refining via temporally organized historical memory). These are architectural additions to standard autoregressive nowcasting frameworks rather than quantities defined in terms of each other or fitted parameters renamed as predictions. No equations reduce by construction to inputs, no uniqueness theorems are imported from self-citations, and no ansatzes are smuggled via prior work. Evaluation on independent external benchmarks (SEVIR, MeteoNet) provides non-circular validation of long-horizon coherence gains. Any self-citations are peripheral and not load-bearing for the central claim.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 3 invented entities

The central claim rests on the effectiveness of newly postulated components (DCBank and its sub-modules) whose performance is asserted without independent evidence or parameter details in the abstract; no free parameters are explicitly listed.

axioms (2)

domain assumption Autoregressive formulations accumulate errors over long rollouts in precipitation nowcasting
Explicitly stated as the core limitation of existing methods in the abstract.
ad hoc to paper Temporally organized memory can actively estimate consistent drift corrections in latent space
Central premise of the proposed DCBank mechanism.

invented entities (3)

Drift-Corrective Memory Bank (DCBank) no independent evidence
purpose: Explicitly estimates temporally consistent drift corrections to calibrate divergent trajectories
New component introduced to perform two-stage correction
Corrective Latent Extractor no independent evidence
purpose: Predicts an initial correction from the current prediction and a reference latent state
First stage of the drift correction process
Correction-Aware Memory Retrieval module no independent evidence
purpose: Refines the initial correction using temporally organized historical memory
Second stage of the drift correction process

pith-pipeline@v0.9.0 · 5569 in / 1507 out tokens · 42017 ms · 2026-05-14T20:32:09.629144+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

37 extracted references · 37 canonical work pages · 1 internal anchor

[1]

Skilful pre- cipitation nowcasting using deep generative models of radar.Nature, 597(7878):672–677, 2021

Suman Ravuri, Karel Lenc, Matthew Willson, Dmitry Kangin, Remi Lam, Piotr Mirowski, Megan Fitzsimons, Maria Athanassiadou, Sheleem Kashem, Sam Madge, et al. Skilful pre- cipitation nowcasting using deep generative models of radar.Nature, 597(7878):672–677, 2021

work page 2021
[2]

Skilful nowcasting of extreme precipitation with nowcastnet.Nature, 619 (7970):526–532, 2023

Yuchen Zhang, Mingsheng Long, Kaiyuan Chen, Lanxiang Xing, Ronghua Jin, Michael I Jordan, and Jianmin Wang. Skilful nowcasting of extreme precipitation with nowcastnet.Nature, 619 (7970):526–532, 2023

work page 2023
[3]

Diffcast: A unified framework via residual diffusion for precipitation nowcasting

Demin Yu, Xutao Li, Yunming Ye, Baoquan Zhang, Chuyao Luo, Kuai Dai, Rui Wang, and Xunlai Chen. Diffcast: A unified framework via residual diffusion for precipitation nowcasting. InConference on Computer Vision and Pattern Recognition, pages 27758–27767, 2024

work page 2024
[4]

AlphaPre: Amplitude-phase disentanglement model for precipitation nowcasting

Kenghong Lin, Baoquan Zhang, Demin Yu, Wenzhi Feng, Shidong Chen, Feifan Gao, Xutao Li, and Yunming Ye. AlphaPre: Amplitude-phase disentanglement model for precipitation nowcasting. InComputer Vision and Pattern Recognition Conference, pages 17841–17850, June 2025

work page 2025
[5]

CasCast: Skillful high-resolution precipitation nowcasting via cascaded modelling

Junchao Gong, Lei Bai, Peng Ye, Wanghan Xu, Na Liu, Jianhua Dai, Xiaokang Yang, and Wanli Ouyang. CasCast: Skillful high-resolution precipitation nowcasting via cascaded modelling. International Conference on Machine Learning, 2024

work page 2024
[6]

Pimmnet: Introducing multi-modal precipitation nowcasting via a physics-informed perspective

Demin Yu, Wenchuan Du, Kenghong Lin, Xutao Li, Yunming Ye, Chuyao Luo, and Xunlai Chen. Pimmnet: Introducing multi-modal precipitation nowcasting via a physics-informed perspective. InACM International Conference on Multimedia, pages 11522–11531, 2025

work page 2025
[7]

Perceptually constrained precipitation nowcasting model

Wenzhi Feng, Xutao Li, Zhe Wu, Kenghong Lin, Demin Yu, Yunming Ye, and Yaowei Wang. Perceptually constrained precipitation nowcasting model. InInternational Conference on Machine Learning, 2025

work page 2025
[8]

Duocast: Duo-probabilistic diffusion for precipitation nowcasting

Penghui Wen, Mengwei He, Patrick Filippi, Na Zhao, Feng Zhang, Thomas Francis Bishop, Zhiyong Wang, and Kun Hu. Duocast: Duo-probabilistic diffusion for precipitation nowcasting. InAAAI Conference on Artificial Intelligence, volume 40, pages 39442–39450, 2026

work page 2026
[9]

Memorybank: Enhancing large language models with long-term memory

Wanjun Zhong, Lianghong Guo, Qiqi Gao, He Ye, and Yanlin Wang. Memorybank: Enhancing large language models with long-term memory. InAAAI Conference on Artificial Intelligence, volume 38, pages 19724–19731, 2024

work page 2024
[10]

Corgi: Cached memory guided video generation

Xindi Wu, Uriel Singer, Zhaojiang Lin, Andrea Madotto, Xide Xia, Yifan Xu, Paul Crook, Xin Luna Dong, and Seungwhan Moon. Corgi: Cached memory guided video generation. In Winter Conference on Applications of Computer Vision, pages 4585–4594. IEEE, 2025

work page 2025
[11]

Context as memory: Scene-consistent interactive long video generation with memory retrieval

Jiwen Yu, Jianhong Bai, Yiran Qin, Quande Liu, Xintao Wang, Pengfei Wan, Di Zhang, and Xihui Liu. Context as memory: Scene-consistent interactive long video generation with memory retrieval. InSIGGRAPH Asia, pages 1–11, 2025

work page 2025
[12]

SEVIR: A storm event imagery dataset for deep learning applications in radar and satellite meteorology.Advances in Neural Information Processing Systems, 33:22009–22019, 2020

Mark Veillette, Siddharth Samsi, and Chris Mattioli. SEVIR: A storm event imagery dataset for deep learning applications in radar and satellite meteorology.Advances in Neural Information Processing Systems, 33:22009–22019, 2020. 10

work page 2020
[13]

Meteonet: An open reference weather dataset for ai by météo-france

Gwennaëlle Larvor and Lea Berthomier. Meteonet: An open reference weather dataset for ai by météo-france. InAmerican Meteorological Society Meeting Abstracts, volume 101, pages 1–ii, 2021

work page 2021
[14]

Convolutional LSTM network: A machine learning approach for precipitation nowcasting

Xingjian Shi, Zhourong Chen, Hao Wang, Dit-Yan Yeung, Wai-Kin Wong, and Wang-chun Woo. Convolutional LSTM network: A machine learning approach for precipitation nowcasting. Advances in Neural Information Processing Systems, 28, 2015

work page 2015
[15]

Prediff: Precipitation nowcasting with latent diffusion models

Zhihan Gao, Xingjian Shi, Boran Han, Hao Wang, Xiaoyong Jin, Danielle Maddix, Yi Zhu, Mu Li, and Yuyang Bernie Wang. Prediff: Precipitation nowcasting with latent diffusion models. Advances in Neural Information Processing Systems, 36, 2024

work page 2024
[16]

A foundation model for the earth system.Nature, 641(8065):1180–1187, 2025

Cristian Bodnar, Wessel P Bruinsma, Ana Lucic, Megan Stanley, Anna Allen, Johannes Brand- stetter, Patrick Garvan, Maik Riechert, Jonathan A Weyn, Haiyu Dong, et al. A foundation model for the earth system.Nature, 641(8065):1180–1187, 2025

work page 2025
[17]

Worldexplorer: Towards generating fully navigable 3d scenes

Manuel-Andreas Schneider, Lukas Höllein, and Matthias Nießner. Worldexplorer: Towards generating fully navigable 3d scenes. InSIGGRAPH Asia, pages 1–11, 2025

work page 2025
[18]

Memory forcing: Spatio-temporal memory for consistent scene generation on minecraft.arXiv preprint arXiv:2510.03198, 2025

Junchao Huang, Xinting Hu, Boyao Han, Shaoshuai Shi, Zhuotao Tian, Tianyu He, and Li Jiang. Memory forcing: Spatio-temporal memory for consistent scene generation on minecraft.arXiv preprint arXiv:2510.03198, 2025

work page arXiv 2025
[19]

Vggt: Visual geometry grounded transformer

Jianyuan Wang, Minghao Chen, Nikita Karaev, Andrea Vedaldi, Christian Rupprecht, and David Novotny. Vggt: Visual geometry grounded transformer. InComputer Vision and Pattern Recognition Conference, pages 5294–5306, 2025

work page 2025
[20]

LMcast: A pretrained language model guided long-term memory transformer for precipitation nowcasting.Neural Networks, page 108168, 2025

Feifan Gao, Chuyao Luo, Guangbo Deng, Xutao Li, Baoquan Zhang, Demin Yu, and Yun- ming Ye. LMcast: A pretrained language model guided long-term memory transformer for precipitation nowcasting.Neural Networks, page 108168, 2025

work page 2025
[21]

Lora: Low-rank adaptation of large language models.International Conference on Learning Representations, 1(2):3, 2022

Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Liang Wang, Weizhu Chen, et al. Lora: Low-rank adaptation of large language models.International Conference on Learning Representations, 1(2):3, 2022

work page 2022
[22]

Deep learning for precipitation nowcasting: A benchmark and a new model

Xingjian Shi, Zhihan Gao, Leonard Lausen, Hao Wang, Dit-Yan Yeung, Wai-kin Wong, and Wang-chun Woo. Deep learning for precipitation nowcasting: A benchmark and a new model. Advances in neural information processing systems, 30, 2017

work page 2017
[23]

Mau: A motion-aware unit for video prediction and beyond.Advances in Neural Information Processing Systems, 34:26950–26962, 2021

Zheng Chang, Xinfeng Zhang, Shanshe Wang, Siwei Ma, Yan Ye, Xiang Xinguang, and Wen Gao. Mau: A motion-aware unit for video prediction and beyond.Advances in Neural Information Processing Systems, 34:26950–26962, 2021

work page 2021
[24]

Simvp: Simpler yet better video prediction

Zhangyang Gao, Cheng Tan, Lirong Wu, and Stan Z Li. Simvp: Simpler yet better video prediction. InConference on Computer Vision and Pattern Recognition, pages 3170–3180, 2022

work page 2022
[25]

FourCastNet: A Global Data-driven High-resolution Weather Model using Adaptive Fourier Neural Operators

Jaideep Pathak, Shashank Subramanian, Peter Harrington, Sanjeev Raja, Ashesh Chattopadhyay, Morteza Mardani, Thorsten Kurth, David Hall, Zongyi Li, Kamyar Azizzadenesheli, et al. Fourcastnet: A global data-driven high-resolution weather model using adaptive fourier neural operators.arXiv preprint arXiv:2202.11214, 2022

work page internal anchor Pith review Pith/arXiv arXiv 2022
[26]

Earthformer: Exploring space-time transformers for earth system forecasting.Advances in Neural Information Processing Systems, 35:25390–25403, 2022

Zhihan Gao, Xingjian Shi, Hao Wang, Yi Zhu, Yuyang Bernie Wang, Mu Li, and Dit-Yan Yeung. Earthformer: Exploring space-time transformers for earth system forecasting.Advances in Neural Information Processing Systems, 35:25390–25403, 2022

work page 2022
[27]

Disentangling physical dynamics from unknown factors for unsupervised video prediction

Vincent Le Guen and Nicolas Thome. Disentangling physical dynamics from unknown factors for unsupervised video prediction. InConference on Computer Vision and Pattern Recognition, pages 11474–11484, 2020

work page 2020
[28]

Earthfarsser: Versatile spatio-temporal dynamical systems modeling in one model

Hao Wu, Yuxuan Liang, Wei Xiong, Zhengyang Zhou, Wei Huang, Shilong Wang, and Kun Wang. Earthfarsser: Versatile spatio-temporal dynamical systems modeling in one model. In AAAI Conference on Artificial Intelligence, volume 38, pages 15906–15914, 2024. 11

work page 2024
[29]

Fourier amplitude and correlation loss: Beyond using l2 loss for skillful precipitation nowcasting

Chiu-Wai Yan, Shi Quan Foo, Van Hoan Trinh, Dit-Yan Yeung, Ka-Hing Wong, and Wai-Kin Wong. Fourier amplitude and correlation loss: Beyond using l2 loss for skillful precipitation nowcasting. InAdvances in Neural Information Processing Systems, 2024

work page 2024
[30]

Tracerouter: Robust safety for large foundation models via path-level intervention.arXiv preprint arXiv:2601.21900, 2026

Chuancheng Shi, Shangze Li, Wenjun Lu, Wenhua Wu, Cong Wang, Zifeng Cheng, Fei Shen, and Tat-Seng Chua. Tracerouter: Robust safety for large foundation models via path-level intervention.arXiv preprint arXiv:2601.21900, 2026

work page arXiv 2026
[31]

Dna: Uncovering universal latent forgery knowledge.arXiv preprint arXiv:2601.22515, 2026

Jingtong Dou, Chuancheng Shi, Yemin Wang, Shiming Guo, Anqi Yi, Wenhua Wu, Li Zhang, Fei Shen, and Tat-Seng Chua. Dna: Uncovering universal latent forgery knowledge.arXiv preprint arXiv:2601.22515, 2026

work page arXiv 2026
[32]

Swin transformer v2: Scaling up capacity and resolution

Ze Liu, Han Hu, Yutong Lin, Zhuliang Yao, Zhenda Xie, Yixuan Wei, Jia Ning, Yue Cao, Zheng Zhang, Li Dong, et al. Swin transformer v2: Scaling up capacity and resolution. InComputer Vision and Pattern Recognition Conference, pages 12009–12019, 2022

work page 2022
[33]

Era5 hourly data on single levels from 1940 to present.Copernicus Climate Change Service (C3S) Climate Data Store (CDS), 10(1.24381):24381, 2023

Hans Hersbach, Bill Bell, Paul Berrisford, Gionata Biavati, András Horányi, Joaquín Muñoz Sabater, Julien Nicolas, Carole Peubey, Raluca Radu, Iryna Rozum, et al. Era5 hourly data on single levels from 1940 to present.Copernicus Climate Change Service (C3S) Climate Data Store (CDS), 10(1.24381):24381, 2023

work page 1940
[34]

The era5 global reanalysis.Quarterly Journal of the Royal Meteorological Society, 146(730):1999–2049, 2020

Hans Hersbach, Bill Bell, Paul Berrisford, Shoji Hirahara, András Horányi, Joaquín Muñoz- Sabater, Julien Nicolas, Carole Peubey, Raluca Radu, Dinand Schepers, et al. The era5 global reanalysis.Quarterly Journal of the Royal Meteorological Society, 146(730):1999–2049, 2020

work page 1999
[35]

Weatherbench 2: A benchmark for the next generation of data-driven global weather models.Journal of Advances in Modeling Earth Systems, 16(6):e2023MS004019, 2024

Stephan Rasp, Stephan Hoyer, Alexander Merose, Ian Langmore, Peter Battaglia, Tyler Russell, Alvaro Sanchez-Gonzalez, Vivian Yang, Rob Carver, Shreya Agrawal, et al. Weatherbench 2: A benchmark for the next generation of data-driven global weather models.Journal of Advances in Modeling Earth Systems, 16(6):e2023MS004019, 2024

work page 2024
[36]

Limitations

Zheng Chang, Xinfeng Zhang, Shanshe Wang, Siwei Ma, and Wen Gao. Strpm: A spatiotempo- ral residual predictive model for high-resolution video prediction. InConference on Computer Vision and Pattern Recognition, pages 13946–13955, 2022. 12 Appendix Contents A Broader Impact 14 B Limitations 14 C Licenses for Existing Assets 14 D Justifications 14 D.1 Just...

work page 2022
[37]

Guidelines: • The answer [N/A] means that the paper does not involve crowdsourcing nor research with human subjects

Institutional review board (IRB) approvals or equivalent for research with human subjects Question: Does the paper describe potential risks incurred by study participants, whether such risks were disclosed to the subjects, and whether Institutional Review Board (IRB) approvals (or an equivalent approval/review based on the requirements of your country or ...

work page