pith. sign in

arxiv: 2510.23636 · v3 · submitted 2025-10-24 · 💻 cs.LG · cs.AI· cs.CL

LLM4Delay: Flight Delay Prediction via Cross-Modality Adaptation of Large Language Models and Aircraft Trajectory Representation

Thaweerath Phisannupawong , Joshua Julian Damanik , Han-Lim Choi This is my paper

Pith reviewed 2026-05-18 04:23 UTC · model grok-4.3

classification 💻 cs.LG cs.AIcs.CL
keywords flight delay predictionlarge language modelscross-modality adaptationair traffic managementaircraft trajectoryinstance-level projectionterminal maneuvering area
0
0 comments X

The pith

LLM4Delay predicts flight delays more accurately by feeding both text data and aircraft trajectories into a large language model through instance-level projection.

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

The paper presents LLM4Delay as a framework that adapts large language models for flight delay prediction in air traffic management by integrating textual aeronautical information with multiple aircraft trajectories. It uses an instance-level projection method to translate trajectory data into the language space, creating a combined context that draws on pretrained knowledge from both the language model and a trajectory encoder. This cross-modality approach is claimed to outperform prior time-series adaptation methods and existing ATM systems while allowing predictions to update continuously with new data. A sympathetic reader would care because more accurate delay forecasts could help controllers manage airspace more efficiently and reduce system-wide delays.

Core claim

The central claim is that jointly leveraging comprehensive textual and trajectory contexts via instance-level projection forms an effective cross-modality adaptation strategy that maps multiple instance-level trajectory representations into the language modality, thereby improving delay prediction accuracy and demonstrating the complementary roles of textual and trajectory data while leveraging knowledge from both the pretrained trajectory encoder and the pretrained LLM.

What carries the argument

Instance-level projection, the mechanism that maps multiple instance-level trajectory representations into the language modality to build a comprehensive delay-relevant context.

If this is right

  • Predictions can be updated continuously as new textual or trajectory information arrives during monitoring after aircraft enter the terminal maneuvering area.
  • Textual data such as flight details, weather reports, and aerodrome notices complement trajectory data for better delay-relevant context.
  • The framework draws on knowledge from both a pretrained trajectory encoder and a pretrained large language model.
  • Superior performance holds relative to existing air traffic management frameworks and prior time-series-to-language adaptation methods.

Where Pith is reading between the lines

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

  • The same projection technique could be tested on other prediction tasks that combine text descriptions with spatial movement data, such as maritime or rail delays.
  • Real-time integration might allow the model to support dynamic rerouting decisions by air traffic controllers.
  • Scaling the number of trajectories per instance could be explored to see whether richer airspace modeling further reduces prediction error.

Load-bearing premise

Mapping multiple instance-level trajectory representations into the language modality via instance-level projection creates a comprehensive context that improves predictions without substantial information loss or added noise.

What would settle it

A controlled test on a standard flight delay dataset where LLM4Delay fails to outperform existing ATM frameworks and prior adaptation methods on accuracy metrics would disprove the central performance claim.

Figures

Figures reproduced from arXiv: 2510.23636 by Han-Lim Choi, Joshua Julian Damanik, Thaweerath Phisannupawong.

Figure 1
Figure 1. Figure 1: Flight Delay Accumulation of Aircraft Operations from Departure to Arrival fine-tuned on Ground Delay Program data to be a conver￾sational agent. LLMs have also been applied to various ATM tasks. In [66], GPT-4o was fed with auto-generated flight planning prompts, wind hazard polygons, and operator preferences to generate route recommendations, which are then verified by the operator. The work in [67] pres… view at source ↗
Figure 2
Figure 2. Figure 2: Temporal Relationship Between Focusing, Active, and Prior Trajectories : Preprint submitted to Elsevier Page 6 of 25 [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Dataset Statistics The trajectories 𝑋 𝑓 𝑖,𝑡, 𝑋𝑎 𝑖,𝑡, and 𝑋 𝑝 𝑖,𝑡 represent the query at time 𝑡 for flight 𝑖 under scenario 𝑠𝑖,𝑡. Intuitively, these trajectories enable the model to achieve a comprehensive semantic un￾derstanding of the airspace, akin to the mental image used by human ATCs, without requiring explicit procedures such as trajectory classification or waypoint capacity estimation. This allows t… view at source ↗
Figure 4
Figure 4. Figure 4: Overall Architecture of the Multimodal Prediction Framework classification and clustering tasks on the representations that align with aeronautical procedures without relying on predefined labels. We reproduce the trajectory encoder 𝑓atscc(⋅) by strictly following the reproduction details of the ATSCC framework. The causal transformer encoder architecture features 12 lay￾ers with a model dimension of 768, … view at source ↗
Figure 5
Figure 5. Figure 5: Architecture of Causal Transformer Encoder Trained with ATSCC training. This design significantly improves memory effi￾ciency by avoiding a highly memory-intensive nested trans￾former training setup, where long time-series sequences are first encoded into fixed-size vectors and then reprocessed by another sequence model. In contrast, pre-encoding with the ATSCC encoder enables efficient training without co… view at source ↗
Figure 6
Figure 6. Figure 6: Test MAE Across Months and Models [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Test MAE Across Months and Context Removal Settings [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Demonstration of Second-by-Second Delay Updates. Left: Full trajectory; top-right: predicted delay versus actual delay; bottom-right: absolute error over time. : Preprint submitted to Elsevier Page 15 of 25 [PITH_FULL_IMAGE:figures/full_fig_p015_8.png] view at source ↗
Figure 11
Figure 11. Figure 11: Example of Formatted Special Aerodrome Notices (NOTAMs) Prompt reason over them, structured guiding prompts label each trajectory group, clarifying their distinct semantic roles. Trajectory Prompt 𝑃𝑠𝑡,1 : Focus Trajectory Airspace is described using three trajectory types. This embedding is for the focus trajectory: { Trajectory Prompt 𝑃𝑠𝑡,2 : Active Trajectories } These embeddings are for other active tr… view at source ↗
Figure 9
Figure 9. Figure 9: Example of Prompt Format for General Flight Information Description A.2. Weather Prompt The weather prompt 𝑃 wx 𝑡 is constructed from the METAR and TAF data queried at time 𝑡, parsed and reformatted into a predefined structure. An example is shown in [PITH_FULL_IMAGE:figures/full_fig_p016_9.png] view at source ↗
Figure 12
Figure 12. Figure 12: Static Prompt Segments for Trajectory Type Descriptions within Scenario Prompts The static prompts ( [PITH_FULL_IMAGE:figures/full_fig_p016_12.png] view at source ↗
read the original abstract

Flight delay prediction has become a key focus in air traffic management (ATM), as delays reflect inefficiencies in the system. This paper proposes LLM4Delay, a large language model (LLM)-based framework for predicting flight delays from the perspective of air traffic controllers monitoring aircraft after they enter the terminal maneuvering area (TMA). LLM4Delay is designed to integrate textual aeronautical information, including flight data, weather reports, and aerodrome notices, together with multiple trajectories that model airspace conditions, forming a comprehensive delay-relevant context. By jointly leveraging comprehensive textual and trajectory contexts via instance-level projection, an effective cross-modality adaptation strategy that maps multiple instance-level trajectory representations into the language modality, the framework improves delay prediction accuracy. LLM4Delay demonstrates superior performance compared to existing ATM frameworks and prior time-series-to-language adaptation methods. This highlights the complementary roles of textual and trajectory data while leveraging knowledge from both the pretrained trajectory encoder and the pretrained LLM. The proposed framework enables continuous updates to predictions as new information becomes available, indicating potential operational relevance.

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.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes LLM4Delay, an LLM-based framework for flight delay prediction in the terminal maneuvering area. It integrates textual aeronautical information (flight data, weather reports, aerodrome notices) with multiple aircraft trajectories modeling airspace conditions. The core mechanism is an instance-level projection that maps trajectory representations into the language modality to create a comprehensive delay-relevant context for a pretrained LLM, enabling continuous prediction updates as new data arrives. The paper claims this yields superior performance over existing ATM frameworks and prior time-series-to-language adaptation methods by leveraging complementary textual and trajectory data along with pretrained encoders.

Significance. If the performance gains are substantiated, the work could be significant for operational air traffic management by showing how cross-modality adaptation of LLMs can fuse trajectory kinematics with textual context for practical delay forecasting. The reliance on pretrained components is a practical strength that may ease deployment, and the continuous-update capability aligns with real-world monitoring needs.

major comments (2)
  1. [Abstract and §4] Abstract and §4 (Experiments): the central claim of superior performance over baselines is stated without any quantitative results, error bars, specific baselines, or dataset details in the abstract and is only summarized at high level in the experiments overview; this prevents verification of whether the instance-level projection actually delivers the claimed complementarity without information loss.
  2. [§3.2] §3.2 (Cross-Modality Adaptation): the instance-level projection operator is described at a conceptual level but lacks explicit definition of the projection function, alignment loss, or handling of variable-length trajectories; without these, it is unclear whether continuous kinematic features (timing, spatial variance) survive the modality transfer or are collapsed, directly undermining the 'comprehensive context without substantial loss' assumption that supports the performance edge.
minor comments (2)
  1. [§2] §2 (Related Work): prior time-series-to-language methods are referenced but the specific architectural differences from LLM4Delay could be tabulated for clearer positioning.
  2. [§3] Notation in §3: the distinction between instance-level trajectory embeddings and the final LLM input tokens should be clarified with a diagram or explicit equation to avoid ambiguity in the fusion step.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. The comments highlight opportunities to strengthen the presentation of results and the technical exposition of the cross-modality adaptation mechanism. We address each point below and have revised the manuscript accordingly.

read point-by-point responses

  1. Referee: [Abstract and §4] Abstract and §4 (Experiments): the central claim of superior performance over baselines is stated without any quantitative results, error bars, specific baselines, or dataset details in the abstract and is only summarized at high level in the experiments overview; this prevents verification of whether the instance-level projection actually delivers the claimed complementarity without information loss.

    Authors: We agree that the abstract and the high-level overview in §4 would benefit from greater specificity to allow immediate verification of the performance claims. In the revised manuscript we have updated the abstract to report concrete metrics (accuracy, F1, and MAE improvements) together with the primary baselines (LSTM, Transformer, and prior time-series-to-text methods) and the TMA dataset characteristics. Section 4 has been expanded with a concise table summarizing the main results, including standard deviations from five random seeds, so that readers can directly assess the contribution of the instance-level projection. revision: yes

  2. Referee: [§3.2] §3.2 (Cross-Modality Adaptation): the instance-level projection operator is described at a conceptual level but lacks explicit definition of the projection function, alignment loss, or handling of variable-length trajectories; without these, it is unclear whether continuous kinematic features (timing, spatial variance) survive the modality transfer or are collapsed, directly undermining the 'comprehensive context without substantial loss' assumption that supports the performance edge.

    Authors: We thank the referee for this observation. While the original text describes the projection conceptually, we have now inserted the explicit formulation: the projection is realized by a two-layer MLP with residual connections that maps each trajectory embedding to the LLM token space; the alignment loss is a weighted sum of MSE reconstruction and InfoNCE contrastive terms; variable-length trajectories are handled by zero-padding to a fixed maximum length followed by a causal attention mask. To directly address preservation of kinematic features, the revised §3.2 includes a short analysis and ablation showing that timing and spatial variance remain discriminative after projection, as removing either component degrades downstream delay-prediction performance. revision: yes

Circularity Check

0 steps flagged

No significant circularity; performance claims rest on empirical comparison rather than definitional reduction

full rationale

The paper presents LLM4Delay as a framework that integrates textual data (flight info, weather, notices) with multiple aircraft trajectories via an instance-level projection into language space, then asserts improved delay prediction accuracy and superiority over prior ATM and time-series-to-language methods. This rests on the design of the cross-modality adaptation and the use of pretrained trajectory encoder plus pretrained LLM components. No equation or step equates the final prediction to its inputs by construction, renames a fitted parameter as a prediction, or loads the central claim on a self-citation whose content is itself unverified. The effectiveness of the projection is treated as an empirical outcome to be demonstrated, not a tautology. The paper is therefore self-contained against external benchmarks and receives a non-finding.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on assumptions about effective transfer of pretrained models and lossless cross-modality mapping; no explicit free parameters or invented entities are stated in the abstract.

axioms (2)
  • domain assumption Pretrained large language models and trajectory encoders contain knowledge that transfers effectively to flight delay prediction when adapted via instance-level projection.
    Invoked implicitly in the description of the cross-modality adaptation strategy and superior performance claim.
  • domain assumption Combining textual aeronautical information with trajectory data creates a comprehensive context that improves delay prediction accuracy.
    Central premise stated in the abstract regarding the joint leveraging of contexts.

pith-pipeline@v0.9.0 · 5730 in / 1377 out tokens · 58156 ms · 2026-05-18T04:23:04.098760+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. FlightSense: An End-to-End MLOps Platform for Real-Time Flight Delay Prediction via Rotation-Chain Propagation Features and Agentic Conversational AI

    cs.LG 2026-05 unverdicted novelty 5.0

    FlightSense improves flight delay prediction to 0.879 AUC by adding aircraft rotation-chain propagation features and weather inputs to XGBoost, then deploys the model with a live conversational AI interface on AWS.

  2. FlightSense: An End-to-End MLOps Platform for Real-Time Flight Delay Prediction via Rotation-Chain Propagation Features and Agentic Conversational AI

    cs.LG 2026-05 unverdicted novelty 5.0

    FlightSense boosts flight delay prediction AUC to 0.879 by adding 11 aircraft rotation-chain propagation features and weather inputs in a deployed AWS system with agentic conversational AI.

Reference graph

Works this paper leans on

86 extracted references · 86 canonical work pages · cited by 1 Pith paper · 17 internal anchors

  1. [1]

    G. Enea, M. Porretta, A comparison of 4d-trajectory operations envisioned for ’nextgen’ and ’sesar’, some preliminary findings, in: Proceedings of the 28th Congress of the International Council of the Aeronautical Sciences, 2012, pp. 23–28

    work page 2012

  2. [2]

    López-Lago, J

    M. López-Lago, J. Serna, R. Casado, A. Bermúdez, Present and future of air navigation: Pbn operations and supporting technologies, International Journal of Aeronautical and Space Sciences 21 (2) (2019) 451–468.doi:10.1007/s42405-019-00216-y

    work page doi:10.1007/s42405-019-00216-y 2019

  3. [3]

    Mueller, G

    E. Mueller, G. Chatterji, Analysis of aircraft arrival and departure delay characteristics, in: AIAA’s Aircraft Technology, Integration, andOperations(ATIO)2002TechnicalForum,AmericanInstituteof Aeronautics and Astronautics, 2002.doi:10.2514/6.2002-5866

    work page doi:10.2514/6.2002-5866 2002

  4. [4]

    Khanal, R

    A. Khanal, R. Bhusal, K. Subbarao, A. Chakravarthy, W. A. Okolo, Gaussian processes for flight delay prediction: Learning a stochastic process, Journal of Aerospace Information Systems 22 (6) (2025) 457–476.doi:10.2514/1.i011539

    work page doi:10.2514/1.i011539 2025

  5. [5]

    Nigam, K

    R. Nigam, K. Govinda, Cloud based flight delay prediction using logistic regression, in: 2017 International Conference on Intelligent Sustainable Systems (ICISS), 2017, pp. 662–667.doi:10.1109/ISS1. 2017.8389254

    work page doi:10.1109/iss1 2017

  6. [6]

    Q. Li, R. Jing, Z. S. Dong, Flight delay prediction with priority information of weather and non-weather features, IEEE Transactions onIntelligentTransportationSystems24(7)(2023)7149–7165. doi: 10.1109/TITS.2023.3270743

    work page doi:10.1109/tits.2023.3270743 2023

  7. [7]

    Y. Tang, Airline flight delay prediction using machine learning models, in: Proceedings of the 2021 5th International Conference on E-Business and Internet, ICEBI ’21, Association for Computing Machinery, New York, NY, USA, 2022, p. 151–154.doi:10.1145/ 3497701.3497725

    work page arXiv 2021

  8. [8]

    Hatıpoğlu, Ö

    I. Hatıpoğlu, Ö. Tosun, Predictive modeling of flight delays at an airport using machine learning methods, Applied Sciences 14 (13) (2024) 5472. doi:10.3390/app14135472

    work page doi:10.3390/app14135472 2024

  9. [9]

    Alfarhood, R

    M. Alfarhood, R. Alotaibi, B. Abdulrahim, A. Einieh, M. Almousa, A.Alkhanifer,Predictingflightdelayswithmachinelearning:Acase study from saudi arabian airlines, International Journal of Aerospace Engineering 2024 (2024) 1–12.doi:10.1155/2024/3385463

    work page doi:10.1155/2024/3385463 2024

  10. [10]

    Vo, T.-V

    M.-T. Vo, T.-V. Tran, D.-T. Pham, T.-H. Do, A practical real- time flight delay prediction system using big data technology, in: 2022 IEEE International Conference on Communication, Networks and Satellite (COMNETSAT), 2022, pp. 160–167. doi:10.1109/ COMNETSAT56033.2022.9994427

    work page arXiv 2022

  11. [11]

    W. Shao, A. Prabowo, S. Zhao, S. Tan, P. Koniusz, J. Chan, X. Hei, B.Feest,F.D.Salim,Flightdelaypredictionusingairportsituational awareness map, in: Proceedings of the 27th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems, SIGSPATIAL ’19, Association for Computing Machinery, New York, NY, USA, 2019, p. 432–435. doi:10....

    work page doi:10.1145/3347146 2019

  12. [12]

    W. Wu, K. Cai, Y. Yan, Y. Li, An improved svm model for flight delayprediction,in:2019IEEE/AIAA38thDigitalAvionicsSystems Conference (DASC), 2019, pp. 1–6.doi:10.1109/DASC43569.2019. 9081611

    work page doi:10.1109/dasc43569.2019 2019

  13. [13]

    B.Yu,Z.Guo,S.Asian,H.Wang,G.Chen,Flightdelaypredictionfor commercial air transport: A deep learning approach, Transportation Research Part E: Logistics and Transportation Review 125 (2019) 203–221.doi:https://doi.org/10.1016/j.tre.2019.03.013

    work page doi:10.1016/j.tre.2019.03.013 2019

  14. [14]

    Z.Wang,C.Liao,X.Hang,L.Li,D.Delahaye,M.Hansen,Distribu- tionpredictionofstrategicflightdelaysviamachinelearningmethods, Sustainability 14 (22) (2022).doi:10.3390/su142215180

    work page doi:10.3390/su142215180 2022

  15. [15]

    Evangeline, R

    A. Evangeline, R. C. Joy, A. A. Rajan, Flight delay prediction using different regression algorithms in machine learning, in: 2023 4th International Conference on Signal Processing and Communication (ICSPC),2023,pp.262–266. doi:10.1109/ICSPC57692.2023.10125675

    work page doi:10.1109/icspc57692.2023.10125675 2023

  16. [16]

    R. T. Reddy, P. Basa Pati, K. Deepa, S. T. Sangeetha, Flight delay prediction using machine learning, in: 2023 IEEE 8th International Conference for Convergence in Technology (I2CT), 2023, pp. 1–5. doi:10.1109/I2CT57861.2023.10126220

    work page doi:10.1109/i2ct57861.2023.10126220 2023

  17. [17]

    Y. J. Kim, S. Choi, S. Briceno, D. Mavris, A deep learning approach toflightdelayprediction,in:2016IEEE/AIAA35thDigitalAvionics SystemsConference(DASC),2016,pp.1–6. doi:10.1109/DASC.2016. 7778092

    work page doi:10.1109/dasc.2016 2016

  18. [18]

    G. Gui, F. Liu, J. Sun, J. Yang, Z. Zhou, D. Zhao, Flight delay prediction based on aviation big data and machine learning, IEEE Transactions on Vehicular Technology 69 (1) (2020) 140–150.doi: 10.1109/TVT.2019.2954094

    work page doi:10.1109/tvt.2019.2954094 2020

  19. [19]

    Q. Li, R. Jing, Flight delay prediction from spatial and temporal perspective, Expert Systems with Applications 205 (2022) 117662. doi:10.1016/j.eswa.2022.117662

    work page doi:10.1016/j.eswa.2022.117662 2022

  20. [20]

    Q. Li, X. Guan, J. Liu, A cnn-lstm framework for flight delay pre- diction, Expert Systems with Applications 227 (2023) 120287.doi: 10.1016/j.eswa.2023.120287

    work page doi:10.1016/j.eswa.2023.120287 2023

  21. [21]

    J. Qu, S. Wu, J. Zhang, Flight delay propagation prediction based on deep learning, Mathematics 11 (3) (2023) 494. doi:10.3390/ math11030494

    work page 2023

  22. [22]

    K. Cai, Y. Li, Y.-P. Fang, Y. Zhu, A deep learning approach for flight delay prediction through time-evolving graphs, IEEE Transactions on Intelligent Transportation Systems 23 (8) (2022) 11397–11407. doi:10.1109/TITS.2021.3103502

    work page doi:10.1109/tits.2021.3103502 2022

  23. [23]

    X. Shen, J. Chen, R. Yan, A spatial–temporal model for network- wideflightdelaypredictionbasedonfederatedlearning,AppliedSoft Computing 154 (2024) 111380.doi:10.1016/j.asoc.2024.111380

    work page doi:10.1016/j.asoc.2024.111380 2024

  24. [24]

    J. L. Franco, M. V. M. Neto, F. A. Verri, D. R. Amancio, Graph machinelearningforflightdelaypredictionduetoholdingmanouver, arXiv preprint arXiv:2502.04233 (2025)

    work page arXiv 2025

  25. [25]

    Chaudhuri, S

    T. Chaudhuri, S. Zhang, Y. Zhang, Attention-based deep learning model for flight delay prediction using real-time trajectory, SESAR Innovation Days Conference 2024 (2024) 2024–006

    work page 2024

  26. [26]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. u. Kaiser, I. Polosukhin, Attention is all you need, in: I. Guyon, U. V. Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vish- wanathan, R. Garnett (Eds.), Advances in Neural Information Pro- cessing Systems, Vol. 30, Curran Associates, Inc., 2017

    work page 2017

  27. [27]

    Radford, K

    A. Radford, K. Narasimhan, T. Salimans, I. Sutskever, et al., Improv- ing language understanding by generative pre-training (2018)

    work page 2018

  28. [28]

    Radford, J

    A. Radford, J. Wu, R. Child, D. Luan, D. Amodei, I. Sutskever, Language models are unsupervised multitask learners (2019)

    work page 2019

  29. [29]

    BERT : Pre-training of Deep Bidirectional Transformers for Language Understanding

    J. Devlin, M.-W. Chang, K. Lee, K. Toutanova, BERT: Pre-training of deep bidirectional transformers for language understanding, in: J. Burstein, C. Doran, T. Solorio (Eds.), Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1, Association for Computational...

    work page doi:10.18653/v1/n19-1423 2019

  30. [30]

    Raffel, N

    C. Raffel, N. Shazeer, A. Roberts, K. Lee, S. Narang, M. Matena, Y. Zhou, W. Li, P. J. Liu, Exploring the limits of transfer learning with a unified text-to-text transformer, Journal of machine learning research 21 (140) (2020) 1–67

    work page 2020

  31. [31]

    Brown, B

    T. Brown, B. Mann, N. Ryder, M. Subbiah, J. D. Kaplan, P. Dhari- wal, A. Neelakantan, P. Shyam, G. Sastry, A. Askell, S. Agarwal, A. Herbert-Voss, G. Krueger, T. Henighan, R. Child, A. Ramesh, D.Ziegler,J.Wu,C.Winter,C.Hesse,M.Chen,E.Sigler,M.Litwin, S. Gray, B. Chess, J. Clark, C. Berner, S. McCandlish, A. Radford, I. Sutskever, D. Amodei, Language model...

    work page 2020

  32. [32]

    J. Wei, M. Bosma, V. Y. Zhao, K. Guu, A. W. Yu, B. Lester, N. Du, A. M. Dai, Q. V. Le, Finetuned language models are zero-shot learners, arXiv preprint arXiv:2109.01652 (2021)

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  33. [33]

    Training language models to follow instructions with human feedback

    L. Ouyang, J. Wu, X. Jiang, D. Almeida, C. Wainwright, P. Mishkin, C. Zhang, S. Agarwal, K. Slama, A. Ray, et al., Training language models to follow instructions with human feedback, arXiv preprint arXiv:2203.02155 (2022)

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  34. [34]

    LLaMA: Open and Efficient Foundation Language Models

    H. Touvron, T. Lavril, G. Izacard, X. Martinet, M.-A. Lachaux, T. Lacroix, B. Rozière, N. Goyal, E. Hambro, F. Azhar, et al., Llama: Open and efficient foundation language models, arXiv preprint arXiv:2302.13971 (2023)

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  35. [35]

    Llama 2: Open Foundation and Fine-Tuned Chat Models

    H. Touvron, L. Martin, K. Stone, P. Albert, A. Almahairi, Y. Babaei, N. Bashlykov, S. Batra, P. Bhargava, S. Bhosale, et al., Llama 2: Open foundation and fine-tuned chat models, arXiv preprint arXiv:2307.09288 (2023)

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  36. [36]

    The Llama 3 Herd of Models

    A. Grattafiori, A. Dubey, A. Jauhri, A. Pandey, A. Kadian, A. Al- Dahle, A. Letman, A. Mathur, A. Schelten, A. Vaughan, et al., The llama 3 herd of models, arXiv preprint arXiv:2407.21783 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  37. [37]

    A. Q. Jiang, A. Sablayrolles, A. Mensch, C. Bamford, D. S. Chaplot, D. de las Casas, F. Bressand, G. Lengyel, G. Lample, L. Saulnier, L.R.Lavaud,M.-A.Lachaux,P.Stock,T.L.Scao,T.Lavril,T.Wang, T.Lacroix,W.E.Sayed,Mistral7b,arXivpreprintarXiv:2310.06825 (2023)

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  38. [38]

    A.Q.Jiang,A.Sablayrolles,A.Roux,A.Mensch,B.Savary,C.Bam- ford, D. S. Chaplot, D. de las Casas, E. B. Hanna, F. Bressand, G. Lengyel, G. Bour, G. Lample, L. R. Lavaud, L. Saulnier, M.-A. Lachaux,P.Stock,S.Subramanian,S.Yang,S.Antoniak,T.L.Scao, T. Gervet, T. Lavril, T. Wang, T. Lacroix, W. E. Sayed, Mixtral of experts, arXiv preprint arXiv:2401.04088 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  39. [39]

    Gemma: Open Models Based on Gemini Research and Technology

    T. Mesnard, et al., Gemma: Open models based on gemini research and technology, arXiv preprint arXiv:2403.08295 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  40. [40]

    Gemma 2: Improving Open Language Models at a Practical Size

    M. Riviere, et al., Gemma 2: Improving open language models at a practical size, arXiv preprint arXiv:2408.00118 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  41. [41]

    Gemma 3 Technical Report

    A. Kamath, et al., Gemma 3 technical report, arXiv preprint arXiv:2503.19786 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  42. [42]

    Textbooks Are All You Need

    S. Gunasekar, Y. Zhang, J. Aneja, C. C. T. Mendes, A. D. Giorno, S. Gopi, M. Javaheripi, P. Kauffmann, G. de Rosa, O. Saarikivi, A. Salim, S. Shah, H. S. Behl, X. Wang, S. Bubeck, R. Eldan, A. T. Kalai, Y. T. Lee, Y. Li, Textbooks are all you need, arXiv preprint arXiv:2306.11644 (2023)

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  43. [43]

    M. Javaheripi, et al., Phi-2: The surprising power of small language models, https://www.microsoft.com/en-us/research/blog/ phi-2-the-surprising-power-of-small-language-models , accessed: 21 February 2025 (2023)

    work page 2025

  44. [44]

    Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone

    M. Abdin, et al., Phi-3 technical report: A highly capable lan- guagemodellocallyonyourphone,arXivpreprintarXiv:2404.14219 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  45. [45]

    J.Bai,etal.,Qwentechnicalreport,arXivpreprintarXiv:2309.16609 (2023)

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  46. [46]

    Qwen Team, Introducing qwen1.5, https://qwenlm.github.io/blog/ qwen1.5/ (February 2024)

    work page 2024

  47. [47]

    Qwen2 Technical Report

    A. Yang, et al., Qwen2 technical report, arXiv preprint arXiv:2407.10671 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  48. [48]

    Qwen2.5 Technical Report

    A. Yang, et al., Qwen2.5 technical report, arXiv preprint arXiv:2412.15115 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  49. [49]

    Qwen3 Technical Report

    A. Yang, et al., Qwen3 technical report, arXiv preprint arXiv:2505.09388 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  50. [50]

    D.Guo,etal.,Deepseek-r1:Incentivizingreasoningcapabilityinllms via reinforcement learning, arXiv preprint arXiv:2501.12948 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  51. [51]

    Olive, L

    X. Olive, L. Basora, B. Viry, R. Alligier, Deep trajectory clustering withautoencoders,in:ICRAT2020,9thInternationalConferencefor Research in Air Transportation, 2020

    work page 2020

  52. [52]

    36, 2022, pp

    Z.Yue,Y.Wang,J.Duan,T.Yang,C.Huang,Y.Tong,B.Xu,TS2vec: Towards universal representation of time series, in: Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 36, 2022, pp. 8980–8987

    work page 2022

  53. [53]

    Zhang, Z

    X. Zhang, Z. Zhao, T. Tsiligkaridis, M. Zitnik, Self-supervised con- trastive pre-training for time series via time-frequency consistency, Advances in neural information processing systems 35 (2022) 3988– 4003

    work page 2022

  54. [54]

    J. Dong, H. Wu, H. Zhang, L. Zhang, J. Wang, M. Long, Simmtm: A simple pre-training framework for masked time-series modeling, in: A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt, S. Levine (Eds.),AdvancesinNeuralInformationProcessingSystems,Vol.36, Curran Associates, Inc., 2023, pp. 29996–30025

    work page 2023

  55. [55]

    D. Luo, W. Cheng, Y. Wang, D. Xu, J. Ni, W. Yu, X. Zhang, Y. Liu, Y. Chen, H. Chen, et al., Time series contrastive learning with information-aware augmentations, in: Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 37, 2023, pp. 4534–4542

    work page 2023

  56. [56]

    Phisannupawong, J

    T. Phisannupawong, J. J. Damanik, H.-L. Choi, Aircraft trajec- tory segmentation-based contrastive coding: A framework for self- supervisedtrajectoryrepresentation,IEEEOpenJournalofIntelligent Transportation Systems (2025) 1–1doi:10.1109/OJITS.2025.3574746

    work page doi:10.1109/ojits.2025.3574746 2025

  57. [57]

    Chang, W.-Y

    C. Chang, W.-Y. Wang, W.-C. Peng, T.-F. Chen, Llm4ts: Aligning pre-trainedllmsasdata-efficienttime-seriesforecasters,ACMTrans- actions on Intelligent Systems and Technology 16 (3) (Apr. 2025). doi:10.1145/3719207

    work page doi:10.1145/3719207 2025

  58. [58]

    e. a. Zhou, One fits all: Power general time series analysis by pre- trained lm, arXiv preprint arXiv:2302.11939 (2023)

    work page arXiv 2023

  59. [59]

    M. Jin, S. Wang, L. Ma, Z. Chu, J. Y. Zhang, X. Shi, P.-Y. Chen, Y. Liang, Y.-F. Li, S. Pan, Q. Wen, Time-LLM: Time series fore- casting by reprogramming large language models, in: International Conference on Learning Representations (ICLR), 2024

    work page 2024

  60. [60]

    L.Wang,J.Chou,A.Tien,X.Zhou,D.Baumgartner,Aviationgpt:A large language model for the aviation domain, in: AIAA AVIATION FORUM AND ASCEND 2024, American Institute of Aeronautics and Astronautics, 2024.doi:10.2514/6.2024-4250

    work page doi:10.2514/6.2024-4250 2024

  61. [61]

    C.Chandra,X.Jing,M.V.Bendarkar,K.Sawant,L.Elias,M.Kirby, D. N. Mavris, Aviation-bert: A preliminary aviation-specific natural languagemodel,in:AIAAAVIATION2023Forum,AmericanInsti- tuteofAeronauticsandAstronautics,2023. doi:10.2514/6.2023-3436

    work page doi:10.2514/6.2023-3436 2023

  62. [62]

    Tikayat Ray, O

    A. Tikayat Ray, O. J. Pinon-Fischer, D. N. Mavris, R. T. White, B. F. Cole, aerobert-ner: Named-entity recognition for aerospace requirementsengineeringusingbert,in:AIAASciTech2023Forum, American Institute of Aeronautics and Astronautics, 2023, p. 2583. doi:10.2514/6.2023-2583

    work page doi:10.2514/6.2023-2583 2023

  63. [63]

    Tikayat Ray, B

    A. Tikayat Ray, B. F. Cole, O. J. Pinon Fischer, R. T. White, D. N. Mavris, aerobert-classifier: Classification of aerospace re- quirements using bert, Aerospace 10 (3) (2023) 279.doi:10.3390/ aerospace10030279

    work page 2023

  64. [64]

    Moreira, D

    G. Moreira, D. Pleffken, W. Santos, C. Cerqueira, M. Gotelip, Using llms to automate means of compliance assignment in aerospace defense systems, techrxiv preprint (2024). doi:10.36227/techrxiv. 173156107.77200429/v1

    work page doi:10.36227/techrxiv 2024

  65. [65]

    Abdulhak, W

    S. Abdulhak, W. Hubbard, K. Gopalakrishnan, M. Z. Li, Chatatc: Large language model-driven conversational agents for supporting strategic air traffic flow management, arXiv preprint arXiv:2402.14850 (2024)

    work page arXiv 2024

  66. [66]

    Saldiran, M

    A. Tabrizian, P. Gupta, A. Taye, J. Jones, E. Thompson, S. Chen, T.Bonin,D.Eberle,P.Wei,Usinglargelanguagemodelstoautomate flight planning under wind hazards, in: 2024 AIAA DATC/IEEE 43rd Digital Avionics Systems Conference (DASC), 2024, pp. 1–8. doi:10.1109/DASC62030.2024.10749512. : Preprint submitted to Elsevier Page 24 of 25

    work page doi:10.1109/dasc62030.2024.10749512 2024

  67. [67]

    W. Zhou, J. Wang, L. Zhu, Y. Wang, Y. Ji, Flight arrival scheduling via large language model, Aerospace 11 (10) (2024) 813.doi:10. 3390/aerospace11100813

    work page 2024

  68. [68]

    Andriuškevičius, J

    J. Andriuškevičius, J. Sun, Automatic control with human-like rea- soning: Exploring language model embodied air traffic agents, arXiv preprint arXiv:2409.09717 (2024)

    work page arXiv 2024

  69. [69]

    J. M. Hoekstra, J. Ellerbroek, Bluesky atc simulator project: an open data and open source approach, in: Proceedings of the 7th international conference on research in air transportation, Vol. 131, FAA/Eurocontrol Washington, DC, USA, 2016, p. 132

    work page 2016

  70. [70]

    D. Guo, E. Q. Wu, Y. Wu, J. Zhang, R. Law, Y. Lin, Flightbert: Binary encoding representation for flight trajectory prediction, IEEE Transactions on Intelligent Transportation Systems 24 (2) (2023) 1828–1842.doi:10.1109/TITS.2022.3219923

    work page doi:10.1109/tits.2022.3219923 2023

  71. [71]

    D. Guo, Z. Zhang, Z. Yan, J. Zhang, Y. Lin, Flightbert++: A non- autoregressive multi-horizon flight trajectory prediction framework, ProceedingsoftheAAAIConferenceonArtificialIntelligence38(1) (2024) 127–134.doi:10.1609/aaai.v38i1.27763

    work page doi:10.1609/aaai.v38i1.27763 2024

  72. [72]

    K.Luo,J.Zhou,Largelanguagemodelsforsingle-stepandmulti-step flight trajectory prediction, arXiv preprint arXiv:2501.17459 (2025)

    work page arXiv 2025

  73. [73]

    Q.Zhang,J.H.Mott,Anexploratoryassessmentofllms’potentialfor flight trajectory reconstruction analysis, Mathematics 13 (11) (2025)

    work page 2025

  74. [74]

    doi:10.3390/math13111775

    work page doi:10.3390/math13111775

  75. [75]

    D. Guo, Z. Zhang, B. Yang, J. Zhang, H. Yang, Y. Lin, Integrating spoken instructions into flight trajectory prediction to optimize au- tomationinairtrafficcontrol,NatureCommunications15(1)(2024). doi:10.1038/s41467-024-54069-5

    work page doi:10.1038/s41467-024-54069-5 2024

  76. [76]

    MinistryofLand,InfrastructureandTransport,SouthKorea,Airpor- tal, https://www.airportal.go.kr/, accessed: 7 April 2023 (2024)

    work page 2023

  77. [77]

    D. W. Wragg, A Dictionaryof Aviation, 1st Edition, Osprey Publish- ing Ltd., Reading, Berkshire, UK, 1973, p. 189

    work page 1973

  78. [78]

    International Civil Aviation Organization, PANS-ATM Doc 4444: Procedures for Air Navigation Services - Air Traffic Management, sixteenth Edition, ICAO, 2016

    work page 2016

  79. [79]

    G. B. Valor, Ogimet: Weather data portal,https://www.ogimet.com/ home.phtml.en, accessed: 9 June 2025 (2025)

    work page 2025

  80. [80]

    MinistryofLand,InfrastructureandTransport,AeronauticalInforma- tion Management (AIM),https://aim.koca.go.kr/xNotam/, accessed: 9 June 2025 (2025)

    work page 2025

Showing first 80 references.