pith. sign in

arxiv: 2605.14855 · v1 · pith:EFFEBI2Vnew · submitted 2026-05-14 · 💻 cs.LG · cs.AI· eess.SP

Exploitation of Hidden Context in Dynamic Movement Forecasting: A Neural Network Journey from Recurrent to Graph Neural Networks and General Purpose Transformers

Lukas Schelenz , Shobha Rajanna , Denis Gosalci , Lucas Heublein , Jonas Pirkl , Jonathan Ott , Felix Ott , Christopher Mutschler
show 1 more author
Tobias Feigl
This is my paper

Pith reviewed 2026-06-30 21:22 UTC · model grok-4.3

classification 💻 cs.LG cs.AIeess.SP
keywords trajectory forecastingLSTMgraph neural networkstransformersNBA player movementscontextual informationfinal displacement errordynamic movement prediction
0
0 comments X

The pith

A context-augmented LSTM achieves the lowest final displacement error of 1.51 meters when forecasting NBA player trajectories, outperforming graph attention networks and transformers.

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

The paper evaluates linear models and several machine learning architectures for predicting the movements of NBA players over short horizons. It shows that ML methods deliver large gains over linear baselines for forecasts up to two seconds ahead. Among the neural models, the hybrid LSTM that receives both past trajectories and explicit contextual features records the smallest final displacement error while also using less training data and time than the graph attention and transformer variants. The work concludes that no architecture dominates every performance dimension and that choices must be made according to the specific demands of fast, interactive motion prediction.

Core claim

When forecasting NBA player trajectories, a hybrid LSTM that receives both temporal history and contextual information attains a final displacement error of 1.51 m. This result is lower than the errors produced by a temporal convolutional network, a graph attention network, and a transformer, while the same LSTM also requires less data and training time than the graph and transformer models.

What carries the argument

Hybrid LSTM augmented with contextual information, which combines recurrent temporal modeling with explicit side features that describe player interactions and game state.

If this is right

  • Machine-learning models produce substantially lower errors than linear methods for prediction horizons up to two seconds.
  • Adding contextual features to an LSTM yields the smallest final displacement error among the tested networks.
  • The context-augmented LSTM uses less training data and wall-clock time than graph attention or transformer alternatives.
  • No single architecture wins on every metric of accuracy, generalizability, and computational cost.

Where Pith is reading between the lines

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

  • The performance edge of the context-augmented LSTM implies that explicit interaction features can substitute for the more expensive relational layers used in graph or attention models.
  • The same modeling pattern may transfer to other multi-agent forecasting settings where state descriptors are already available, such as traffic or pedestrian flows.
  • Task-specific tuning of history length and context richness appears more decisive than the choice of base architecture.

Load-bearing premise

The NBA player trajectory dataset together with its supplied contextual features stands in for the chaotic, interactive dynamics that would appear in real-world deployment.

What would settle it

Retraining and testing the same four architectures on a different multi-agent trajectory corpus, such as soccer match data, and checking whether the 1.51 m FDE advantage and resource ranking persist.

Figures

Figures reproduced from arXiv: 2605.14855 by Christopher Mutschler, Denis Gosalci, Felix Ott, Jonas Pirkl, Jonathan Ott, Lucas Heublein, Lukas Schelenz, Shobha Rajanna, Tobias Feigl.

Figure 1
Figure 1. Figure 1: Preprocessing of the NBA dataset. Summary. Existing literature does not sufficiently address the complex interplay between temporal dependencies, inter￾player interactions, and the dynamic context inherent in sports environments. Traditional statistical models are overly sim￾plistic, relying on assumptions that limit their applicability in capturing the nuanced, non-linear motion patterns charac￾teristic o… view at source ↗
Figure 2
Figure 2. Figure 2: The game court in the NBA dataset with context. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Pipeline of the CNN-LSTM model [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Pipeline of the GNN model. Our model utilizes two LMU layers. The first LMU layer transforms the input, which consists of 22 dimensions (11 objects, each with two features, i.e., positions and velocities), into a hidden size of d. A dropout layer is applied after the first LMU layer, followed by a second LMU layer with the same dimensionality. The final fully connected layer projects the output back to two… view at source ↗
Figure 5
Figure 5. Figure 5: Architecture (left) with positional encoding (right) of [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Visualization of the data flow of an encoder block. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Comparison of trajectory forecasting for different input lengths (from left to right: 0.04 [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Evaluation error (in m) of different input lengths. believe this is due to the highly dynamic nature of basketball: players frequently change direction, making motion patterns beyond 2s less predictive. In contrast, preliminary tests on soccer and pedestrian data suggest that significantly longer input horizons are beneficial in those domains, as movements are generally more stable and less prone to sudden… view at source ↗
Figure 9
Figure 9. Figure 9: Evaluation of all models. An optimized constant velocity KF works best below 0.12, [PITH_FULL_IMAGE:figures/full_fig_p010_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Evaluation for unseen games. TABLE III: Difference in error between Experiment 2 and 3. Model ∆ADE [m] ∆FDE [m] ∆AAE [◦] ∆FAE [◦] CNN-LSTM 0.01 0.02 0.87 0.68 LMU 0.04 0.07 4.02 2.92 Transformer 0.02 0.05 4.70 2.84 GNN 0.04 0.03 2.47 2.78 may outperform attention-based models such as GNNs (with GAT) and Transformers, particularly in data-scarce settings and dynamic sports. VI. SUMMARY This paper evaluated… view at source ↗
read the original abstract

Forecasting within signal processing pipelines is crucial for mitigating delays, particularly in predicting the dynamic movements of objects such as NBA players. This task poses significant challenges due to the inherently interactive and unpredictable nature of sports, where abrupt changes in velocity and direction are prevalent. Traditional approaches, including (S)ARIMA(X), Kalman filters (KF), and Particle filters (PF), often struggle to model the non-linear dynamics present in such scenarios. Machine learning (ML) methods, such as long short-term memory (LSTM) networks, graph neural networks (GNNs), and Transformers, offer greater flexibility and accuracy but frequently fail to explicitly capture the interplay between temporal dependencies and contextual interactions, which are critical in chaotic sports environments. In this paper, we evaluate these models and assess their strengths and weaknesses. Experimental results reveal key performance trade-offs across input history length, generalizability, and the ability to incorporate contextual information. ML-based methods demonstrated substantial improvements over linear models across forecast horizons of up to 2s. Among the tested architectures, our hybrid LSTM augmented with contextual information achieved the lowest final displacement error (FDE) of 1.51m, outperforming temporal convolutional neural network (TCNN), graph attention network (GAT), and Transformers, while also requiring less data and training time compared to GAT and Transformers. Our findings indicate that no single architecture excels across all metrics, emphasizing the need for task-specific considerations in trajectory prediction for fast-paced, dynamic environments such as NBA gameplay.

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 / 0 minor

Summary. The manuscript conducts an empirical benchmark of machine learning models (LSTM variants, TCNN, GAT, Transformers) for forecasting NBA player trajectories up to 2s horizons. It reports that a hybrid LSTM augmented with contextual information achieves the lowest FDE of 1.51m, outperforms the other architectures, and requires less data and training time than GAT and Transformers, while highlighting performance trade-offs across input history length and generalizability.

Significance. If the reported performance differences hold under rigorous verification, the work provides concrete evidence that task-specific incorporation of contextual features can improve trajectory forecasting in interactive dynamic settings and that architecture choice involves explicit trade-offs in accuracy, data efficiency, and compute; the concrete FDE numbers and qualitative comparisons constitute a useful reference point for similar signal-processing applications.

major comments (2)
  1. [performance comparison section] Performance comparison section: the central claim that the hybrid LSTM achieves the lowest FDE of 1.51m and outperforms TCNN, GAT, and Transformers is presented without error bars, statistical significance tests, details on train/validation/test splits, or hyperparameter selection protocol, rendering the quantitative outperformance unverifiable from the reported results.
  2. [Abstract and performance comparison section] Abstract and performance comparison section: the claim that results generalize to chaotic, interactive dynamics rests on the NBA dataset and its contextual features being representative, yet no quantitative characterization of dataset variability (e.g., statistics on abrupt velocity changes or multi-agent interaction density) or comparison against other trajectory corpora is supplied to support this assumption.

Simulated Author's Rebuttal

2 responses · 0 unresolved

Thank you for the opportunity to respond to the referee's comments. We address each major comment below and indicate the revisions we plan to make to the manuscript.

read point-by-point responses

  1. Referee: [performance comparison section] Performance comparison section: the central claim that the hybrid LSTM achieves the lowest FDE of 1.51m and outperforms TCNN, GAT, and Transformers is presented without error bars, statistical significance tests, details on train/validation/test splits, or hyperparameter selection protocol, rendering the quantitative outperformance unverifiable from the reported results.

    Authors: We agree that the absence of error bars, statistical tests, split details, and hyperparameter protocol limits the verifiability of the results. The reported FDE of 1.51m is from our primary experimental run, but to address this, we will conduct additional experiments with multiple seeds to report means and standard deviations, perform statistical significance tests, and include a detailed description of the train/validation/test splits (e.g., chronological split to avoid leakage) and the hyperparameter search protocol in the revised manuscript. revision: yes

  2. Referee: [Abstract and performance comparison section] Abstract and performance comparison section: the claim that results generalize to chaotic, interactive dynamics rests on the NBA dataset and its contextual features being representative, yet no quantitative characterization of dataset variability (e.g., statistics on abrupt velocity changes or multi-agent interaction density) or comparison against other trajectory corpora is supplied to support this assumption.

    Authors: The manuscript positions the NBA dataset as an exemplar of chaotic, interactive dynamics based on its established use in trajectory forecasting literature for sports. However, we acknowledge the lack of explicit quantitative characterization and cross-corpus comparison in the current version. In the revision, we will add a dataset analysis section providing statistics on velocity changes and interaction density, and discuss the generalizability to other domains with appropriate caveats. revision: yes

Circularity Check

0 steps flagged

No circularity: purely empirical benchmark study

full rationale

The paper is an empirical comparison of LSTM, TCNN, GAT, and Transformer models on NBA trajectory data, reporting measured FDE values such as 1.51m for the hybrid LSTM. No equations, derivations, fitted parameters renamed as predictions, or self-citation chains appear in the abstract or described content. All performance claims rest on direct experimental outcomes rather than any reduction to inputs by construction. The dataset representativeness assumption is external to any derivation and does not trigger the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No mathematical derivations, free parameters, axioms, or invented entities appear in the abstract; the work is an empirical model comparison.

pith-pipeline@v0.9.1-grok · 5841 in / 1035 out tokens · 30378 ms · 2026-06-30T21:22:31.847901+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

49 extracted references · 11 canonical work pages · 6 internal anchors

  1. [1]

    Guest editorial: Special issue on sports analytics,

    U. Brefeld and A. Zimmermann, “Guest editorial: Special issue on sports analytics,” inData Mining and Knowledge Discovery Jo., vol. 31, no. 7, 2017, pp. 1577–1579

    2017

  2. [2]

    Stock Market Prediction via Deep Learning Techniques: A Survey,

    J. Zou, Q. Zhao, Y . Jiao, H. Cao, Y . Liu, Q. Yan, E. Abbasnejad, L. Liu, and J. Q. Shi, “Stock Market Prediction via Deep Learning Techniques: A Survey,” 2023

    2023

  3. [3]

    Forecasting Economics and Financial Time Series: ARIMA vs. LSTM

    S. Siami-Namini and A. S. Namin, “Forecasting Economics and Finan- cial Time Series: ARIMA vs. LSTM,” inarXiv:1803.06386 [cs.CV], 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  4. [4]

    Joint Classification and Trajectory Regression of Online Handwriting using a Multi-Task Learning Approach,

    F. Ott, D. R ¨ugamer, L. Heublein, B. Bischl, and C. Mutschler, “Joint Classification and Trajectory Regression of Online Handwriting using a Multi-Task Learning Approach,” inProc. IEEE Conf. Winter Conf. on Applications of Computer Vision (WACV), Waikoloa, HI, 2022

    2022

  5. [5]

    Forecasting of Electricity Prices Using Deep Learning Networks,

    W. Zhang, F. Cheema, and D. Srinivasan, “Forecasting of Electricity Prices Using Deep Learning Networks,” inAsia Pacific Power and Energy Engineering Conf. (APPEEC), 2018, pp. 451–456

    2018

  6. [6]

    A Review on Weather Forecasting Techniques,

    G. Jain and S. Singh, “A Review on Weather Forecasting Techniques,” inIJARCCE, vol. 5, 2016, pp. 177–180

    2016

  7. [7]

    ViPR: Visual-Odometry- aided Pose Regression for 6DoF Camera Localization,

    F. Ott, T. Feigl, C. L ¨offler, and C. Mutschler, “ViPR: Visual-Odometry- aided Pose Regression for 6DoF Camera Localization,” inProc. of the IEEE/CVF Intl. Conf. on Computer Vision and Pattern Recognition Workshops (CVPRW), Seattle, W A, 2020, pp. 187–198

    2020

  8. [8]

    Fusing Structure from Motion and Simulation-Augmented Pose Regression from Optical Flow for Challenging Indoor Environments,

    F. Ott, L. Heublein, D. R ¨ugamer, B. Bischl, and C. Mutschler, “Fusing Structure from Motion and Simulation-Augmented Pose Regression from Optical Flow for Challenging Indoor Environments,” inJo. Visual Communication and Image Representation (JVCIR), vol. 104256, 2024

    2024

  9. [9]

    Sports Analytics and the Big- data Era,

    E. Morgulev, O. H. Azar, and R. Lidor, “Sports Analytics and the Big- data Era,” inIntl. Jo. Data Science and Analytics, vol. 5, no. 1, 2018, pp. 213–222

    2018

  10. [10]

    One-match-ahead Forecasting in Two-team Sports with Stacked Bayesian Regressions,

    M. W. Lam, “One-match-ahead Forecasting in Two-team Sports with Stacked Bayesian Regressions,” inJo. Artificial Intelligence and Soft Computing Research, vol. 8, no. 3, 2018, pp. 159–171

    2018

  11. [11]

    Daten-getriebene Methoden zur Bestimmung von Position und Orientierung in Funk- und Tr ¨agheits-basierter Koppelnavigation,

    T. Feigl, “Daten-getriebene Methoden zur Bestimmung von Position und Orientierung in Funk- und Tr ¨agheits-basierter Koppelnavigation,” Ph.D. dissertation, FAU Erlangen-N ¨urnberg, 2021

    2021

  12. [12]

    Rethinking Trajectory Prediction via Team Game,

    Z. Wei, X. Zhu, B. Dai, and D. Lin, “Rethinking Trajectory Prediction via Team Game,” inarXiv:2210.08793 [cs.CV], 2022, pp. 1–13

    work page arXiv 2022

  13. [13]

    Stochastic trajectory prediction with social graph network

    L. Zhang, Q. She, and P. Guo, “Stochastic Trajectory Prediction with Social Graph Network,” inarXiv:1907.10233 [cs.CV], 2019, pp. 1–12

    work page internal anchor Pith review Pith/arXiv arXiv 1907

  14. [14]

    Hybrid Basketball Game Outcome Prediction Model by Integrating Data Mining Methods for the National Basketball Association,

    W.-J. Chen, M.-J. Jhou, T.-S. Lee, and C.-J. Lu, “Hybrid Basketball Game Outcome Prediction Model by Integrating Data Mining Methods for the National Basketball Association,” inEntropy Jo., vol. 23, no. 477, 2021, pp. 1–17

    2021

  15. [15]

    Ans ¨atze zur Verbesserung KI-basierter Systeme f¨ur das autonome Fahren,

    D. Br ¨uggemann, S. Bracke, H. Gottschalk, M. Rottmann, K. Maag, R. Chan, and M. Schubert, “Ans ¨atze zur Verbesserung KI-basierter Systeme f¨ur das autonome Fahren,” inQualit ¨atsmanagement in den 20er Jahren - Trends und Perspektiven, B. Leyendecker, Ed., 2021, pp. 100– 119

    2021

  16. [16]

    An Introduction to the Kalman Filter,

    G. Welch and G. Bishop, “An Introduction to the Kalman Filter,” University of North Carolina at Chapel Hill, USA, Tech. Rep., 1995

    1995

  17. [17]

    Particle Filtering,

    P. Djuric, J. Kotecha, J. Zhang, Y . Huang, T. Ghirmai, M. Bugallo, and J. Miguez, “Particle Filtering,” inIEEE Signal Processing Magazine, vol. 20, no. 5, 2003, pp. 19–38

    2003

  18. [18]

    Machine Learning Predictive Analytics for Player Movement Prediction in NBA: Applications, Opportunities, and Challenges,

    D. K. Stephanos, G. Husari, B. T. Bennett, and E. Stephanos, “Machine Learning Predictive Analytics for Player Movement Prediction in NBA: Applications, Opportunities, and Challenges,” inProc. Intl. Conf. ACM Southeast (ACM SE), 2021, pp. 2–8

    2021

  19. [19]

    Understanding LSTM – a tutorial into Long Short-Term Memory Recurrent Neural Networks,

    R. C. Staudemeyer and E. R. Morris, “Understanding LSTM – a tutorial into Long Short-Term Memory Recurrent Neural Networks,” inarXiv:1909.09586 [cs.CV], 2019, pp. 1–32

    work page arXiv 1909

  20. [20]

    Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling

    J. Chung, C. Gulcehre, K. Cho, and Y . Bengio, “Empirical Evalua- tion of Gated Recurrent Neural Networks on Sequence Modeling,” in arXiv:1412.3555 [cs.NE], 2014, pp. 1–9

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  21. [21]

    Legendre Memory Units: Continuous-Time Representation in Recurrent Neural Networks,

    A. V oelker, I. Kaji ´c, and C. Eliasmith, “Legendre Memory Units: Continuous-Time Representation in Recurrent Neural Networks,” in Proc. of Conf. Advances in Neural Information Processing Systems (NIPS), 2019, pp. 15 544–15 553

    2019

  22. [22]

    Attention Is All You Need

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, “Attention Is All You Need,” in arXiv:1706.03762 [cs.CV], 8 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  23. [23]

    A Comprehensive Survey on Graph Neural Networks,

    Z. Wu, S. Pan, F. Chen, G. Long, C. Zhang, and S. Y . Philip, “A Comprehensive Survey on Graph Neural Networks,” inIEEE Trans. on Neural Networks and Learning Systems, vol. 32, no. 1. IEEE, 2020

    2020

  24. [24]

    Con- necting the dots: Multivariate time series forecasting with graph neural networks,

    Z. Wu, S. Pan, G. Long, J. Jiang, X. Chang, and C. Zhang, “Con- necting the dots: Multivariate time series forecasting with graph neural networks,” inProc. Intl. conf. on knowledge discovery and data mining (KDD), 2020, pp. 753–763

    2020

  25. [25]

    Adaptive Trajectory Prediction via Transferable GNN,

    Y . Xu, L. Wang, Y . Wang, and Y . Fu, “Adaptive Trajectory Prediction via Transferable GNN,” inProc. of the IEEE/CVF Intl. Conf. on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 6520–6531

    2022

  26. [26]

    The graph neural network model,

    Scarselli, Franco and Gori, Marco and Tsoi, Ah Chung and Hagen- buchner, Markus and Monfardini, Gabriele, “The graph neural network model,” inIEEE Trans. on Neural Networks, vol. 20, no. 1, 2009, pp. 61–80

    2009

  27. [27]

    Graph neural network for traffic forecasting: A survey,

    W. Jiang and J. Luo, “Graph neural network for traffic forecasting: A survey,” inExpert Systems with Applications, vol. 207, 2022, p. 117921

    2022

  28. [28]

    Graph and Recurrent Neural Network-based Vehicle Trajectory Prediction for Highway Driving,

    X. Mo, Y . Xing, and C. Lv, “Graph and Recurrent Neural Network-based Vehicle Trajectory Prediction for Highway Driving,” inProc. Intl. Conf. IEEE Intelligent Transportation Systems Conf. (ITSC), 2021, pp. 1934– 1939

    2021

  29. [29]

    Graph attention networks,

    P. Velickovic, G. Cucurull, A. Casanova, A. Romero, P. Lio, Y . Bengio et al., “Graph attention networks,” instat, vol. 1050, no. 20, 2017

    2017

  30. [30]

    A spatio-temporal transformer for 3d human motion prediction,

    E. Aksan, M. Kaufmann, P. Cao, and O. Hilliges, “A spatio-temporal transformer for 3d human motion prediction,” in3DV, 2021, pp. 565– 574

    2021

  31. [31]

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

    J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding,” in arXiv:1810.04805 [cs.CV], 2019

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  32. [32]

    End-to- end object detection with transformers,

    N. Carion, F. Massa, G. Synnaeve, R. Cadene, and C. Schmid, “End-to- end object detection with transformers,” inEuropean Conf. on Computer Vision (ECCV), 2020

    2020

  33. [33]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, and B. Schiele, “An image is worth 16x16 words: Transformers for image recognition at scale,” inarXiv:2010.11929 [cs.CV], 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  34. [34]

    Social LSTM: Human Trajectory Prediction in Crowded Spaces,

    A. Alahi, K. Goel, V . Ramanathan, A. Robicquet, L. Fei-Fei, and S. Savarese, “Social LSTM: Human Trajectory Prediction in Crowded Spaces,” inProc. Intl. Conf. IEEE Computer Vision and Pattern Recog- nition (CVPR), 2016, pp. 961–971

    2016

  35. [35]

    Social-stgcnn: A Social Spatio-temporal Graph Convolutional Neural Network for Human Trajectory Prediction,

    A. Mohamed, K. Qian, M. Elhoseiny, and C. Claudel, “Social-stgcnn: A Social Spatio-temporal Graph Convolutional Neural Network for Human Trajectory Prediction,” inProc. Intl. Conf. IEEE Computer Vision and Pattern Recognition (CVPR), 2020, pp. 14 424–14 432

    2020

  36. [36]

    arXiv preprint arXiv:1907.07792 (2019)

    X. Li, X. Ying, and M. C. Chuah, “Grip++: Enhanced Graph-based Interaction-aware Trajectory Prediction for Autonomous Driving,” in arXiv:1907.07792 [cs.CV], 2019, pp. 1–16

    work page arXiv 1907

  37. [37]

    Goodfellow, Y

    I. Goodfellow, Y . Bengio, and A. Courville,Convolutional Networks. MIT Press, 2016

    2016

  38. [38]

    Review of Deep Learning: Concepts, CNN Architectures, Challenges, Applications, Future Directions,

    L. Alzubaidi, J. Zhang, A. Humaidi, A. Al-Dujaili, Y . Duan, O. Al- Shamma, J. Santamar ´ıa, M. Fadhel, M. Al-Amidie, and L. Farhan, “Review of Deep Learning: Concepts, CNN Architectures, Challenges, Applications, Future Directions,” inJo. of Big Data, vol. 8, no. 2, 2021, pp. 1–74

    2021

  39. [39]

    Structured prediction helps 3d human motion modelling,

    Aksan, Emre and Kaufmann, Manuel and Hilliges, Otmar, “Structured prediction helps 3d human motion modelling,” inProc. IEEE Conf. Winter Conf. on Applications of Computer Vision (WACV), 2019, pp. 7144–7153

    2019

  40. [40]

    Convolutional Neural Network for Trajec- tory Prediction,

    N. Nikhil and B. T. Morris, “Convolutional Neural Network for Trajec- tory Prediction,” inProc. European Conf. on Computer Vision (ECCV), 2019, pp. 186–196

    2019

  41. [41]

    Long Short-term Memory,

    S. Hochreiter and J. Schmidhuber, “Long Short-term Memory,” in Neural Computation Jo., vol. 9, no. 8, 1997, pp. 1735–1780

    1997

  42. [42]

    Transformer Networks for Trajectory Forecasting,

    F. Giuliari, I. Hasan, M. Cristani, and F. Galasso, “Transformer Networks for Trajectory Forecasting,” inarXiv:2003.08111 [cs.CV], 2020, pp. 1– 14

    work page arXiv 2003

  43. [43]

    Robust Human Motion Forecasting using Transformer-based Model,

    E. V . Mascaro, S. Ma, H. Ahn, and D. Lee, “Robust Human Motion Forecasting using Transformer-based Model,” inProc. Intl. Conf. on Intelligent Robots and Systems (IROS), 2022, pp. 10 674–10 680

    2022

  44. [44]

    Interaction-aware Trajectory Prediction of Connected Vehicles Using CNN-LSTM Networks,

    X. Mo, Y . Xing, and C. Lv, “Interaction-aware Trajectory Prediction of Connected Vehicles Using CNN-LSTM Networks,” inProc. Intl. Conf. IEEE Industrial Electronics Society (IECON), 2020, pp. 5057–5062

    2020

  45. [45]

    A Holistic and Proactive Approach to Forecasting Cyber Threats,

    Z. Almahmoud, P. D. Yoo, O. Alhussein, I. Farhat, and E. Damiani, “A Holistic and Proactive Approach to Forecasting Cyber Threats,” in Nature Scientific Reports Jo., vol. 13, no. 1, 2023, pp. 8049–8061

    2023

  46. [46]

    wav2vec 2.0: A framework for self-supervised learning of speech representations,

    A. Baevski, H. Zhou, A. Mohamed, and M. Auli, “wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations,” inCoRR, vol. abs/2006.11477, 2020

    work page arXiv 2006

  47. [47]

    Temporal convolutional networks for action segmentation and detection,

    C. Lea, M. D. Flynn, R. Vidal, A. Reiter, and G. D. Hager, “Temporal convolutional networks for action segmentation and detection,” inProc. Intl. Conf. on Computer Vision and Pattern Recognition (CVPR), 2017, pp. 156–165

    2017

  48. [48]

    Gatraj: A graph-and attention-based multi-agent trajectory prediction model,

    H. Cheng, M. Liu, L. Chen, H. Broszio, M. Sester, and M. Y . Yang, “Gatraj: A graph-and attention-based multi-agent trajectory prediction model,”ISPRS JPRS, vol. 205, pp. 163–175, 2023

    2023

  49. [49]

    Linou, D

    K. Linou, D. Linou, and M. de Boer. (2016) NBA-Player- Movements Dataset. [Online]. Available: https://github.com/linouk23/ NBA-Player-Movements

    2016