Pith. sign in

REVIEW 4 major objections 4 minor 26 references

Learning Latent Memory States from Longitudinal Athlete Monitoring Data

T0 review · 4 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper proposes the Latent Memory Table as a reusable unit of analysis for longitudinal data, arguing that methods should be compared by the quality of the exported table, not by the encoder that produced it.

desk verdict Interesting proposal for a reusable latent state table, but the key empirical support rests on an unfair baseline comparison and a P2 score that contradicts the paper's own definition. read the letter →

arxiv 2608.06290 v1 pith:5MMRAYBO submitted 2026-08-06 stat.CO stat.MEstat.ML

Dae-Jin Lee This is my paper · ORCID
classification stat.COstat.MEstat.ML MSC 62H2562M1062P1068T07
keywords LatentMemoryTablelongitudinaldataunitofanalysisoperatorpropertysystemqualityindexathletemonitoringmultivariatetimeseries
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper proposes a new unit of analysis for longitudinal data: the Latent Memory Table, a rectangular export of latent states that summarize each unit's recent windowed history, each with an attached uncertainty estimate. The claim is that this table deserves the same treatment as PCA scores or estimated random effects—stored, queried, and reused across analyses—so methods should be compared by the table's quality, not by the encoder that produced it. To support that claim, the paper defines a six-property validation system with a composite quality index Q, shows in simulation that Q discriminates genuine multivariate memory from negative controls and reacts to window misspecification, and reports a soccer-monitoring case study in which the learned table attains Q≈0.73 versus about 0.40 for classical and lagged principal-component baselines. If the claim is right, one validated table can feed many downstream statistical operators without retraining, making the exported table the scientific product rather than the model.

What carries the argument

The central object is the Latent Memory Table T, collected from a memory operator Mθ : HW → Rd that maps a masked, windowed history Hi,t to a d-dimensional latent memory state mi(t), together with per-row ensemble covariance bΣi(t) obtained from a Procrustes-aligned replicate ensemble. The table carries the scientific content, and its validity is assessed by a property system P = {P1, ..., P6} aggregated into the composite quality Q(T) = (1/6) Σ Sj(T). Because latent coordinates are rotation non-identifiable, comparisons use geometry after orthogonal Procrustes alignment rather than raw coordinates, which is why the export is a table of states with uncertainty instead of a named-axis embedding.

What would settle it

Re-run the SoccerMon P6 comparison with a classical baseline that includes the outcome's own recent history (e.g., lagged outcome or an EWMA of it): if latent PCs no longer add held-out R² beyond that baseline, the reusability advantage over classical summaries fails. Conversely, in the simulation framework, add a univariate-memory scenario where the true memory depends on a single channel; if the learned table still scores Q≈0.73 while classical summaries score 0.40, that would suggest Q is inflated by training on the outcome channel.

Watch

Extended reading notes

Core claim

The central claim is that the Latent Memory Table T = {(bmi(t), bΣi(t)) : i = 1,...,N, t ∈ Ti}, formed by a memory operator Mθ mapping each masked windowed history to a finite-dimensional state together with its ensemble covariance, is a reusable unit of analysis on the same footing as principal-component scores or estimated random effects. Methods should be compared by the quality Q of the table, which is an equally weighted aggregate of six property scores measuring recoverability, personalization, temporal coherence, interpretability, stability, and reusability, rather than by the encoder or by endpoint accuracy. Classical fixed summaries such as exponentially weighted moving averages are shown to be degenerate univariate special cases of the same operator class. Empirically, the constructed table attains Q≈0.73 versus about 0.40 for classical and lagged-PCA baselines on the SoccerMon case study, with incremental held-out value for some wellness targets, while simulations with known memory mechanisms show that Q, not regime classification accuracy alone, separates genuine memory from negative controls and responds to window misspecification.

Load-bearing premise

The reusability claim depends on a fair comparison: latent states are built from all channels including the wellness outcome being predicted, while the classical baseline excludes that channel, so the reported gains could partly reflect outcome autocorrelation rather than the value of the table.

Editorial extensions

If this is right

  • Any sequence encoder—Transformer, LSTM, linear map, or PCA of lags—is only an estimator of the memory operator, so comparisons should be made by the quality of the resulting table, not by the architecture.
  • Classical EWMA-style workload summaries and acute:chronic ratios are degenerate cases of the same operator class, so the framework subsumes standard sports-science dashboards rather than competing with them.
  • A P-validated table licenses a family of statistical operators—paths, anomaly scores, neighbourhoods, partitions, and covariate use in mixed, survival, or functional models—without retraining the encoder.
  • Regime classification accuracy is not sufficient to certify a memory table: in the negative-control simulation, structure stays high while interpretability and state recovery collapse, catching a failure that accuracy alone would miss.
  • On SoccerMon, the learned table's quality Q≈0.73 substantially exceeds classical summaries (≈0.41) and lagged PCA (≈0.39), with incremental held-out R² gains for fatigue and readiness and a latent-only advantage for sleep quality.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial inference: the reported reusability gains rest on comparing latent states trained on all twelve channels, including wellness outcomes, against classical summaries that exclude the outcome channel; a baseline that included lagged outcomes might erase the incremental R², so the reusability advantage should be re-tested with outcome-aware baselines.
  • Editorial inference: if the Latent Memory Table becomes a standard export, the same P/Q ladder could be applied to clinical or device-monitoring panels, and the framework predicts that Q, not endpoint accuracy, is the quantity that transfers across settings.
  • Editorial inference: the simulation finding that Q peaks near the true memory horizon suggests window length is an empirical property of the data, so practitioners could use Q-sensitivity grids to choose windows instead of importing domain conventions.
  • Editorial inference: the Procrustes ensemble provides per-row uncertainty, and a natural next step the paper does not take is to propagate that uncertainty into downstream models via measurement-error or Bayesian mixed models on table rows.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper proposes a new unit of analysis for longitudinal data, called the Latent Memory Table (LMT): a table of estimated latent states and uncertainties, produced by a learned memory operator over masked windowed histories, intended to be stored, queried, and reused in the same way as PCA scores or estimated random effects. The method is validated through a six-property system P1–P6 with a composite quality index Q, on a SoccerMon athlete-monitoring case study and on simulations with known memory mechanisms. The paper reports that the learned ensemble table attains Q ≈ 0.73 versus ≈ 0.41 for classical summaries and ≈ 0.39 for lagged PCA, and that the LMT adds incremental held-out predictive value for some wellness targets beyond classical summaries.

Significance. The concept of an exported, validated table of latent memory states is an interesting and potentially useful framing for longitudinal statistics, and the paper includes reproducible code and a clearly stated validation philosophy. If the empirical claims were sound, the proposal could give practitioners a principled alternative to ad hoc embeddings or task-specific features. However, the central evidence for the table's reusability and superiority over classical summaries is undermined by the flawed P6 comparison, an internally inconsistent P2 score, and simulation results in which baseline methods recover the oracle memory better than the learned encoder. The paper's strengths—a formal property system, a composite index, simulation diagnostics, and code release—do not compensate for the fact that the headline quantitative claims currently rest on invalid comparisons.

major comments (4)
  1. [Section 7, Table 7 (P6)] The incremental-value comparison is not a fair test of the LMT's reusability. The latent PCs are computed from bmi(t), whose encoder was trained on all twelve channels including the wellness outcomes (fatigue, sleep quality, readiness) that then serve as next-window prediction targets. The classical baseline, by contrast, explicitly excludes the outcome channel. Under this design, the reported R2 gains (e.g., sleep quality 0.539 vs 0.286) may simply reflect autocorrelation of the target rather than the incremental longitudinal information in the table. Because the Reuse score (0.60 in Table 8) feeds directly into Q, the Q advantage of 0.734 vs 0.407 is also inflated. A valid comparison would include the outcome channel, or its lagged values, in the classical baseline (e.g., an AR(1) term for each target).
  2. [Section 6, Definition 3.5, Table 4, Table 8 (P2)] The personalization property P2 is defined as a strict inequality: within-athlete similarity must dominate cross-athlete similarity after removing adjacency. The reported median cosines are 0.622 for same-athlete pairs and 0.661 for different-athlete/same-regime pairs, so the defining inequality fails. Yet Table 8 assigns the learned table a personalization score of 0.62, approximately equal to the same-athlete cosine. This is internally inconsistent: the paper either needs to revise Definition 3.5 (for example, to require within-athlete similarity to dominate only cross-athlete/different-regime similarity) or correct the score. As written, the paper's own numbers show that the learned table is not P2-valid, and this also affects the composite Q.
  3. [Section 8, Table 11 (Simulation recovery)] The simulation results do not support the claim that the learned LMT recovers known memory mechanisms better than baselines. In Block I, the Transformer's Procrustes correlation with the oracle memory is 0.39, while classical and PCA-lagged baselines achieve 0.46 and 0.47, respectively. In Block II, the Transformer again scores lower (0.40) than classical (0.47). The text acknowledges 'partial recovery' but does not directly address the fact that the baselines recover the oracle memory more accurately. Since the paper's central argument is that the LMT deserves the same status as PCA scores or random effects, a simulation in which PCA-lagged windows outperform the learned state is a direct counterexample that must be confronted, not merely reported in a table.
  4. [Section 4 and Section 6 (P1 recoverability)] The P1 recoverability property is evaluated using the same Team×Season classification criterion that is used as the auxiliary training objective. Although the evaluation is on held-out athletes, the test essentially verifies whether the auxiliary training succeeded, rather than providing independent evidence that the table retains generalizable longitudinal memory beyond the chosen supervision. This circularity weakens the recoverability claim and should be acknowledged explicitly; for example, the authors could evaluate P1 with a held-out regime rotation or a different auxiliary task not used in training.
minor comments (4)
  1. [Section 6, Figure 3] The text describing the three-way cosine comparison states 'Same-athlete similarity remains far above cross-regime similarity', but the more relevant comparison to different-athlete/same-regime similarity (0.622 vs 0.661) is not highlighted; this should be discussed to avoid misleading readers about personalization.
  2. [Section 3, Definition 3.6] Definition 3.6 says a table is P-valid if it satisfies all six properties, yet the paper later reports Q as an equally weighted average and uses it to rank tables. If P-validity requires all properties to hold, then the learned table, by the paper's own numbers, is not P-valid because P2 fails; the relationship between P-validity and Q should be clarified.
  3. [Appendix F] The sensitivity analysis covers only the latent dimension d; the paper states in Section 11 that robustness to LSTM/linear baselines remains future work. This is acceptable as a limitation, but the claim of 'stability' (P5) is only partially demonstrated.
  4. [Section 3, Eq. (6)] The statement 'Alternative weights are sensitivity analyses' is not accompanied by any reported sensitivity results; either a small table or a reference to an appendix would strengthen this point.

Circularity Check

2 steps flagged · score 7.0 of 10

P1 recoverability and P6 incremental value are built into the training objective and input channels, so the headline Q advantage is partly constructed.

  1. self definitional [Section 4 (Estimation); Definition 3.5 (P1); Table 8]
    "We train with an auxiliary Team×Season classification criterion: each window is labelled by the crossed factor of club and season, yielding four classes {Team A–2020, Team A–2021, Team B–2020, Team B–2021}, optimized by weighted cross-entropy. The criterion is not the estimand: it only forces rows of T to retain regime information for testing P1."

    P1 in Definition 3.5 is exactly 'Regime information retained in m_i(t)' and is tested by regime classification. Because the encoder is trained with the Team×Season classification objective and checkpoints are selected by held-out regime accuracy, the P1 test verifies whether that training objective generalized; the regime information is inserted into the table by construction rather than discovered independently. This trained property is then counted in Q (Structure score 0.38 in Table 8), making part of the quality index self-confirmatory.

  2. fitted input called prediction [Definition 3.5 (P6); Section 7, Table 7]
    "P6. Reusability. For a family of targets {Y(1), ..., Y(K)} not used to train Mθ, the fixed table retains incremental predictive information beyond a classical baseline for more than one Y(k). ... we compare next-window fatigue, sleep quality and readiness under player-grouped holdout using (i) classical load/wellness summaries excluding the outcome channel, (ii) the first five PCs of bmi(t) and (iii) both."

    The P6 targets are fatigue, sleep quality and readiness, which are channels in the twelve value channels used to estimate bmi(t) (Sections 4–5). Thus the 'targets not used to train' clause is violated: the table contains the current value of the outcome, while the classical baseline explicitly excludes the outcome channel. The reported gains, e.g. sleep quality 0.54 versus 0.29, can therefore be carried by the outcome's own current value inside the latent state rather than by reusable longitudinal memory. This inflates the Reuse score (0.60) and the headline Q=0.73 versus 0.40 comparison.

full rationale

The central quantitative claim—that the learned Latent Memory Table reaches Q≈0.73 versus about 0.40 for classical and lagged-PCA baselines—is partly constructed. P1 is a self-definitional validation: the table is trained on the Team×Season classification objective and then scored on retaining Team×Season information. P6 is a fitted-input comparison: the model is estimated on all twelve channels including the wellness outcomes, while the classical baseline is required to exclude the outcome channel, so the incremental R2 for fatigue, sleep quality and readiness is statistically forced by outcome autocorrelation. The paper also contains internal consistency concerns that are not themselves circularity but reinforce that Q is not an independent oracle: P2 is scored 0.62 even though the reported median same-athlete cosine (0.622) is below the different-athlete/same-regime cosine (0.661), so the defining inequality in Definition 3.5 fails; and in simulation Block I the learned Transformer Procrustes recovery (0.39) is below the classical baseline (0.46). The manuscript candidly states that classical interpretability is 'partly tautological because the same constructs enter both the table and the association library' and that Q 'should be read as a profile, not a single oracle ranking,' which further undercuts the aggregate comparison. There is genuine independent content—the SoccerMon external dataset, negative controls, code release, and the simulation machinery—so the paper is not wholly circular, but the headline evidence for reusability is partially circular by construction.

Assumptions & free parameters 5 free parameters · 5 assumptions · 1 invented entities

The central claim rests on several ad hoc choices: the property system P and Q are defined by the authors, the auxiliary regime-classification objective is a training choice, and the P6 baseline is biased. The LMT is an invented statistical object whose validation is mostly self-referential, with the simulation providing the only external anchor, but that anchor does not favor the proposed method.

free parameters (5)
  • Window length W=28 days = 28
    Chosen by hand as the chronic-load horizon; the paper explicitly disclaims universal optimality. Simulation shows Q peaks at fitted W=28 when true W0=28, but W is still an operating choice.
  • Latent dimension d=32 = 32
    Chosen with sensitivity analysis over d=16,32,64. d=16 and d=32 perform similarly, d=64 slightly worse; the choice affects the geometry of the table.
  • Stride=7 days = 7
    Chosen as a reporting cadence; affects window overlap and personalization/temporal metrics.
  • Equal Q weights = 1/6 each
    The composite quality index is an equally weighted average of the six property scores; alternative weights are mentioned as sensitivity but not reported.
  • Persistence threshold 0.90 for P3 = 0.90
    Used to define the persistence horizon (first lag with median cosine below 0.90); chosen without formal justification.
assumptions (5)
  • domain assumption A finite-dimensional latent state is a working summary of the windowed history
    Remark 3.2 states that the latent state is a working summary, not a formal sufficient statistic, acknowledging that formal sufficiency would require a fully specified likelihood and target functional.
  • ad hoc to paper The six properties P1-P6 and their operationalizations capture 'usefulness as a reusable table'
    The property system P is introduced by the authors and the scores Sj are empirical tests designed by them; there is no external standard for what makes a reusable table useful. This is acknowledged as a construction rather than a discovered principle.
  • domain assumption Team-by-Season regime is a meaningful recoverable label
    Regime labels are used as the auxiliary training criterion and as the recoverability target for P1. The paper notes these labels mix coaching, calendars, and measurement systems, and within-regime partial correlations mitigate but do not eliminate regime confounding.
  • standard math Rotation non-identifiability can be resolved by Procrustes alignment and geometric summaries
    Remark 3.3 and Section 7 assume that after Procrustes alignment, ensemble means and covariances are meaningful; this is a standard technique in shape analysis and representation learning.
  • ad hoc to paper The classical baseline in the P6 test that excludes the outcome channel is a valid comparator
    In Section 7, the incremental-value comparison uses classical summaries that exclude the target outcome, while the latent table includes the outcome channel. This assumption is questionable because it gives the latent table an inherent advantage through autocorrelation.
invented entities (1)
  • Latent Memory Table (LMT)
    purpose: A reusable statistical object summarizing longitudinal histories, intended to be stored, queried, and reused in downstream analyses without retraining the encoder.
    The LMT is a new abstract object proposed by the paper. Independent evidence would require a falsifiable handle outside the paper, such as a simulation showing the learned table recovers known memory better than classical alternatives. The simulations actually show classical summaries recover the true memory slightly better (Block I Procrustes 0.46 vs 0.39), and the empirical Q advantage is compromised by the unfair P6 baseline, so the entity lacks strong independent support in the paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning Latent Memory States from Longitudinal Athlete Monitoring Data." pith.science (2026). https://pith.science/paper/5MMRAYBO

@misc{pith2026260806290,
  author       = {Pith},
  title        = {Pith review of: Learning Latent Memory States from Longitudinal Athlete Monitoring Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5MMRAYBO}},
  note         = {Machine review of arXiv:2608.06290}
}
abstract

We propose a new unit of analysis for longitudinal data: the Latent Memory Table. The scientific contribution is not the encoder. It is that table, treated as a reusable statistical object on the same footing as a matrix of principal-component scores, a table of estimated random effects, or a table of predicted probabilities. We estimate a statistical table that summarizes recent longitudinal history and is intended to be stored, queried, analysed and reused throughout the statistical workflow. A memory operator maps each masked windowed history to a finite-dimensional state; collecting those states with uncertainty yields the Latent Memory Table. Validation is organized around six properties---recoverability, personalization, temporal coherence, interpretability, stability and reusability---summarized by a composite quality index \(Q\); the Transformer, the SoccerMon case study and the simulations exist to argue that this table deserves that status. Classical exponentially weighted moving averages and related short- and long-horizon scalar summaries arise as restricted, typically univariate special cases of the same operator class. A simulation study with known memory mechanisms shows that \(Q\) and rotation-invariant recovery scores discriminate genuine multivariate or personalized memory from negative controls and from misspecified windows, whereas regime classification accuracy alone does not. SoccerMon serves as an empirical case study: a constructed Latent Memory Table attains \(Q\approx 0.73\) versus about \(0.40\) for classical and lagged principal-component baselines, with incremental held-out value for some wellness targets and Procrustes ensembles for row-wise reliability.

Figures

Figures reproduced from arXiv: 2608.06290 by the authors.

Figure 2
Figure 2. Geometry of the table on held-out athletes: leading principal components (PC1–PC2) [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Three-way personalization geometry. Same-athlete pairs exclude adjacent windows [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Temporal persistence ρbm(h) with 95% player-bootstrap bands and a 0.90 reference line [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figures from the paper (16 more)
Figure 6
Figure 6. Figure 6: 7 Statistical characterization of the table Having checked P1–P4 on geometry, we characterize how variance is shared across regimes and athletes, whether T adds information beyond classical scalars (P6), how tables score under the composite index Q, and how uncertain t…
Figure 5
Figure 5. Figure 5: Latent trajectories in PC2–PC4 for representative athletes from each Team [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]
Figure 6
Figure 6. Figure 6: mgcv GAMM partial smooths for PC2–sleep, PC3–CTL28 and PC4–ACWR (Team×Season fixed effects and player random intercept; bands are 95% pointwise intervals for s(x); x on the raw physiological scale). PC1 PC2 PC3 PC4 PC5 0.0 0.2 0.4 0.6 0.8 1.0 Share of total sum of squa…
Figure 7
Figure 7. Figure 7: Variance decomposition of PC1–PC5 of the Latent Memory Table. [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]
Figure 8
Figure 8. Figure 8: Incremental value of the Latent Memory Table versus classical summaries for next [PITH_FULL_IMAGE:figures/full_fig_p014_8.png]
Figure 9
Figure 9. Figure 9: Six-property quality profiles for learned, classical and PCA tables. [PITH_FULL_IMAGE:figures/full_fig_p015_9.png]
Figure 10
Figure 10. Figure 10: Initialization stability: held-out accuracy across seeds and Procrustes correlation of [PITH_FULL_IMAGE:figures/full_fig_p017_10.png]
Figure 11
Figure 11. Figure 11: Simulation window sensitivity: quality profile versus fitted [PITH_FULL_IMAGE:figures/full_fig_p018_11.png]
Figure 12
Figure 12. Figure 12: Schematic anomaly operator ai(t) with quantile coding of the score’s own distribution [PITH_FULL_IMAGE:figures/full_fig_p019_12.png]
Figure 13
Figure 13. Figure 13: Monthly observation coverage for subjective wellness versus objective GPS and HR by [PITH_FULL_IMAGE:figures/full_fig_p024_13.png]
Figure 14
Figure 14. Figure 14: Functional PCA of PC3 (load-related) trajectories: mean curve, [PITH_FULL_IMAGE:figures/full_fig_p024_14.png]
Figure 15
Figure 15. Figure 15: Row-wise state uncertainty versus observation coverage. [PITH_FULL_IMAGE:figures/full_fig_p025_15.png]
Figure 16
Figure 16. Figure 16: Aligned PC trajectories with ensemble bands ( [PITH_FULL_IMAGE:figures/full_fig_p025_16.png]
Figure 17
Figure 17. Figure 17: Heatmap of tr(Σbi(t)) over weeks for athletes with dense coverage (B = 50 ensemble). E Software details Code and reproducibility notes are at https://github.com/idaejin/latent-memory-states. The intended pipeline is raw histories → windows → operator estimation → T → …
Figure 18
Figure 18. Figure 18: Training curves for the regime-supervised estimator ( [PITH_FULL_IMAGE:figures/full_fig_p027_18.png]
Figure 19
Figure 19. Figure 19: Held-out Team×Season accuracy versus latent dimension d. 27 [PITH_FULL_IMAGE:figures/full_fig_p027_19.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 25 canonical work pages

  1. [1]

    Recurrent neural networks for multivariate time series with missing values

    Che, Z., Purushotham, S., Cho, K., Sontag, D.\ and Liu, Y.\ (2018). Recurrent neural networks for multivariate time series with missing values. Scientific Reports, 8, 6085

    work page 2018

  2. [2]

    W.\ (1991)

    Banister, E. W.\ (1991). Modeling elite athletic performance. In MacDougall, J. D., Wenger, H. A.\ and Green, H. J.\ (eds), Physiological Testing of the High-Performance Athlete, 2nd ed., pp.\ 403--424. Human Kinetics

    work page 1991

  3. [3]

    A., Franklin, J., Gottschall, L., Hrovatin, L

    Foster, C., Florhaug, J. A., Franklin, J., Gottschall, L., Hrovatin, L. A., Parker, S., Doleshal, P.\ and Dodge, C.\ (2001). A new approach to monitoring exercise training. Journal of Strength and Conditioning Research, 15, 109--115

    work page 2001

  4. [4]

    J.\ (2016)

    Gabbett, T. J.\ (2016). The training--injury prevention paradox: should athletes be training smarter and harder? British Journal of Sports Medicine, 50, 273--280

    work page 2016

  5. [5]

    C.\ (1975)

    Gower, J. C.\ (1975). Generalized Procrustes analysis. Psychometrika, 40, 33--51

    work page 1975

  6. [6]

    T., Gabbett, T

    Hulin, B. T., Gabbett, T. J., Lawson, D. W., Caputi, P.\ and Sampson, J. A.\ (2016). The acute:chronic workload ratio predicts injury British Journal of Sports Medicine, 50, 231--236

    work page 2016

  7. [7]

    M., Hawkins, R., Kelly, D

    Lolli, L., Batterham, A. M., Hawkins, R., Kelly, D. M., Strudwick, A. J., Cook, G., Atkinson, G.\ and MacManus, A. M.\ (2019). Mathematical coupling causes spurious correlation in the acute:chronic workload ratio. British Journal of Sports Medicine, 53, 1518--1521

    work page 2019

  8. [8]

    Decoupled weight decay regularization

    Loshchilov, I.\ and Hutter, F.\ (2019). Decoupled weight decay regularization. In 7th International Conference on Learning Representations, New Orleans, 6--9 May 2019

    work page 2019

Show all 26 references
  1. [9]

    B., Gabbett, T

    Murray, N. B., Gabbett, T. J., Townshend, A. D.\ and Blanch, P.\ (2017). Calculating acute:chronic workload ratios using exponentially weighted moving averages provides a more sensitive indicator of injury likelihood than rolling averages. British Journal of Sports Medicine, 5...

    2017

  2. [10]

    Eekhout, I., van Buuren, S., Visser, B., Bink, M. C. A. M.\ and Huisman, A.\ (2023). Longitudinal individual predictions from irregular repeated measurements data. Scientific Reports, 13, 952. https://doi.org/10.1038/s41598-022-26933-1

    2023 doi

  3. [11]

    A.\ and Dixon, W

    Lunt, M., Selby, D. A.\ and Dixon, W. G.\ (2025). Describing variability of intensively collected longitudinal ordinal data with latent spline models. Scientific Reports, 15, 34151. https://doi.org/10.1038/s41598-025-13993-2

    2025 doi

  4. [12]

    O g retir, M., Koskinen, M., Sinisalo, J., Renkonen, R.\ and L \

    \" O g retir, M., Koskinen, M., Sinisalo, J., Renkonen, R.\ and L \"a hdesm \"a ki, H.\ (2025). SeqRisk : Transformer-augmented latent variable model for robust survival prediction with longitudinal data. arXiv:2409.12709

    2025

  5. [13]

    Deep representation learning for clustering longitudinal survival data from electronic health records

    Qiu, J., Hu, Y., Li, L.\ et al.\ (2025). Deep representation learning for clustering longitudinal survival data from electronic health records. Nature Communications, 16, 2534. https://doi.org/10.1038/s41467-025-56625-z

    2025 doi

  6. [14]

    O.\ and Silverman, B

    Ramsay, J. O.\ and Silverman, B. W.\ (2005). Functional Data Analysis, 2nd ed. Springer

    2005

  7. [15]

    Joint Models for Longitudinal and Time-to-Event Data

    Rizopoulos, D.\ (2012). Joint Models for Longitudinal and Time-to-Event Data. CRC Press

    2012

  8. [16]

    M., Fern \'a ndez, J.\ and Medina, D.\ (2018)

    Rossi, A., Pappalardo, L., Cintia, P., Iaia, F. M., Fern \'a ndez, J.\ and Medina, D.\ (2018). Effective injury forecasting in soccer with GPS training data and machine learning. PLOS ONE, 13, e0201264

    2018

  9. [17]

    H.\ (1966)

    Sch \"o nemann, P. H.\ (1966). A generalized solution of the orthogonal Procrustes problem. Psychometrika, 31, 1--10

    1966

  10. [18]

    N.\ and Marlin, B

    Shukla, S. N.\ and Marlin, B. M.\ (2021). Multi-time attention networks for irregularly sampled time series. In ICLR

    2021

  11. [19]

    A., Johansen, D., Riegler, M., Halvorsen, P.\ and Hicks, S.\ (2024)

    Midoglu, C., Kj reng Winther, A., Boeker, M., Pettersen, S. A., Johansen, D., Riegler, M., Halvorsen, P.\ and Hicks, S.\ (2024). A large-scale multivariate soccer athlete health, performance, and position monitoring dataset. Scientific Data, 11, 553. https://doi.org/10.1038/s4...

    2024 doi

  12. [20]

    K.\ (2022)

    Tipirneni, S.\ and Reddy, C. K.\ (2022). Self-supervised transformer for sparse and irregularly sampled multivariate clinical time-series. ACM TKDD, 16, 1--17

    2022

  13. [21]

    C.\ and Bialek, W.\ (2000)

    Tishby, N., Pereira, F. C.\ and Bialek, W.\ (2000). The information bottleneck method. arXiv:physics/0004057

    2000 arXiv

  14. [22]

    A.\ and Davidian, M.\ (2004)

    Tsiatis, A. A.\ and Davidian, M.\ (2004). Joint modeling of longitudinal and time-to-event data: an overview. Statistica Sinica, 14, 809--834

    2004

  15. [23]

    J.\ and Jordan, M

    Wainwright, M. J.\ and Jordan, M. I.\ (2008). Graphical models, exponential families, and variational inference. Foundations and Trends in Machine Learning, 1, 1--305

    2008

  16. [24]

    J.\ (2019)

    Windt, J.\ and Gabbett, T. J.\ (2019). Is it all for naught? What does mathematical coupling really mean for acute:chronic workload ratios? British Journal of Sports Medicine, 53, 988--990

    2019

  17. [25]

    N.\ (2017)

    Wood, S. N.\ (2017). Generalized Additive Models: An Introduction with R , 2nd ed. CRC Press

    2017

  18. [26]

    LaTiM : Longitudinal representation learning in continuous-time models to predict disease progression

    Zeghlache, R., Conze, P.-H., El Habib Daho, M., Li, Y., Le Boit \'e , H., Tadayoni, R., Massin, P., Cochener, B., Rezaei, A., Brahim, I., Quellec, G.\ and Lamard, M.\ (2024). LaTiM : Longitudinal representation learning in continuous-time models to predict disease progression....

    2024 doi

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.