pith. sign in

arxiv: 2605.15995 · v1 · pith:RUF4W2GFnew · submitted 2026-05-15 · 💻 cs.LG · cs.AI

Constrained latent state modeling: A unifying perspective on representation learning under competing constraints

Gwenol\'e Quellec This is my paper

Pith reviewed 2026-05-20 21:20 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords constrained latent state modelingrepresentation learninglatent statestrade-offsidentifiabilitytemporal coherenceunifying framework
0
0 comments X

The pith

Latent state models in machine learning suffer from ambiguity because their objectives leave key properties unspecified.

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

Current methods for learning latent states from complex data remain fragmented because their training objectives fail to specify what properties the states must satisfy, allowing multiple valid but structurally different representations to emerge. The paper proposes constrained latent state modeling as a unifying lens that names six core properties—predictive sufficiency, minimality, temporal coherence, observation compatibility, invariance to nuisance factors, and structural constraints—and shows these properties are linked by unavoidable trade-offs. By re-examining major modeling families through this lens, the work demonstrates that each family enforces only a subset of the properties and therefore sits in a distinct region of a shared design space. This reframing treats problems such as non-identifiability not as isolated defects but as direct consequences of underconstrained formulations.

Core claim

Constrained latent state modeling (CLSM) provides a unifying perspective by identifying six core properties that latent states should satisfy—predictive sufficiency, minimality, temporal coherence, observation compatibility, invariance to nuisance factors, and structural constraints—and demonstrating that these properties are intrinsically coupled through fundamental trade-offs, so that existing approaches can be reinterpreted as enforcing different subsets of the constraints and thereby occupying distinct positions in a common design space.

What carries the argument

The CLSM framework, which treats the six listed properties as the basic building blocks whose mutual trade-offs organize the space of possible latent state models.

If this is right

  • Lack of identifiability is a predictable outcome of underconstrained objectives rather than an independent technical obstacle.
  • Different modeling families can be compared directly by mapping which subset of constraints each one enforces.
  • Design decisions in new models become explicit choices about which constraints to prioritize given the task and data.
  • Persistent challenges such as poor generalization or sensitivity to nuisance factors can be diagnosed as violations of specific properties.

Where Pith is reading between the lines

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

  • The framework could be extended to guide automated search over constraint combinations for new applications in multimodal or partially observed systems.
  • Similar trade-off structures may appear in related areas such as causal discovery or disentangled representation learning.
  • Empirical tests on controlled synthetic data could quantify the exact shape of the trade-off surfaces between pairs of properties.

Load-bearing premise

That these six properties form a fundamental and exhaustive set of constraints whose trade-offs fully explain the behavior of existing latent state models without missing important distinctions.

What would settle it

An empirical case in which a single latent state model satisfies all six properties at once with no measurable degradation in any of them, or the identification of a widely used modeling approach whose essential behavior cannot be expressed as any combination of the six properties.

read the original abstract

Learning latent representations from complex data is central to modern machine learning, spanning temporal, multimodal, and partially observed systems. In such settings, representations are better understood as latent states capturing underlying system dynamics, rather than as mere compressed summaries of observations. Yet current approaches remain fragmented, relying on distinct -- and often implicit -- assumptions about what these states should represent. We argue that this fragmentation reflects a more fundamental limitation: latent representations are typically learned from underconstrained objectives that fail to specify the properties that meaningful latent states should satisfy. As a result, multiple representations can satisfy the same objective, leading to ambiguity in their structure and interpretation. While many of the underlying principles have been explored in isolation, their interactions have not been explicitly formalized. In this work, we propose constrained latent state modeling (CLSM) as a unifying perspective. We identify a set of core properties -- predictive sufficiency, minimality, temporal coherence, observation compatibility, invariance to nuisance factors, and structural constraints -- and show that they are intrinsically coupled through fundamental trade-offs. Revisiting major modeling families through this lens, we show that existing approaches can be interpreted as enforcing different subsets of constraints, thereby occupying distinct regions of a common design space. This perspective reframes persistent challenges such as lack of identifiability as consequences of underconstrained formulations, rather than isolated technical limitations. More broadly, CLSM provides a principled framework to make design choices explicit, to analyze trade-offs, and to guide the development of more interpretable, robust, and task-aligned latent state models.

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 paper proposes Constrained Latent State Modeling (CLSM) as a unifying perspective on learning latent states from complex data. It identifies six core properties—predictive sufficiency, minimality, temporal coherence, observation compatibility, invariance to nuisance factors, and structural constraints—and claims these are intrinsically coupled via fundamental trade-offs. Existing modeling families are reinterpreted as occupying distinct regions of a shared design space by enforcing different subsets of the constraints, reframing issues such as non-identifiability as consequences of underconstrained objectives rather than isolated limitations.

Significance. If the asserted couplings could be formalized with explicit relations or bounds, CLSM would offer a useful organizing lens for representation learning. As presented, the contribution is primarily taxonomic: it collects known desiderata and asserts (without derivation) that they trade off in model-independent ways. This may help practitioners make design choices explicit but does not yet supply the predictive or quantitative framework needed to resolve persistent ambiguities in latent-state methods.

major comments (2)
  1. [Abstract] Abstract and introduction: the central claim that the six properties 'are intrinsically coupled through fundamental trade-offs' is asserted at a high level but is not supported by any explicit derivation, inequality, or dynamical-systems argument (e.g., an information-theoretic relation between predictive sufficiency and minimality, or a bound showing how temporal coherence quantitatively relaxes observation compatibility). Without such relations the re-interpretation of existing methods remains a post-hoc taxonomy rather than a framework that predicts new trade-offs.
  2. [Introduction] The manuscript states that underconstrained objectives lead to representational ambiguity, yet provides neither a formal definition of 'underconstrained' in terms of the six properties nor a demonstration that the listed properties are exhaustive or minimal. This leaves the unifying perspective vulnerable to the objection that important distinctions among existing methods are lost when they are forced into the CLSM taxonomy.
minor comments (2)
  1. Notation for the six properties is introduced in the abstract but not consistently referenced with the same symbols or acronyms in later sections, making cross-references difficult to follow.
  2. The paper would benefit from a small table or diagram that explicitly maps each major modeling family (VAEs, RNNs, contrastive methods, etc.) to the subset of CLSM constraints it is claimed to enforce.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed review. We value the recognition of CLSM as an organizing lens for representation learning and the identification of opportunities to strengthen the formal aspects of the framework. We address the major comments below and outline specific revisions.

read point-by-point responses

  1. Referee: [Abstract] Abstract and introduction: the central claim that the six properties 'are intrinsically coupled through fundamental trade-offs' is asserted at a high level but is not supported by any explicit derivation, inequality, or dynamical-systems argument (e.g., an information-theoretic relation between predictive sufficiency and minimality, or a bound showing how temporal coherence quantitatively relaxes observation compatibility). Without such relations the re-interpretation of existing methods remains a post-hoc taxonomy rather than a framework that predicts new trade-offs.

    Authors: We agree that the manuscript currently motivates the couplings through conceptual arguments, literature examples, and reinterpretations of existing methods rather than through new formal derivations or quantitative bounds. This approach was chosen to synthesize perspectives across sub-communities, but we recognize that explicit relations would make the framework more predictive. In the revision we will add a dedicated subsection deriving information-theoretic trade-offs (e.g., a mutual-information bound relating predictive sufficiency to minimality) and a simple dynamical-systems illustration showing how temporal coherence can quantitatively relax observation-compatibility constraints. These additions will be placed after the property definitions and before the method reinterpretations. revision: yes

  2. Referee: [Introduction] The manuscript states that underconstrained objectives lead to representational ambiguity, yet provides neither a formal definition of 'underconstrained' in terms of the six properties nor a demonstration that the listed properties are exhaustive or minimal. This leaves the unifying perspective vulnerable to the objection that important distinctions among existing methods are lost when they are forced into the CLSM taxonomy.

    Authors: We will insert a precise definition: an objective is underconstrained with respect to the CLSM properties when it fails to specify a unique (up to equivalence) distribution over latent states that simultaneously satisfies all six properties. We do not claim the six properties are exhaustive or minimal in an absolute sense; they are the intersection of desiderata most frequently invoked across the cited literature. The revision will explicitly note this scope, list two candidate additional properties (e.g., explicit causality and robustness to covariate shift), and demonstrate that the current taxonomy still distinguishes methods by the precise subset and weighting of constraints each enforces. This preserves the distinctions the referee is concerned about while clarifying the framework's intended coverage. revision: partial

Circularity Check

0 steps flagged

No significant circularity; CLSM is a self-contained conceptual perspective

full rationale

The manuscript advances a unifying perspective by enumerating six properties and asserting their intrinsic couplings and trade-offs, then re-interpreting existing model families as occupying different regions of the resulting design space. No equations or derivations are supplied that reduce a claimed prediction or coupling to a fitted parameter or self-referential definition by construction. The central claims rest on interpretive re-framing rather than on any load-bearing self-citation chain or ansatz smuggled from prior author work. The framework is therefore self-contained as a taxonomy and does not exhibit the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The paper rests on the domain assumption that latent representations should be interpreted as states with specific properties; no free parameters or invented physical entities are introduced.

axioms (1)
  • domain assumption Latent representations are better understood as latent states capturing underlying system dynamics rather than as mere compressed summaries of observations.
    Stated in the opening of the abstract as the starting point for the fragmentation argument.
invented entities (1)
  • Constrained latent state modeling (CLSM) no independent evidence
    purpose: Unifying perspective that makes design choices explicit and analyzes trade-offs among core properties.
    Introduced as the central proposal of the paper; no independent evidence or falsifiable prediction is provided.

pith-pipeline@v0.9.0 · 5804 in / 1200 out tokens · 34878 ms · 2026-05-20T21:20:19.873022+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.

Reference graph

Works this paper leans on

89 extracted references · 89 canonical work pages · 1 internal anchor

  1. [1]

    Rao, R. P. N. & Ballard, D. H. Predictive coding in the visual cortex: A functional interpretation of some extra-classical receptive-field effects. Nat Neurosci 2, 79–87 (1999)

    work page 1999

  2. [2]

    Hafner, D. et al. Xing, E. (ed.) Learning latent dynamics for planning from pixels . (ed.Xing, E.) Proc ICML, ICML’19, 2555–2565 (PMLR, 2019). 25

    work page 2019

  3. [3]

    & Schmidhuber, J

    Ha, D. & Schmidhuber, J. World models. Zenodo 21 pages (2018)

    work page 2018

  4. [4]

    Kingma, D. P. & Welling, M. Bengio, Y. & LeCun, Y. (eds) Auto-encoding variational Bayes . (eds Bengio, Y. & LeCun, Y.) Proc ICLR, ICLR’14, 14 pages (OpenReview.net, 2014)

    work page 2014

  5. [5]

    J., Mohamed, S

    Rezende, D. J., Mohamed, S. & Wierstra, D. McAllester, D. (ed.) Stochastic back- propagation and approximate inference in deep generative models. (ed.McAllester, D.) Proc ICML, ICML’14, II–1278–II–1286 (JMLR, 2014)

    work page 2014

  6. [6]

    Hinton, G. E. & Salakhutdinov, R. R. Reducing the dimensionality of data with neural networks. Science 313, 504–507 (2006)

    work page 2006

  7. [7]

    & Manzagol, P.-A

    Vincent, P., Larochelle, H., Bengio, Y. & Manzagol, P.-A. Cohen, W. (ed.) Extracting and composing robust features with denoising autoencoders. (ed.Cohen, W.) Proc ICML, ICML’08, 1096–1103 (PMLR, 2008)

    work page 2008

  8. [8]

    & Vinyals, O

    van den Oord, A., Li, Y. & Vinyals, O. Representation learning with contrastive predictive coding (2019)

    work page 2019

  9. [9]

    Assran, M. et al. Brown, M. S., Li, F.-F., Mori, G. & Sato, Y. (eds) Self-supervised learning from images with a joint-embedding predictive architecture . (eds Brown, M. S., Li, F.-F., Mori, G. & Sato, Y.) Proc CVPR, CVPR’23 (IEEE, 2023)

    work page 2023

  10. [10]

    Ngiam, J. et al. Ghahramani, Z. (ed.) Multimodal deep learning. (ed.Ghahramani, Z.) Proc ICML, Vol. 28 of ICML’11, 689–696 (PMLR, 2011)

    work page 2011

  11. [11]

    Radford, A. et al. Langford, J. (ed.) Learning transferable visual models from natural language supervision . (ed.Langford, J.) Proc ICML, ICML’21, 8748–8763 (PMLR, 2021)

    work page 2021

  12. [12]

    & Morency, L.-P

    Baltrusaitis, T., Ahuja, C. & Morency, L.-P. Multimodal machine learning: A survey and taxonomy. IEEE Trans Pattern Anal Mach Intell 41, 423–443 (2019)

    work page 2019

  13. [13]

    Uelwer, T. et al. A survey on self-supervised methods for visual representation learning. Mach Learn 114, 111 (2025)

    work page 2025

  14. [14]

    Matta, S. et al. A systematic review of generalization research in medical image classification. Comput Biol Med 183, 109256 (2024)

    work page 2024

  15. [15]

    & Murphy, K

    Alemi, A., Fischer, I., Dillon, J. & Murphy, K. Bengio, Y. & LeCun, Y. (eds) Deep variational information bottleneck . (eds Bengio, Y. & LeCun, Y.) Proc ICLR, ICLR’17, 19 pages (OpenReview.net, 2017)

    work page 2017

  16. [16]

    K., Gelly, S

    Tschannen, M., Djolonga, J., Rubenstein, P. K., Gelly, S. & Lucic, M. Rush, A. (ed.) On mutual information maximization for representation learning . (ed.Rush, A.) Proc ICLR, ICLR’20 (OpenReview.net, 2020). 26

    work page 2020

  17. [17]

    Hénaff, O. Blei, D. (ed.) Data-efficient image recognition with contrastive predictive coding. (ed.Blei, D.) Proc ICML, ICML’20, 4182–4192 (PMLR, 2020)

    work page 2020

  18. [18]

    & Bengio, Y

    Alain, G. & Bengio, Y. What regularized auto-encoders learn from the data- generating distribution. J Mach Learn Res 15, 3743–3773 (2014)

    work page 2014

  19. [19]

    Fonteijn, H. M. et al. An event-based model for disease progression and its application in familial Alzheimer’s disease and Huntington’s disease. Neuroimage 60, 1880–1889 (2012)

    work page 2012

  20. [20]

    Young, A. L. et al. Uncovering the heterogeneity and temporal complexity of neurodegenerative diseases with Subtype and Stage Inference. Nat Commun 9, 4273 (2018)

    work page 2018

  21. [21]

    Locatello, F. et al. Xing, E. (ed.) Challenging common assumptions in the unsupervised learning of disentangled representations . (ed.Xing, E.) Proc ICML, ICML’19, 4114–4124 (PMLR, 2019)

    work page 2019

  22. [22]

    Kalman, R. E. A new approach to linear filtering and prediction problems. J Basic Eng 82, 35–45 (1960)

    work page 1960

  23. [23]

    Kalman, R. E. & Bucy, R. S. New results in linear filtering and prediction theory. J Basic Eng 83, 95–108 (1961)

    work page 1961

  24. [24]

    E., Petrie, T., Soules, G

    Baum, L. E., Petrie, T., Soules, G. & Weiss, N. A maximization technique occurring in the statistical analysis of probabilistic functions of Markov chains. Ann Math Stat 41, 164–171 (1970)

    work page 1970

  25. [25]

    A tutorial on hidden Markov models and selected applications in speech recognition

    Rabiner, L. A tutorial on hidden Markov models and selected applications in speech recognition. Proc IEEE 77, 257–286 (1989)

    work page 1989

  26. [26]

    & van Der Merwe, R

    Wan, E. & van Der Merwe, R. IEEE (ed.) The unscented Kalman filter for nonlinear estimation . (ed.IEEE) Proc IEEE AS-SPCC , 153–158 (IEEE, 2000)

    work page 2000

  27. [27]

    & Xiong, M

    Sun, X., Jin, L. & Xiong, M. Extended Kalman filter for estimation of parameters in nonlinear state-space models of biochemical networks. PLOS ONE 3, e3758 (2008)

    work page 2008

  28. [28]

    S., Maciejowski, J

    Kantas, N., Doucet, A., Singh, S. S., Maciejowski, J. & Chopin, N. On particle methods for parameter estimation in state-space models. Stat Sci 30, 328–351 (2015)

    work page 2015

  29. [29]

    & Hinton, G

    Ghahramani, Z. & Hinton, G. E. Variational learning for switching state-space models. Neural Comput 12, 831–864 (2000)

    work page 2000

  30. [30]

    B., Jordan, M

    Fox, E., Sudderth, E. B., Jordan, M. I. & Willsky, A. S. Bayesian nonparametric inference of switching dynamic linear models. IEEE Trans Sig Proc 59, 1569–1585 (2011). 27

    work page 2011

  31. [31]

    & Purdon, P

    He, M., Das, P., Hotan, G. & Purdon, P. L. Switching state-space modeling of neural signal dynamics. PLoS Comput Biol 19, e1011395 (2023)

    work page 2023

  32. [32]

    Linderman, S. et al. Singh, A. & Zhu, J. (eds) Bayesian learning and inference in recurrent switching linear dynamical systems . (eds Singh, A. & Zhu, J.) Proc AISTATS, Vol. 54, 914–922 (PMLR, 2017)

    work page 2017

  33. [33]

    Feng, S. et al. A review: State estimation based on hybrid models of Kalman filter and neural network. Syst Sci Control Eng 11, 2173682 (2023)

    work page 2023

  34. [34]

    & Choi, W

    Hashempoor, H. & Choi, W. Globerson, A. et al. (eds) Gated inference network: Inference and learning state-space models . (eds Globerson, A. et al. ) Adv Neural Inf Process Syst, Vol. 37 of NIPS’24, 39036–39073 (Curran Associates, Inc., 2024)

    work page 2024

  35. [35]

    A., Thomas, L., Wilcox, C., Ovaskainen, O

    Patterson, T. A., Thomas, L., Wilcox, C., Ovaskainen, O. & Matthiopoulos, J. State-space models of individual animal movement. Trends Ecol Evol 23, 87–94 (2008)

    work page 2008

  36. [36]

    Paninski, L. et al. A new look at state-space models for neural data. J Comput Neurosci 29, 107–126 (2010)

    work page 2010

  37. [37]

    & Zhao, X

    Han, J., Ding, Q., Xiong, A. & Zhao, X. A state-space EMG model for the estimation of continuous joint movements. IEEE Trans Ind Electron 62, 4267– 4275 (2015)

    work page 2015

  38. [38]

    On lines and planes of closest fit to systems of points in space

    Pearson, K. On lines and planes of closest fit to systems of points in space. Lond Edinb Dubl Phil Mag 2, 559–572 (1901)

    work page 1901

  39. [39]

    Analysis of a complex of statistical variables into principal components

    Hotelling, H. Analysis of a complex of statistical variables into principal components. J Educ Psychol 24, 417–441 (1933)

    work page 1933

  40. [40]

    General intelligence, objectively determined and measured

    Spearman, C. General intelligence, objectively determined and measured. Am J Psychol 15, 201–293 (1904)

    work page 1904

  41. [41]

    Lawley, D. N. & Maxwell, A. E. Factor analysis as a statistical method (London, Butterworths, 1971)

    work page 1971

  42. [42]

    J., Knott, M

    Bartholomew, D. J., Knott, M. & Moustaki, I. Latent variable models and factor analysis: A unified approach (Wiley, Chichester, UK, 2011)

    work page 2011

  43. [43]

    Tipping, M. E. & Bishop, C. M. Probabilistic principal component analysis. J R Stat Soc, B 61, 611–622 (1999)

    work page 1999

  44. [44]

    Independent component analysis, a new concept? Signal Process 36, 287–314 (1994)

    Comon, P. Independent component analysis, a new concept? Signal Process 36, 287–314 (1994)

    work page 1994

  45. [45]

    & Oja, E

    Hyvärinen, A. & Oja, E. Independent component analysis: Algorithms and applications. Neural Netw 13, 411–430 (2000). 28

    work page 2000

  46. [46]

    Probabilistic non-linear principal component analysis with Gaus- sian process latent variable models

    Lawrence, N. Probabilistic non-linear principal component analysis with Gaus- sian process latent variable models. J Mach Learn Res 6, 1783–1816 (2005)

    work page 2005

  47. [47]

    & Pajunen, P

    Hyvärinen, A. & Pajunen, P. Nonlinear independent component analysis: Existence and uniqueness results. Neural Netw 12, 429–439 (1999)

    work page 1999

  48. [48]

    & Cunningham, J

    Wang, Y., Blei, D. & Cunningham, J. P. Ranzato, M. & Beygelzimer, A. (eds) Posterior collapse and latent variable non-identifiability . (eds Ranzato, M. & Beygelzimer, A.) Adv Neural Inf Process Syst , Vol. 34 of NIPS’21, 5443–5455 (Curran Associates, Inc., 2021)

    work page 2021

  49. [49]

    E., Hinton, G

    Rumelhart, D. E., Hinton, G. E. & Williams, R. J. Learning representations by back-propagating errors. Nature 323, 533–536 (1986)

    work page 1986

  50. [50]

    & Bengio, Y

    Rifai, S., Vincent, P., Muller, X., Glorot, X. & Bengio, Y. Ghahramani, Z. (ed.) Contractive auto-encoders: Explicit invariance during feature extraction . (ed.Ghahramani, Z.) Proc ICML, ICML’11, 833–840 (PMLR, 2011)

    work page 2011

  51. [51]

    & Goodfellow, I

    Makhzani, A., Shlens, J., Jaitly, N. & Goodfellow, I. Bengio, Y. & LeCun, Y. (eds) Adversarial autoencoders. (eds Bengio, Y. & LeCun, Y.) Proc ICLR Works, ICLR’16, 16 pages (OpenReview.net, 2016)

    work page 2016

  52. [52]

    & kavukcuoglu, k

    van den Oord, A., Vinyals, O. & kavukcuoglu, k. Guyon, I. & von Luxburg, U. (eds) Neural discrete representation learning. (eds Guyon, I. & von Luxburg, U.) Adv Neural Inf Process Syst , Vol. 30 of NIPS’17, 6309–6318 (Curran Associates, Inc., 2017)

    work page 2017

  53. [53]

    K., Raiko, T., Maaløe, L., Sønderby, S

    Sønderby, C. K., Raiko, T., Maaløe, L., Sønderby, S. K. & Winther, O. Lee, D. D. & Sugiyama, M. (eds) Ladder variational autoencoders . (eds Lee, D. D. & Sugiyama, M.) Adv Neural Inf Process Syst , NIPS’16, 3745–3753 (Curran Associates, Inc., 2016)

    work page 2016

  54. [54]

    Chen, X. et al. Context autoencoder for self-supervised representation learning. Int J Comput Vision 132, 208–223 (2023)

    work page 2023

  55. [55]

    Higgins, I. et al. Bengio, Y. & LeCun, Y. (eds) beta-V AE: Learning basic visual concepts with a constrained variational framework . (eds Bengio, Y. & LeCun, Y.) Proc ICLR, ICLR’17, 13 pages (OpenReview.net, 2017)

    work page 2017

  56. [56]

    S., Li, Y

    Berahmand, K., Daneshfar, F., Salehi, E. S., Li, Y. & Xu, Y. Autoencoders and their applications in machine learning: A survey. Artif Intell Rev 57, 28 (2024)

    work page 2024

  57. [57]

    Bardes, A. et al. Revisiting feature prediction for learning visual representations from video (2024)

    work page 2024

  58. [58]

    & Norouzi, M

    Hafner, D., Lillicrap, T., Ba, J. & Norouzi, M. Rush, A. (ed.) Dream to control: Learning behaviors by latent imagination . (ed.Rush, A.) Proc ICLR , ICLR’20 29 (OpenReview.net, 2020)

    work page 2020

  59. [59]

    Poole, B., Ozair, S., Oord, A. V. D., Alemi, A. & Tucker, G. Xing, E. (ed.) On variational bounds of mutual information . (ed.Xing, E.) Proc ICML , ICML’19, 5171–5180 (PMLR, 2019)

    work page 2019

  60. [60]

    & Ahmed, F

    Song, B., Zhou, R. & Ahmed, F. Multi-modal machine learning in engineering design: A review and future directions. J Comput Inf Sci Eng 24, 010801 (2023)

    work page 2023

  61. [61]

    Li, Y. et al. A review of deep learning-based information fusion techniques for multimodal medical image classification. Comput Biol Med 177, 108635 (2024)

    work page 2024

  62. [62]

    & Salakhutdinov, R

    Murphy, K., Schölkopf, B., Srivastava, N. & Salakhutdinov, R. Multimodal learning with deep Boltzmann machines. J Mach Learn Res 15, 2949–2980 (2014)

    work page 2014

  63. [63]

    & Matsuo, Y

    Suzuki, M., Nakayama, K. & Matsuo, Y. Bengio, Y. & LeCun, Y. (eds) Joint multimodal learning with deep generative models . (eds Bengio, Y. & LeCun, Y.) Proc ICLR, ICLR’17, 12 pages (OpenReview.net, 2017)

    work page 2017

  64. [64]

    & Goodman, N

    Wu, M. & Goodman, N. Bengio, S. (ed.) Multimodal generative models for scalable weakly-supervised learning. (ed.Bengio, S.) Adv Neural Inf Process Syst , Vol. 31 of NIPS’18, 5575–5585 (Curran Associates, Inc., 2018)

    work page 2018

  65. [65]

    & Torr, P

    Shi, Y., Siddharth, N., Paige, B. & Torr, P. H. S. Wallach, H. (ed.) Varia- tional mixture-of-experts autoencoders for multi-modal deep generative models . (ed.Wallach, H.) Adv Neural Inf Process Syst , Vol. 32 of NIPS’19, 15718–15729 (Curran Associates, Inc., 2019)

    work page 2019

  66. [66]

    & Livescu, K

    Andrew, G., Arora, R., Bilmes, J. & Livescu, K. Littman, M. (ed.) Deep canonical correlation analysis. (ed.Littman, M.) Proc ICML, ICML’13, III–1247–III–1255 (PMLR, 2013)

    work page 2013

  67. [67]

    Tsai, Y.-H. H. et al. Korhonen, A., Traum, D. & Màrquez, L. (eds) Multimodal transformer for unaligned multimodal language sequences . (eds Korhonen, A., Traum, D. & Màrquez, L.) Proc ACL, Vol. 57, 6558–6569 (ACL, 2019)

    work page 2019

  68. [68]

    Jia, C. et al. Langford, J. (ed.) Scaling up visual and vision-language representa- tion learning with noisy text supervision . (ed.Langford, J.) Proc ICML, Vol. 38 of ICML’21, 4904–4916 (PMLR, 2021)

    work page 2021

  69. [69]

    & Lee, S

    Lu, J., Batra, D., Parikh, D. & Lee, S. Wallach, H. (ed.) ViLBERT: Pretrain- ing task-agnostic visiolinguistic representations for vision-and-language tasks . (ed.Wallach, H.) Adv Neural Inf Process Syst , Vol. 33 of NIPS’19, 13–23 (Curran Associates, Inc., 2019)

    work page 2019

  70. [70]

    Alayrac, J.-B. et al. Koyejo, S. & Mohamed, S. (eds) Flamingo: A visual language model for few-shot learning . (eds Koyejo, S. & Mohamed, S.) Adv Neural Inf 30 Process Syst, Vol. 35 of NIPS’22, 23716–23736 (Curran Associates, Inc., 2022)

    work page 2022

  71. [71]

    Jack, C. R. et al. Tracking pathophysiological processes in Alzheimer’s disease: An updated hypothetical model of dynamic biomarkers. Lancet Neurol 12, 207–216 (2013)

    work page 2013

  72. [72]

    Chan, P. L. & Holford, N. H. Drug treatment effects on disease progression. Annu Rev Pharmacol Toxicol 41, 625–659 (2001)

    work page 2001

  73. [73]

    Dahl, S. G. et al. Incorporating physiological and biochemical mechanisms into pharmacokinetic-pharmacodynamic models: A conceptual framework. Basic Clin Pharmacol Toxicol 106, 2–12 (2010)

    work page 2010

  74. [74]

    Zeghlache, R. et al. Linguraru, M. G. et al. (eds) LaTiM: Longitudinal repre- sentation learning in continuous-time models to predict disease progression . (eds Linguraru, M. G. et al. ) Proc MICCAI, 404–414 (Springer, 2024)

    work page 2024

  75. [75]

    Young, A. L. et al. A data-driven model of biomarker changes in sporadic Alzheimer’s disease. Brain 137, 2564–2577 (2014)

    work page 2014

  76. [76]

    Zhang, X. et al. Bayesian model reveals latent atrophy factors with dissocia- ble cognitive trajectories in Alzheimer’s disease. Proc Natl Acad Sci USA 113, E6535–E6544 (2016)

    work page 2016

  77. [77]

    Donohue, M. C. et al. Estimating long-term multivariate progression from short- term data. Alzheimers Dement 10, S400–S410 (2014)

    work page 2014

  78. [78]

    & Durrleman, S

    Schiratti, J.-B., Allassonnière, S., Colliot, O. & Durrleman, S. Cortes, C. & Lawrence, N. D. (eds) Learning spatiotemporal trajectories from manifold-valued longitudinal data . (eds Cortes, C. & Lawrence, N. D.) Adv Neural Inf Process Syst, Vol. 2 of NIPS’15, 2404–2412 (Curran Associates, Inc., 2015)

    work page 2015

  79. [79]

    Proust-Lima, C., Séne, M., Taylor, J. M. G. & Jacqmin-Gadda, H. Joint latent class models for longitudinal and time-to-event data: A review. Stat Methods Med Res 23, 74–90 (2014)

    work page 2014

  80. [80]

    & Jacqmin-Gadda, H

    Proust-Lima, C., Dartigues, J.-F. & Jacqmin-Gadda, H. Joint modeling of repeated multivariate cognitive measures and competing risks of dementia and death: A latent process and latent class approach. Stat Med 35, 382–398 (2016)

    work page 2016

Showing first 80 references.