Deep Dynamic Probabilistic Canonical Correlation Analysis

Shiqin Tang; Shujian Yu; S. Joe Qin; Yining Dong

REVIEW 4 major objections 6 minor 1 cited by

Deep Dynamic Probabilistic Canonical Correlation Analysis

T0 review · 4 major / 6 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read D2PCCA replaces the linear dynamics of probabilistic CCA with neural networks, and on financial data it reports higher ELBO and lower RMSE than the linear DPCCA.

desk verdict A reasonable deep extension of DPCCA, but the headline comparison against a weakened DPCCA baseline does not support the claimed win, and the flow objective has an apparent error. read the letter →

arxiv 2502.05155 v1 pith:BVWJLKKW submitted 2025-02-07 cs.LG stat.ML

Shiqin Tang , Shujian Yu , Yining Dong , S. Joe Qin This is my paper · ORCID

classification cs.LGstat.ML

keywords probabilisticcanonicalcorrelationanalysisdeepMarkovmodelnonlineardynamicalsystemsvariationalinferencenormalizingflowsKLannealingmultisetCCAfinancialtimeseries

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

The paper is trying to establish that probabilistic canonical correlation analysis, a method for finding shared and view-specific structure in paired data streams, can be made nonlinear and dynamic by replacing its linear Gaussian transitions and emissions with neural networks. The proposed D2PCCA model keeps the three-chain latent structure of DPCCA—one chain common to all views and one private chain per view—and trains it with amortized variational inference rather than the EM algorithm, because nonlinearity makes exact posterior inference intractable. The authors further show that KL annealing and normalizing flows can be added to improve training and posterior flexibility, and that the model extends naturally to more than two observed variables. On daily stock-price data from five sectors, the nonlinear model reports a higher evidence lower bound and lower reconstruction error than its linear predecessor, which is taken as evidence that it captures nonlinear latent dynamics with uncertainty estimates.

What carries the argument

The central machinery is the three-chain latent state $z_t = (z^0_t, z^1_t, z^2_t)$, where $z^0_t$ is shared by all observed views and $z^1_t, z^2_t$ are private to each view. The generative model uses gated transition networks $G_i(z^i_{t-1})$ and variance networks $S_i(z^i_{t-1})$, so the transition is $N(z^i_t | G_i(z^i_{t-1}), diag(S_i(z^i_{t-1})))$, and emission networks map $(z^0_t, z^j_t)$ to the mean and variance of each observation. The inference network is a structured posterior that factorizes as $q(z_{1:T}|x_{1:T}) = \prod_t q(z_t|z_{t-1}, x_{t:T})$, implemented by a backward RNN over the observations whose hidden state is combined with $z_{t-1}$ by a combiner network. This structured posterior is what makes variational training tractable; the extensions add a KL-annealing weight $\beta$ and normalizing-flow transformations $f_\omega(u_t)$ on the latent samples to enrich the posterior.

What would settle it

Re-run the comparison on a benchmark whose true latent dynamics are known, such as a synthetic multi-view system with a nonlinear shared factor; if D2PCCA's ELBO advantage over DPCCA vanishes or it fails to recover the ground-truth latent states, the claim that it captures nonlinear latent dynamics is not supported.

Watch

Extended reading notes

Core claim

The central claim is that D2PCCA, a deep probabilistic extension of DPCCA, captures nonlinear latent dynamics in sequential multi-view data and consistently outperforms its linear counterpart in both ELBO and RMSE on the tested financial dataset. The model parameterizes each transition as a Gaussian whose mean and variance are outputs of a gated neural network, and each emission as a Gaussian whose mean and variance depend on the current common and view-specific latent states; this lets the model adjust its uncertainty dynamically instead of using fixed variances. Because exact posterior inference is intractable, a structured inference network—a backward RNN combined with the previous latent state—approximates the posterior, and training maximizes the ELBO. The extensions, KL annealing and inverse autoregressive flows, raise the ELBO further, though the paper notes the RMSE gain is modest because stock data may have limited nonlinear structure.

Load-bearing premise

The model assumes that the approximate probability distribution learned by its inference network is close enough to the true distribution of the hidden states; if that approximation is poor, the reported ELBO and latent states no longer reflect how well the model actually fits the data.

Editorial extensions

If this is right

If D2PCCA's claim holds, nonlinear dynamic CCA becomes a drop-in upgrade over linear DPCCA for sequential multi-view data, with better ELBO and comparable or better reconstruction.
The multiset version can track a common latent factor across many observed variables simultaneously, making it applicable to problems like monitoring several sectors or sensors that share one underlying signal.
The probabilistic output includes dynamic variance estimates, so the model gives confidence intervals around its reconstructions instead of only point predictions.
Because the emission network can be exchanged for different output distributions, the model extends to continuous or discrete observations without changing the latent dynamics machinery.

Reading between the lines

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

A testable extension the paper does not run: on datasets with stronger nonlinear dynamics, such as speech or video, the ELBO gap between D2PCCA and DPCCA should widen; the paper's own stock-data results hint that simpler patterns understate the model's advantage.
The shared-latent structure could be repurposed as an interpretable decomposition of common versus view-specific variability, but the paper does not analyze whether the learned $z^0_t$ actually corresponds to a recognizable market factor.
The posterior approximation quality is unexamined; comparing the structured RNN posterior against a more expressive alternative, or checking calibration of the reported confidence intervals, would indicate how much of the ELBO gain is real model fit rather than inference-network flexibility.

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.

Desk Editor's Note

read the letter

Quick take: the paper does one thing—it takes the DPCCA latent structure (shared plus view-specific chains) and pipes it through Deep Markov Model machinery: gated transitions, structured ST-R inference, KL annealing, normalizing flows. That is a legitimate, if incremental, piece of model plumbing, and the authors are honest that it is a combination of existing pieces. The core ELBO derivation for the basic model is standard. Code is released. So there is a real artifact here, not just a sketch.

The soft spots are mostly empirical. The main claim is that D2PCCA "consistently outperforms" DPCCA on ELBO and RMSE. The ELBO gap (131 vs 69.77) looks large, but the DPCCA baseline is not the model in Section II. Section V says the E-step was replaced with an inference network. For this linear-Gaussian model the exact posterior is available via RTS smoothing; an amortized variational approximation can only make the bound worse. So the comparison is between a deep model's variational bound and a deliberately weakened linear bound. That undermines the headline. The RMSE difference (0.0179 vs 0.0181) is roughly a percent, with no repeated runs or error bars. The paper itself concedes the stock data may be too simple.

Also, Eq. 20 for the normalizing flow ELBO looks wrong as written—the per-timestep log-det terms and base distribution ratios don't line up with the change of variables. The multiset extension is described mostly in words; the generative and inference equations for D≥2 views are not written down. These are the two things I'd want fixed or clarified before trusting the details.

On the positive side: the authors clearly state the limitation of the naive spherical-Gaussian deep PCCA, they are upfront that RMSE gains are not significant, and the paper is readable. Not a fatal set of flaws; the architecture likely works as intended. But the empirical evidence as reported is not strong enough to establish the main claim. A serious referee would ask for a proper DPCCA baseline (RTS-smoother EM), multiple runs with uncertainty, a DMM baseline, corrections to the flow objective, and explicit multiset equations.

Who is this for? People building multi-view sequence models in industrial or process settings, or someone who wants a template for adding structured latent chains to DMMs. It deserves peer review—conditional accept, not desk reject. I'd bring it to a reading group as an example of how quickly variational comparisons can be biased by the baseline.

Referee Report

4 major / 6 minor

Summary. This paper proposes Deep Dynamic Probabilistic Canonical Correlation Analysis (D2PCCA), a nonlinear extension of the linear dynamic probabilistic CCA (DPCCA) model. The generative model replaces the linear-Gaussian transition and emission functions with neural networks, and the posterior is approximated using a structured inference network with a backward RNN, trained by maximizing the ELBO. The paper further describes KL annealing, a normalizing-flow posterior, and a multiset extension, and reports experiments on NASDAQ stock data comparing D2PCCA variants against a DPCCA baseline in terms of test ELBO and reconstruction RMSE.

Significance. If validated, D2PCCA would be a useful addition to the family of deep sequential latent-variable models, offering a principled probabilistic treatment of shared and view-specific dynamics in multi-view time series. The paper builds on established components (deep Markov models, structured inference networks, normalizing flows) and makes code available, which strengthens reproducibility. The two-view ELBO derivation for the basic model is standard and largely correct, and the paper honestly notes that the stock dataset may not contain enough nonlinear structure for the model to show large gains. However, the headline comparison against DPCCA is compromised by a modified baseline and by an apparently incorrect expression for the normalizing-flow objective, so the empirical claims are not yet secure.

major comments (4)

[Section V, Table I] The baseline labeled 'DPCCA' is not the DPCCA model defined in Section II. The text states that the E-step in (6) is replaced by an inference network similar to that used in D2PCCA. For the linear-Gaussian model in (1)-(2), the exact posterior is available from the RTS smoother, and the EM objective in (5) is then exact. Replacing the E-step with an amortized network cannot improve the ELBO; it can only make it lower than the exact ELBO. The reported test ELBO of 69.77 is therefore not the ELBO of the actual DPCCA model, and the comparison in Table I is biased in favor of D2PCCA. Please re-run the DPCCA baseline using the exact RTS smoother or report both the exact and amortized DPCCA ELBOs so the comparison is fair.
[Section IV, Eq. (20)] The normalizing-flow objective is not written correctly. The displayed expression contains a ratio with a product of two base densities, q_phi(u_{t-1}|...) and q_phi(u_t|...), followed by an additional '- log q_phi(u_t|...)' term. Under the standard change of variables, the ELBO should contain one base density q_phi(u_t|...) and one log-determinant of the Jacobian, with the prior and likelihood terms evaluated at z_t = f_omega(u_t). As written, Eq. (20) appears to count the base density twice and does not clearly define the flow ELBO used for the IAF variant in Table I. Please correct the equation or provide the exact objective used in the implementation.
[Section IV, Structural Extensions; Section V] The multiset extension is asserted but not specified. The paper states that D2PCCA extends easily to D observed variables by letting i in {0,...,D} and j in {1,...,D} in (1), but (1) is the linear DPCCA transition/emission. The deep generative equations (7)-(9), the structured posterior factorization (14), and the inference network (15)-(16) are all written for two views only. The experiments use D=5 sectors, but it is not described how the shared latent z^0 interacts with each view-specific z^j in the emission network, nor how the backward RNN and combiner are generalized to D>2. Please provide the multiset formulation, or clarify that the experiments are run as pairwise or otherwise reduced comparisons.
[Section V, Table I] The empirical claim that D2PCCA 'consistently outperforms' DPCCA is based on a single run without standard deviations, confidence intervals, or significance tests. The RMSE differences (0.0179 vs 0.0181, and 0.0183/0.0184 for the IAF variants) are on the order of 1%, and the test set consists of 20 sequences of length 30. Moreover, the IAF variants achieve higher ELBO but worse RMSE than D2PCCA+KL, which the authors partially attribute to the limited nonlinear structure in stock data. Please report multiple random seeds and appropriate measures of variability before drawing the comparative conclusion.

minor comments (6)

[Abstract and Introduction] The phrase 'nonlinear dynamical system analysis' is awkward and appears to be a typo for 'nonlinear dynamical systems.'
[Section IV] The sentence 'In this section, we details three extensions' contains a subject-verb agreement typo: 'details' should be 'detail.'
[Section IV, Structural Extensions] The word 'mutliple' should be 'multiple,' and 'the the dynamics' should be 'the dynamics.'
[Section V] The phrase 'each with70 dimensions' is missing a space: it should be 'each with 70 dimensions.'
[Section V] The ClippedAdam optimizer is cited as reference [20], but [20] is the original Adam paper; please provide a citation for the ClippedAdam variant or clarify that it is a Pyro implementation detail.
[Section II, Eq. (3)] The block matrix for \hat V is displayed as a stacked column of V0, V1, V2, but it should be a block-diagonal matrix with V0, V1, V2 along the diagonal; please fix the notation to match the intended block structure.

Circularity Check

0 steps flagged · score 1.0 of 10

No meaningful circularity: D2PCCA's derivation is a standard deep Markov model variational bound, and the held-out ELBO/RMSE comparisons are not forced by construction; the main caveat is an experimental fairness issue, not circularity.

full rationale

The derivation of D2PCCA is self-contained relative to its stated inputs. Equations (7)-(9) define a nonlinear generative model with gated transitions and neural-network emissions; Equations (14)-(16) define a structured variational posterior of the ST-R form introduced in [8]; and Equation (17) is the standard ELBO used to train the model. The paper does not define one claimed quantity in terms of the quantity it claims to predict. The reported test ELBO is computed on a held-out 50-point test window (Section V: 'we use the first 453 data points for training and the remaining 50 for testing'), so optimizing the training ELBO does not by construction force the test ELBO comparison. Similarly, the RMSE in Equation (21) is evaluated on held-out test sequences and is a reconstruction metric, not a re-used training target. The only self-citation, [6] (Dong and Qin), appears in the introduction as a pointer to DiPLS and is not load-bearing for the model derivation or the empirical claims. The one legitimate concern is experimental fairness, not circularity: Section V states that the multiset DPCCA baseline is 'implemented according to (1), with the E-step in (6) replaced by an inference network, similar to that used in D2PCCA, to achieve faster convergence,' even though the linear-Gaussian DPCCA of Section II admits an exact RTS smoother. This weakens the baseline and biases the headline claim that D2PCCA 'consistently outperforms' DPCCA, but it does not make the D2PCCA derivation circular. No circular step meets the standard of exhibiting an equation that reduces to its own input.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central claim rests on the assumed CCA-style latent factorization and on the adequacy of the structured variational approximation. Many neural network weights and hyperparameters are free. No new physical entities are introduced.

free parameters (5)

Latent state dimensions (shared and specific) = shared: 1, specific: 2 per view
Chosen by hand; not fitted. Determines model capacity and the dimensionality of the extracted dynamics.
KL annealing schedule = beta from 0.01 to 1 over 100 epochs
Chosen by hand; affects training dynamics and final ELBO.
Number and dimension of normalizing flows = 5 affine autoregressive flows, 70 dimensions
Chosen by hand; affects posterior flexibility and test ELBO.
Sliding window length = T=30
Chosen by hand; determines the temporal context for inference.
Optimizer hyperparameters = lr=0.0003, Adam betas=(0.96,0.999), grad clip=10, weight decay=2.0
Chosen by hand; standard for the ClippedAdam optimizer.

assumptions (5)

domain assumption First-order Markov independence structure of the three latent chains
This is the generative structure inherited from DPCCA (Eq. 1) and used in the D2PCCA transition model (Eq. 7). If the true system has cross-coupling between chains, the model cannot represent it.
domain assumption ST-R structured inference network approximates the true posterior
The paper relies on the DMM inference network from [8] without proving approximation quality. If this variational family is too restrictive, the learned latent dynamics and ELBO are unreliable.
domain assumption Gaussian observation likelihood with diagonal state-dependent variance
This is a modeling choice; stock returns may have heavier tails or correlations not captured.
standard math Reparameterization trick yields unbiased ELBO gradients
Implicitly used in training; standard for VAE-style models.
ad hoc to paper Modified DPCCA baseline with inference network remains a fair comparison
The paper replaces the exact EM E-step for DPCCA with an inference network, which may understate DPCCA performance relative to the original algorithm.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Deep Dynamic Probabilistic Canonical Correlation Analysis." pith.science (2026). https://pith.science/paper/BVWJLKKW

@misc{pith2026250205155,
  author       = {Pith},
  title        = {Pith review of: Deep Dynamic Probabilistic Canonical Correlation Analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BVWJLKKW}},
  note         = {Machine review of arXiv:2502.05155}
}

read the original abstract

This paper presents Deep Dynamic Probabilistic Canonical Correlation Analysis (D2PCCA), a model that integrates deep learning with probabilistic modeling to analyze nonlinear dynamical systems. Building on the probabilistic extensions of Canonical Correlation Analysis (CCA), D2PCCA captures nonlinear latent dynamics and supports enhancements such as KL annealing for improved convergence and normalizing flows for a more flexible posterior approximation. D2PCCA naturally extends to multiple observed variables, making it a versatile tool for encoding prior knowledge about sequential datasets and providing a probabilistic understanding of the system's dynamics. Experimental validation on real financial datasets demonstrates the effectiveness of D2PCCA and its extensions in capturing latent dynamics.

Figures

Figures reproduced from arXiv: 2502.05155 by the authors.

**Figure 1.** Graphical models for DPCCA and D2 PCCA. The shaded nodes denote observed variables, while the unshaded ones denote latent variables. The arrows represent transition and emission models, and arrows with solid squares denote the usage of neural networks. underlying both observations, and let z 1 t and z 2 t be unique to x 1 t and x 2 t respectively. The transition and emission models of DPCCA are given by: p(z i t |z … view at source ↗

**Figure 2.** Graphical representations for (a) Multiset DPCCA, (b) DP [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Convergence of ELBO during (left) training and (right) testing. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: Comparison of model predictions against the ground truth over [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

InfoDPCCA: Information-Theoretic Dynamic Probabilistic Canonical Correlation Analysis
cs.LG 2025-06 conditional novelty 6.0 of 10

InfoDPCCA combines a dynamic probabilistic CCA model with an information-bottleneck objective so the shared latent state is trained to contain only the mutual information of the two sequences and still predict the nex...

Reference graph

Works this paper leans on

24 extracted references · 17 canonical work pages · cited by 1 Pith paper

[1]

A probabilistic interpretation of canonical correlation analysis,

F. R. Bach and M. I. Jordan, “A probabilistic interpretation of canonical correlation analysis,” 2005

work page 2005
[2]

Probabilistic approach to detecting dependencies between data sets,

A. Klami and S. Kaski, “Probabilistic approach to detecting dependencies between data sets,” Neurocomputing, vol. 72, no. 1-3, pp. 39–46, 2008

work page 2008
[3]

Probabilistic partial least squares model: Identifiability, estimation and application,

S. el Bouhaddani, H.-W. Uh, C. Hayward, G. Jongbloed, and J. Houwing-Duistermaat, “Probabilistic partial least squares model: Identifiability, estimation and application,” Journal of Multivariate Analysis, vol. 167, pp. 331–346, 2018

work page 2018
[4]

K. P. Murphy, Machine learning: a probabilistic perspective . MIT press, 2012

2012
[5]

Bayesian sparse partial least squares,

D. Vidaurre, M. A. van Gerven, C. Bielza, P. Larranaga, and T. Heskes, “Bayesian sparse partial least squares,” Neural computation, vol. 25, no. 12, pp. 3318–3339, 2013

work page 2013
[6]

Dynamic latent variable analytics for process operations and control,

Y . Dong and S. J. Qin, “Dynamic latent variable analytics for process operations and control,” Computers & Chemical Engineering, vol. 114, pp. 69–80, 2018

work page 2018
[7]

Dynamic prob- abilistic cca for analysis of affective behavior and fusion of continuous annotations,

M. A. Nicolaou, V . Pavlovic, and M. Pantic, “Dynamic prob- abilistic cca for analysis of affective behavior and fusion of continuous annotations,” IEEE transactions on pattern analysis and machine intelligence , vol. 36, no. 7, pp. 1299–1311, 2014

work page 2014
[8]

Structured inference networks for nonlinear state space models,

R. Krishnan, U. Shalit, and D. Sontag, “Structured inference networks for nonlinear state space models,” in Proceedings of the AAAI Conference on Artificial Intelligence , vol. 31, no. 1, 2017

work page 2017

Show all 24 references

[9]

A view of the em algorithm that justifies incremental, sparse, and other variants,

R. M. Neal and G. E. Hinton, “A view of the em algorithm that justifies incremental, sparse, and other variants,” in Learning in graphical models. Springer, 1998, pp. 355–368

1998
[10]

Auto-encoding variational bayes,

D. P. Kingma and M. Welling, “Auto-encoding variational bayes,” arXiv preprint arXiv:1312.6114 , 2013

2013 arXiv
[11]

Variational inference: A review for statisticians,

D. M. Blei, A. Kucukelbir, and J. D. McAuliffe, “Variational inference: A review for statisticians,” Journal of the American statistical Association, vol. 112, no. 518, pp. 859–877, 2017

2017
[12]

A unifying review of linear gaussian models,

S. Roweis and Z. Ghahramani, “A unifying review of linear gaussian models,” Neural computation, vol. 11, no. 2, pp. 305– 345, 1999

1999
[13]

Auto-encoding variational bayes,

D. P. Kingma and M. Welling, “Auto-encoding variational bayes,” in Proceedings of the International Conference on Learning Representations (ICLR) , 2014

2014
[14]

Cyclical annealing schedule: A simple approach to mitigating kl vanishing,

H. Fu, C. Li, X. Liu, J. Gao, A. Celikyilmaz, and L. Carin, “Cyclical annealing schedule: A simple approach to mitigating kl vanishing,” arXiv preprint arXiv:1903.10145 , 2019

1903 arXiv
[15]

Variational inference with normalizing flows,

D. Rezende and S. Mohamed, “Variational inference with normalizing flows,” in International conference on machine learning. PMLR, 2015, pp. 1530–1538

2015
[16]

Improved variational inference with inverse autoregressive flow,

D. P. Kingma, T. Salimans, R. Jozefowicz, X. Chen, I. Sutskever, and M. Welling, “Improved variational inference with inverse autoregressive flow,”Advances in neural information processing systems, vol. 29, 2016

2016
[17]

Normalizing kalman filters for multivariate time series analysis,

E. de B ´ezenac, S. S. Rangapuram, K. Benidis, M. Bohlke- Schneider, R. Kurle, L. Stella, H. Hasson, P. Gallinari, and T. Januschowski, “Normalizing kalman filters for multivariate time series analysis,” Advances in Neural Information Process- ing Systems, vol. 33, pp. 2995–3007, 2020

2020
[18]

Improving sequential latent variable models with autoregressive flows,

J. Marino, L. Chen, J. He, and S. Mandt, “Improving sequential latent variable models with autoregressive flows,” in Symposium on advances in approximate bayesian inference . PMLR, 2020, pp. 1–16

2020
[19]

Factorial hidden markov mod- els,

Z. Ghahramani and M. Jordan, “Factorial hidden markov mod- els,” Advances in neural information processing systems , vol. 8, 1995

1995
[20]

Adam: A method for stochastic optimization,

D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” arXiv preprint arXiv:1412.6980 , 2014

2014 arXiv
[21]

Pyro: Deep universal probabilistic programming,

E. Bingham, J. P. Chen, M. Jankowiak, F. Obermeyer, N. Prad- han, T. Karaletsos, R. Singh, P. Szerlip, P. Horsfall, and N. D. Goodman, “Pyro: Deep universal probabilistic programming,” Journal of machine learning research , vol. 20, no. 28, pp. 1–6, 2019

2019
[22]

S&P 500 Stock Data Dataset,

Kaggle, “S&P 500 Stock Data Dataset,” https://www.kaggle. com/datasets/camnugent/sandp500, 2024, accessed: 2024-03-17

2024
[23]

A recurrent latent variable model for sequen- tial data,

J. Chung, K. Kastner, L. Dinh, K. Goel, A. C. Courville, and Y . Bengio, “A recurrent latent variable model for sequen- tial data,” Advances in neural information processing systems , vol. 28, 2015

2015
[24]

Dynamical variational autoencoders: A comprehensive review,

L. Girin, S. Leglaive, X. Bie, J. Diard, T. Hueber, and X. Alameda-Pineda, “Dynamical variational autoencoders: A comprehensive review,”arXiv preprint arXiv:2008.12595, 2020

2008 arXiv

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.

[1] [1]

A probabilistic interpretation of canonical correlation analysis,

F. R. Bach and M. I. Jordan, “A probabilistic interpretation of canonical correlation analysis,” 2005

work page 2005

[2] [2]

Probabilistic approach to detecting dependencies between data sets,

A. Klami and S. Kaski, “Probabilistic approach to detecting dependencies between data sets,” Neurocomputing, vol. 72, no. 1-3, pp. 39–46, 2008

work page 2008

[3] [3]

Probabilistic partial least squares model: Identifiability, estimation and application,

S. el Bouhaddani, H.-W. Uh, C. Hayward, G. Jongbloed, and J. Houwing-Duistermaat, “Probabilistic partial least squares model: Identifiability, estimation and application,” Journal of Multivariate Analysis, vol. 167, pp. 331–346, 2018

work page 2018

[4] [4]

K. P. Murphy, Machine learning: a probabilistic perspective . MIT press, 2012

2012

[5] [5]

Bayesian sparse partial least squares,

D. Vidaurre, M. A. van Gerven, C. Bielza, P. Larranaga, and T. Heskes, “Bayesian sparse partial least squares,” Neural computation, vol. 25, no. 12, pp. 3318–3339, 2013

work page 2013

[6] [6]

Dynamic latent variable analytics for process operations and control,

Y . Dong and S. J. Qin, “Dynamic latent variable analytics for process operations and control,” Computers & Chemical Engineering, vol. 114, pp. 69–80, 2018

work page 2018

[7] [7]

Dynamic prob- abilistic cca for analysis of affective behavior and fusion of continuous annotations,

M. A. Nicolaou, V . Pavlovic, and M. Pantic, “Dynamic prob- abilistic cca for analysis of affective behavior and fusion of continuous annotations,” IEEE transactions on pattern analysis and machine intelligence , vol. 36, no. 7, pp. 1299–1311, 2014

work page 2014

[8] [8]

Structured inference networks for nonlinear state space models,

R. Krishnan, U. Shalit, and D. Sontag, “Structured inference networks for nonlinear state space models,” in Proceedings of the AAAI Conference on Artificial Intelligence , vol. 31, no. 1, 2017

work page 2017