nucleation

LIT-1296: Slow Feature Analysis as Variational Inference Objective

2 min read

Properties6
StatusActive — read 2026-10-20 (NOTE-1015) in full, from arXiv v1; worth reading for an exact identity, not a limit: with a Markov prior p(z_t | z_{t−1}) = N(z_{t−1}, I) (a Gaussian random walk, no decay), an instantaneous encoder q(z_t | x_t) = N(g(x_t), I) and any forward model p_θ(x_t | z_t), the KL half of the ELBO is the unordered slowness loss ½Σ‖g(x_{t−1}) − g(x_t)‖² (plus a constant the paper drops), so the ELBO is a per-point reconstruction term minus a slowness term (Eq. 40), and the reconstruction term takes the place of SFA's unit-variance and decorrelation constraints; and for the linear Gaussian case (Eq. 41) and its necessary condition VᵀV = WᵀĊW (Eq. 42). Read it knowing that the paper does not claim, and does not have, equivalence to SFA: it derives no optimum, so what its objective learns is the record's derivation in THEORY-775 (on whitened input, the span of the slowest SFA directions whose lag-one autocorrelation exceeds ½, rotation unidentified; on raw input, not SFA's subspace and not scale-invariant); that the constant d/2 per step is cancelled by an error in Eq. 35; that Eq. 41 has the sign of the reconstruction error reversed (the appendix's Eq. 54 is right); that Eq. 42 is necessary, not "equivalent" to stationarity as §7 says; that the prior builds in independent, non-decaying latents, so closure of the latent coordinates is assumed, not found; and that its account of Turner and Sahani (2007) is two sentences of related work, which the record has not checked against that paper (LIT-1287, Unavailable).
Version2
Tagsrepresentation-learning, probabilistic-modeling, anthology-candidate
SupportsCLAIM-394 — In a stationary linear Gaussian VAR(1) that is time-reversible (AΠ = ΠAᵀ, Π the stationary covariance), every slow-feature direction is a left eigenvector of A, so any span of slow features is an exactly dynamically independent coarse-graining; off reversibility a slow feature is in general neither independent nor free of memory
Cited as “paper” byNOTE-1015 — Schüler and Wiskott — Slow feature analysis as variational inference
Cited as “source” byTHEORY-775 — With a Gaussian random-walk prior and an instantaneous unit-variance Gaussian encoder, the KL half of the evidence lower bound is the unordered slowness loss plus a constant; with a linear Gaussian encoder and decoder on whitened input, the bound is maximized on the span of the slowest SFA directions whose lag-one autocorrelation exceeds one half, with the basis inside that span unidentified, and on unwhitened input the maximizer is neither SFA's subspace nor invariant to rescaling the input

Graph View

Schüler, M. & Wiskott, L. (2025). Slow Feature Analysis as Variational Inference Objective. arXiv:2506.00580 [cs.LG]. Institute for Neural Computation, Ruhr University Bochum. Preprint, one version.

Read in full in NOTE-1015.

Key takeaways

  • An identity, not a limit. Under the paper’s assumptions (prior N(z_{t−1}, I), encoder N(g(x_t), I) factorized per time step), the KL term of the ELBO equals ½Σ_t‖g(x_{t−1}) − g(x_t)‖² + (T − 1)d/2. The paper writes it without the constant, through a slip in Eq. 35 that is harmless for optimization.
  • The constraints become a reconstruction term. SFA’s unit-variance and decorrelation constraints are replaced by E_q log p_θ(x_t | z_t). The argument for the substitution is informal (§7): the fixed encoder noise forbids shrinking the features, and a bottleneck discourages redundancy.
  • Linear case. With q(z | x) = N(Wᵀx + b, I) and p(x | z) = N(Vz + o, I) the paper gives the objective and the necessary condition VᵀV = WᵀĊW (Ċ the second moment of the step differences). It does not solve for the optimum. THEORY-775 does: on whitened input the optimum spans the slowest SFA directions with lag-one autocorrelation above ½; on raw input it is not SFA.
  • Turner and Sahani, as reported here only. §2 says PSFA derives SFA from a linear Gaussian state-space model with linear emissions, that linear SFA is “the maximum-likelihood estimate of the emission model”, with exact equivalence “only in the limit of zero observation noise and annealing of the internal dynamics towards white noise”, the latter “mainly for scale correction”, and (§4) that its prior is z_t ~ N(Λz_{t−1}, ·) with Λ diagonal and ordered. That is this paper’s summary, not a reading of LIT-1287.

Standing in the record

Filed as the open substitute for LIT-1287, for QUESTION-217’s generative-assumptions item, CLAIM-331 and THEORY-733. It answers the first only partly: the paper’s generative model puts slowness into the prior as independent random walks, so slow latents are closed by assumption, and the paper says nothing on predictive sufficiency. The record’s own derivations for the linear, noise-free PSFA-type model are in NOTE-1015.

Backlinks