Factorial hidden Markov model
Model Families AIMech:FactorialHiddenMarkovModel DRAFT
Aliases: FHMM, Factorial HMM
A directed latent sequence-model family in which several independently evolving hidden Markov chains jointly determine the observation distribution at each time.
Identity Notes & Concept Scope
This record covers the independent-chain factorial HMM developed by Ghahramani and Jordan, with a shared observation distribution. It is not a mixture selecting one HMM, a globally normalized product of complete HMM likelihoods, an input-conditioned HMM, or the variational algorithm used to fit the model. Each combination of chain states is one joint state; factorial refers to its distributed component representation. The later coupled-chain and input-conditioned extensions in Section 5 are outside this narrow definition. No normalized identity match was found across all 442 pinned AIO labels/scoped synonyms and 376 preceding native IDs/names/aliases. Full-checkout fixed-string searches included ignored and hidden files in worktree and main, returning 5 and 5 lines. All 4 unique matched contents were fully read: preceding boundary discussions, citations, reviews and pending leads, not existing records of these three identities. All 64 semantic label/synonym hits, 17 exact fully annotated AIO nodes with outgoing edges, their generated records and four native neighbors were inspected or verified unchanged after prior full reading. Absence is bounded to this pin and inspected neighborhood.
Claims & Grounded Evidence
1 Claim
Mechanism
#method
With independent initial states and chain-specific transitions, the joint distribution factors into the product of chain priors, the product over times and chains of p(s_t^m|s_(t-1)^m), and one shared p(y_t|s_t^1,...,s_t^M) per time. The paper’s worked emission is Gaussian with mean sum_m W_m s_t^m for one-hot states and common covariance C. Although chains are independent a priori, observations generally couple their posterior paths through the shared emission. For M chains each having K states, the joint latent state has K^M configurations. Exact forward–backward can exploit the transition factorization by contracting one chain dimension at a time, with O(T M K^(M+1)) transition work under the paper’s fixed-observation-dimension setting; this remains exponential in M. EM uses posterior state, temporal-pair and cross-chain moments. In the additive Gaussian model its weight update is a linear-system solution with a pseudoinverse because chain contributions have redundant mean offsets. Exact inference, Gibbs sampling and variational approximations are separate procedures, not defining alternatives for model identity.
⚠️ Curated Limitations & Negative Results
- Independent latent dynamics do not imply posterior independence after shared observations. Special emission structures may decouple chains; for the Gaussian case a sufficient condition is cross-chain orthogonality in the C^-1 metric. Note 3’s ordinary orthogonality with a merely diagonal covariance is not sufficient in general.
- The Gaussian additive mean is a worked parameterization, not the only possible factorial-HMM emission. Its closed-form parameter updates do not automatically transfer to nonlinear or discrete emissions.
- Factorized transitions reduce exact computation relative to a naive unconstrained joint-state HMM but do not remove exponential state growth. A compact parameterization is not a universal scalability or accuracy guarantee.
- Approximate E-steps may improve a variational bound without monotonically improving the true likelihood. Additive offsets are not individually identifiable without conventions; the paper uses a pseudoinverse.
- Section 5’s coupled-transition and input-conditioned extension displays contain indexing/factorization ambiguities. They are not adopted as the defining joint distribution here. No experimental performance, author implementation or general convergence guarantee is certified.
Artificial Intelligence Ontology (AIO) Grounding
Mapped AIO Classes
| AIO Target Class | Mapping Relation | Mapping Rationale |
|---|---|---|
AIO:BayesianNetwork
Bayesian Network
|
broad_match | For a finite sequence, the independent-chain model with shared observation children is a directed acyclic probabilistic network. The pinned BayesianNetwork definition has that broader scope and is parented under Network; the local factorial temporal structure is more specific. This is a local mapping, not an upstream axiom. |
AIO:MarkovChain
Markov Chain
|
related_match | The model contains multiple latent Markov chains; shared emissions and marginalization do not make its observations a first-order Markov chain. This is a component association. |
Evidence Sources & Primary Literature
1 Source| Source ID | Publication / Document | Type / Access | Reference & Checksum |
|---|---|---|---|
method-source
|
Zoubin Ghahramani, Michael I. Jordan
|
primary paper
full_text
|
ID: doi:10.1023/A:1007425814087
sha256:ab3b4f8f4c8d…
|
Curation Provenance
Curator: Codex
Method: ai_assisted
Model: GPT-6
Created: 2026-10-11T19:44:52Z
Updated: 2026-10-11T19:44:52Z
Source File:
data/curated/FactorialHiddenMarkovModel.yaml