Dynamic Bayesian network

A Bayesian network template replicated across time slices to represent probabilistic dependencies within and between successive states of a stochastic process.

Version

v2 · 2026-09-19 · History

Domain-specific #

4286

Domain group

Applied Sciences & Engineering

Origin domain

Computer Science & Software Engineering

Subdomain

Probabilistic Graphical Models → Computer Science & Software Engineering

Aliases

DBN, Two-timeslice Bayesian network

Core Idea

A first-order two-slice template is common but not mandatory, stationarity and conditional-independence assumptions must be declared and unrolling creates a larger ordinary Bayesian network. Variables are indexed by time, a directed acyclic intra-slice graph and forward inter-slice edges factor the joint distribution, and filtering, smoothing or prediction propagates beliefs through the unrolled model. The abstraction is therefore identified by a declared carrier, a transformation or constraint over that carrier, and an invariant that tells an analyst whether the named structure is genuinely present.

How would you explain it like I'm…

Day-by-Day Guessing Map

Imagine a picture with arrows showing which things affect which, like "clouds" pointing to "rain." Now make a copy of that picture for every day, and add arrows from yesterday to today, but never from today back to yesterday. That chain of pictures helps you guess what is happening now, and what might happen next, from the clues you've seen.

Time-Slice Clue Network

A dynamic Bayesian network is a way to make smart guesses about things that change over time. You draw the things you care about, like 'cloudy' and 'rain', as dots, and arrows show what affects what. You copy this drawing for each moment in time, and add arrows that only go forward from earlier moments to later ones. When you learn new clues, like seeing someone with an umbrella, you can update your guesses about now, the past, or the future. Copying the drawing for many moments makes one big network.

Time-Sliced Probability Network

A Dynamic Bayesian network (DBN) models how random variables evolve over time. Each time slice has its own variables, linked inside the slice by a directed graph with no cycles. Further edges point only forward, from earlier slices to later ones. Each variable has a conditional probability table given its parents, and together these define the joint probability of everything. A common shortcut is a two-slice template: describe the first slice, then one repeating step. This assumes the rules don't change over time (stationarity) and that only the previous step matters (first-order Markov). Neither assumption is required, but whichever you use must be stated. Unrolling the template over many steps gives one big ordinary Bayesian network. With it you can do filtering (now), smoothing (the past), or prediction (the future).

 

A dynamic Bayesian network extends a Bayesian network to sequences by indexing variables by time. Within each slice the variables form a directed acyclic graph. Inter-slice edges go only forward in time, and the local conditional probability distributions factor the joint distribution over the whole trajectory. A typical specification is a first-order two-slice template: an initial-slice distribution plus a transition network. That form is common but not required, and higher Markov orders are allowed. Stationarity, meaning the same transition model at every step, and the conditional-independence assumptions must be stated explicitly. Unrolling the template over a horizon yields a larger ordinary Bayesian network. On that network, given an evidence sequence, inference performs filtering, smoothing, or prediction, and parameters or structure can be learned. Hidden Markov models and many state-space models are special cases of this framework.

Scope of Application

Dynamic Bayesian network belongs to probabilistic graphical models and is useful where the analyst can specify the typed probabilistic graphical models carrier, including objects, relations, parameters, conventions, evidence, boundaries, and comparison targets, then evaluate the time-indexed variables and state space, initial-slice distribution, intra-slice DAG, inter-slice parent links and temporal order, conditional probability distributions, stationarity and Markov-order assumptions, unrolling horizon, evidence sequence and filtering smoothing prediction or learning task are explicit. The scope is broad within that domain but bounded by the need for the time-indexed variables and state space, initial-slice distribution, intra-slice DAG, inter-slice parent links and temporal order, conditional probability distributions, stationarity and Markov-order assumptions, unrolling horizon, evidence sequence and filtering smoothing prediction or learning task are explicit.

Clarity

The abstraction clarifies a crowded vocabulary by making the time-indexed variables and state space, initial-slice distribution, intra-slice DAG, inter-slice parent links and temporal order, conditional probability distributions, stationarity and Markov-order assumptions, unrolling horizon, evidence sequence and filtering smoothing prediction or learning task are explicit the center of the account. A claim should name the carrier, the governing operation or relation, the applicable assumptions, and the recognition test.

Manages Complexity

Without the abstraction, an analyst must reason directly over many local details: the carrier roles, admissibility assumptions, competing conventions, derived invariants, boundary cases, and proof or validation obligations specific to Dynamic Bayesian network. Dynamic Bayesian network compresses them into the roles in the structural signature. That compression permits comparison across instances without erasing the variables that determine validity. It also exposes which details may be varied safely and which are constitutive.

Abstract Reasoning

1. Identify the carrier. State what the elements, states, objects, or observations are: the typed probabilistic graphical models carrier, including objects, relations, parameters, conventions, evidence, boundaries, and comparison targets. Reject examples whose alleged carrier belongs to a different problem.
2. Lock the constitutive rule. Express the time-indexed variables and state space, initial-slice distribution, intra-slice DAG, inter-slice parent links and temporal order, conditional probability distributions, stationarity and Markov-order assumptions, unrolling horizon, evidence sequence and filtering smoothing prediction or learning task are explicit independently of one notation or implementation.

Knowledge Transfer

Knowledge transfers strongly among subfields of probabilistic graphical models because they reuse the typed probabilistic graphical models carrier, including objects, relations, parameters, conventions, evidence, boundaries, and comparison targets, Variables are indexed by time, a directed acyclic intra-slice graph and forward inter-slice edges factor the joint distribution, and filtering, smoothing or prediction propagates beliefs through the unrolled model., and type the carrier, state every parameter and convention in the definition, test that the time-indexed variables and state space, initial-slice distribution, intra-slice DAG, inter-slice parent links and temporal order, conditional probability distributions, stationarity and Markov-order assumptions, unrolling horizon, evidence sequence and filtering smoothing prediction or learning task are explicit, compare the nearest accepted identity, and report counterexamples, uncertainty, and limiting cases.

Relationships to Other Abstractions

Current abstraction Dynamic Bayesian network Domain-specific

Parents (1) — outgoing typed relations

• Dynamic Bayesian network is a kind of Relation Prime

  It remains its own entry because its identity is fixed by the time-indexed variables and state space, initial-slice distribution, intra-slice DAG, inter-slice parent links and temporal order, conditional probability distributions, stationarity and Markov-order assumptions, unrolling horizon, evidence sequence and filtering smoothing prediction or learning task.

Hierarchy path (1) — routes to 1 parentless root

• Dynamic Bayesian network → is a kind of Relation

Neighborhood in Abstraction Space

Dynamic Bayesian network sits in a crowded region of the domain-specific corpus (19^th percentile for distinctiveness): several abstractions share nearly its structure, so a description that fits it tends to fit its neighbors too.

Family — Graphical & Network Models of Dependence (16 abstractions)

Nearest neighbors

• Variable-order Bayesian network — 0.96
• Factor graph — 0.92
• Widely applicable information criterion — 0.90
• Bayesian model reduction — 0.89
• Ancestral graph — 0.89

Computed from structural-signature embeddings · 2026-10-25