Forward-backward algorithm
Encyclopedia
The forward–backward algorithm is an inference
algorithm
for hidden Markov models which computes the posterior
marginals of all hidden state variables given a sequence of observations/emissions
, i.e. it computes, for all hidden state variables 
, the distribution 
. This inference task is usually called smoothing. The algorithm makes use of the principle of dynamic programming
to efficiently compute the values that are required to obtain the posterior marginal distributions in two passes. The first pass goes forward in time while the second goes backward in time; hence the name forward–backward algorithm.
The term forward–backward algorithm is also used to refer to any algorithm belonging to the general class of algorithms that operate on sequence models in a forward–backward manner. In this sense, the descriptions in the remainder of this article refer but to one specific instance of this class.
, the probability of ending up in any particular state given the first 
observations in the sequence, i.e. 
. In the second pass, the algorithm computes a set of backward probabilities which provide the probability of observing the remaining observations given any starting point 
, i.e. 
. These two sets of probability distributions can then be combined to obtain the distribution over states at any specific point in time given the entire observation sequence:
The last step follows from an application of Bayes' rule
and the conditional independence
of
and 
given 
.
As outlined above, the algorithm involves three steps:
The forward and backward steps are often called "forward message pass" and "backward message pass". The wording originates from the way the algorithm processes the given observation sequence. First the algorithm moves forward starting with the first observation in the sequence and going to the last, and then returning back to the first. At each single observation in the sequence, probabilities to be used for calculations at the next observation are computed. During the backward pass the algorithm simultaneously performs the smoothing step. This step allows the algorithm to take into account any past observations of output for computing more accurate results.
The forward–backward algorithm can be used to find the most likely state for any point in time. It cannot, however, be used to find the most likely sequence of states (see Viterbi algorithm
).
into matrix notation as follows.
The transition probabilities
of a given random variable 
representing all possible states in the hidden Markov model will be represented by the matrix 
where the row index, i, will represent the start state and the column index, j, represents the target state. The example below represents a system where the probability of staying in the same state after each step is 70% and the probability of transitioning to the other state is 30%. The transition matrix is then:
In a typical Markov model we would multiply a state vector by this matrix to obtain the probabilities for the subsequent state. In a hidden Markov model the state is unknown, and we instead observe events associated with the possible states. An event matrix of the form:
provides the probabilities for observing events given a particular state. In the above example, event 1 will be observed 90% of the time if we are in state 1 while event 2 has a 10% probability of occurring in this state. In contrast, event 1 will only be observed 20% of the time if we are in state 2 and event 2 has an 80% chance of occurring. Given a state vector (
), the probability of observing event j is then:
This can be represented in matrix form by multiplying the state vector (
) by an observation matrix (
) containing only diagonal entries. Each entry is the probability of the observed event given each state. Continuing the above example, an observation of event 1 would be:
This allows us to calculate the probabilities associated with transitioning to a new state and observing the given event as:
The probability vector that results contains entries indicating the probability of transitioning to each state and observing the given event. This process can be carried forward with additional observations using:
This value is the forward probability vector. The i'th entry of this vector provides:
Typically, we will normalize the probability vector at each step so that its entries sum to 1. A scaling factor is thus introduced at each step such that:
where
represents the scaled vector from the previous step and 
represents the scaling factor that causes the resulting vector's entries to sum to 1. The product of the scaling factors is the total probability for observing the given events irrespective of the final states:
This allows us to interpret the scaled probability vector as:
We thus find that the product of the scaling factors provides us with the total probability for observing the given sequence up to time t and that the scaled probability vector provides us with the probability of being in each state at this time.
That is, we now want to assume that we start in a particular state (
), and we are now interested in the probability of observing all future events from this state. Since the initial state is assumed as given (i.e. the prior probability of this state = 100%), we begin with:
Notice that we are now using a column vector while the forward probabilities used row vectors. We can then work backwards using:
While we could normalize this vector as well so that its entries sum to one, this is not usually done. Noting that each entry contains the probability of the future event sequence given a particular initial state, normalizing this vector would be equivalent to applying Bayes' theorem to find the likelihood of each initial state given the future events (assuming uniform priors for the final state vector). However, it is more common to scale this vector using the same
constants used in the forward probability calculations. 
is not scaled, but subsequent operations use:
where
represents the previous, scaled vector. This result is that the scaled probability vector is related to the backward probabilities by:
This is useful because it allows us to find the total probability of being in each state at a given time, t, by multiplying these values:
To understand this, we note that
provides the probability for observing the given events in a way that passes through state 
at time t. This probability includes the forward probabilities covering all events up to time t as well as the backward probabilities which include all future events. This is the numerator we are looking for in our equation, and we divide by the total probability of the observation sequence to normalize this value and extract only the probability that 
. These values are sometimes called the "smoothed values" as they combine the forward and backward probabilities to compute a final probability.
The values
thus provide the probability of being in each state at time t. As such, they are useful for determining the most probable state at any time. It should be noted, however, that the term "most probable state" is somewhat ambiguous. While the most probable state is the most likely to be correct at a given point, the sequence of individually probable states is not likely to be the most probable sequence. This is because the probabilities for each point are calculated independently of each other. They do not take into account the transition probabilities between states, and it is thus possible to get states at two moments (t and t+1) that are both most probable at those time points but which have very little probability of occurring together, i.e. 
. The most probable sequence of states that produced an observation sequence can be found using the Viterbi algorithm
.
We also assume each state generates 2 events: event 1 = umbrella, event 2 = no umbrella. The conditional probabilities for these occurring in each state are given by the probability matrix:
We then observe the following sequence of events: {umbrella, umbrella, no umbrella, umbrella, umbrella} which we will represent in our calculations as:
Note that
differs from the others because of the "no umbrella" observation.
In computing the forward probabilities we begin with:
which is our prior state vector indicating that we don't know which state the weather is in before our observations. While a state vector should be given as a row vector, we will use the transpose of the matrix so that the calculations below are easier to read. Our calculations are then written in the form:
instead of:
Notice that the transformation matrix is also transposed, but in our example the transpose is equal to the original matrix. Performing these calculations and normalizing the results provides:
For the backward probabilities we start with:
We are then able to compute (using the observations in reverse order and normalizing with different constants):
Finally, we will compute the smoothed probability values. These result also must be scaled so that its entries sum to 1 because we did not scale the backward probabilities with the
's found earlier. The backward probability vectors above thus actually represent the likelihood of each state at time t given the future observations. Because these vectors are proportional to the actual backward probabilities, the result has to be scaled an additional time.
Notice that the value of
is equal to 
and that 
is equal to 
. This follows naturally because both 
and 
begin with uniform priors over the initial and final state vectors (respectively) and take into account all of the observations. However, 
will only be equal to 
when our initial state vector represents a uniform prior (i.e. all entries are equal). When this is not the case 
needs to be combined with the initial state vector to find the most likely initial state. We thus find that the forward probabilities by themselves are sufficient to calculate the most likely final state. Similarly, the backward probabilities can be combined with the initial state vector to provide the most probable initial state given the observations. The forward and backward probabilities need only be combined to infer the most probable states between the initial and final points.
The calculations above reveal that the most probable weather state on every day except for the third one was "rain." They tell us more than this, however, as they now provide a way to quantify the probabilities of each state at different times. Perhaps most importantly, our value at
quantifies our knowledge of the state vector at the end of the observation sequence. We can then use this to predict the probability of the various weather states tomorrow as well as the probability of observing an umbrella.
state sequences and calculating the joint probability of each state sequence with the observed series of events. This approach has time complexity
, where 
is the length of sequences and 
is the number of symbols in the state alphabet. This is intractable for realistic problems, as the number of possible hidden node sequences typically is extremely high. However, the forward–backward algorithm has time complexity 
.
Several enhancements are known to the general forward–backward algorithm which allow for computations to take place in constant space. In addition, as it grows, algorithms have been developed to compute
efficiently through online smoothing such as the fixed-lag smoothing (FLS) algorithm Russel & Norvig 2003 pp. 552.
Inference
Inference is the act or process of deriving logical conclusions from premises known or assumed to be true. The conclusion drawn is also called an idiomatic. The laws of valid inference are studied in the field of logic.Human inference Inference is the act or process of deriving logical conclusions...
algorithm
Algorithm
In mathematics and computer science, an algorithm is an effective method expressed as a finite list of well-defined instructions for calculating a function. Algorithms are used for calculation, data processing, and automated reasoning...
for hidden Markov models which computes the posterior
Posterior probability
In Bayesian statistics, the posterior probability of a random event or an uncertain proposition is the conditional probability that is assigned after the relevant evidence is taken into account...
marginals of all hidden state variables given a sequence of observations/emissions
, i.e. it computes, for all hidden state variables , the distribution . This inference task is usually called smoothing. The algorithm makes use of the principle of dynamic programmingDynamic programming
In mathematics and computer science, dynamic programming is a method for solving complex problems by breaking them down into simpler subproblems. It is applicable to problems exhibiting the properties of overlapping subproblems which are only slightly smaller and optimal substructure...
to efficiently compute the values that are required to obtain the posterior marginal distributions in two passes. The first pass goes forward in time while the second goes backward in time; hence the name forward–backward algorithm.
The term forward–backward algorithm is also used to refer to any algorithm belonging to the general class of algorithms that operate on sequence models in a forward–backward manner. In this sense, the descriptions in the remainder of this article refer but to one specific instance of this class.
Overview
In the first pass, the forward–backward algorithm computes a set of forward probabilities which provide, for all, the probability of ending up in any particular state given the first observations in the sequence, i.e. . In the second pass, the algorithm computes a set of backward probabilities which provide the probability of observing the remaining observations given any starting point , i.e. . These two sets of probability distributions can then be combined to obtain the distribution over states at any specific point in time given the entire observation sequence:The last step follows from an application of Bayes' rule
Bayes' rule
In probability theory and applications, Bayes' rule relates the odds of event A_1 to event A_2, before and after conditioning on event B. The relationship is expressed in terms of the Bayes factor, \Lambda. Bayes' rule is derived from and closely related to Bayes' theorem...
and the conditional independence
Conditional independence
In probability theory, two events R and B are conditionally independent given a third event Y precisely if the occurrence or non-occurrence of R and the occurrence or non-occurrence of B are independent events in their conditional probability distribution given Y...
of
and given .As outlined above, the algorithm involves three steps:
- computing forward probabilities
- computing backward probabilities
- computing smoothed values.
The forward and backward steps are often called "forward message pass" and "backward message pass". The wording originates from the way the algorithm processes the given observation sequence. First the algorithm moves forward starting with the first observation in the sequence and going to the last, and then returning back to the first. At each single observation in the sequence, probabilities to be used for calculations at the next observation are computed. During the backward pass the algorithm simultaneously performs the smoothing step. This step allows the algorithm to take into account any past observations of output for computing more accurate results.
The forward–backward algorithm can be used to find the most likely state for any point in time. It cannot, however, be used to find the most likely sequence of states (see Viterbi algorithm
Viterbi algorithm
The Viterbi algorithm is a dynamic programming algorithm for finding the most likely sequence of hidden states – called the Viterbi path – that results in a sequence of observed events, especially in the context of Markov information sources, and more generally, hidden Markov models...
).
Forward probabilities
The following description takes as its base matrices of probability values rather than probability distributions. We transform the probability distributions related to a given hidden Markov modelHidden Markov model
A hidden Markov model is a statistical Markov model in which the system being modeled is assumed to be a Markov process with unobserved states. An HMM can be considered as the simplest dynamic Bayesian network. The mathematics behind the HMM was developed by L. E...
into matrix notation as follows.
The transition probabilities
of a given random variable representing all possible states in the hidden Markov model will be represented by the matrix where the row index, i, will represent the start state and the column index, j, represents the target state. The example below represents a system where the probability of staying in the same state after each step is 70% and the probability of transitioning to the other state is 30%. The transition matrix is then:In a typical Markov model we would multiply a state vector by this matrix to obtain the probabilities for the subsequent state. In a hidden Markov model the state is unknown, and we instead observe events associated with the possible states. An event matrix of the form:
provides the probabilities for observing events given a particular state. In the above example, event 1 will be observed 90% of the time if we are in state 1 while event 2 has a 10% probability of occurring in this state. In contrast, event 1 will only be observed 20% of the time if we are in state 2 and event 2 has an 80% chance of occurring. Given a state vector (
), the probability of observing event j is then:This can be represented in matrix form by multiplying the state vector (
) by an observation matrix () containing only diagonal entries. Each entry is the probability of the observed event given each state. Continuing the above example, an observation of event 1 would be:This allows us to calculate the probabilities associated with transitioning to a new state and observing the given event as:
The probability vector that results contains entries indicating the probability of transitioning to each state and observing the given event. This process can be carried forward with additional observations using:
This value is the forward probability vector. The i'th entry of this vector provides:
Typically, we will normalize the probability vector at each step so that its entries sum to 1. A scaling factor is thus introduced at each step such that:
where
represents the scaled vector from the previous step and represents the scaling factor that causes the resulting vector's entries to sum to 1. The product of the scaling factors is the total probability for observing the given events irrespective of the final states:This allows us to interpret the scaled probability vector as:
We thus find that the product of the scaling factors provides us with the total probability for observing the given sequence up to time t and that the scaled probability vector provides us with the probability of being in each state at this time.
Backward probabilities
A similar procedure can be constructed to find backward probabilities. These intend to provide the probabilities:That is, we now want to assume that we start in a particular state (
), and we are now interested in the probability of observing all future events from this state. Since the initial state is assumed as given (i.e. the prior probability of this state = 100%), we begin with:Notice that we are now using a column vector while the forward probabilities used row vectors. We can then work backwards using:
While we could normalize this vector as well so that its entries sum to one, this is not usually done. Noting that each entry contains the probability of the future event sequence given a particular initial state, normalizing this vector would be equivalent to applying Bayes' theorem to find the likelihood of each initial state given the future events (assuming uniform priors for the final state vector). However, it is more common to scale this vector using the same
constants used in the forward probability calculations. is not scaled, but subsequent operations use:where
represents the previous, scaled vector. This result is that the scaled probability vector is related to the backward probabilities by:This is useful because it allows us to find the total probability of being in each state at a given time, t, by multiplying these values:
To understand this, we note that
provides the probability for observing the given events in a way that passes through state at time t. This probability includes the forward probabilities covering all events up to time t as well as the backward probabilities which include all future events. This is the numerator we are looking for in our equation, and we divide by the total probability of the observation sequence to normalize this value and extract only the probability that . These values are sometimes called the "smoothed values" as they combine the forward and backward probabilities to compute a final probability.The values
thus provide the probability of being in each state at time t. As such, they are useful for determining the most probable state at any time. It should be noted, however, that the term "most probable state" is somewhat ambiguous. While the most probable state is the most likely to be correct at a given point, the sequence of individually probable states is not likely to be the most probable sequence. This is because the probabilities for each point are calculated independently of each other. They do not take into account the transition probabilities between states, and it is thus possible to get states at two moments (t and t+1) that are both most probable at those time points but which have very little probability of occurring together, i.e. . The most probable sequence of states that produced an observation sequence can be found using the Viterbi algorithmViterbi algorithm
The Viterbi algorithm is a dynamic programming algorithm for finding the most likely sequence of hidden states – called the Viterbi path – that results in a sequence of observed events, especially in the context of Markov information sources, and more generally, hidden Markov models...
.
Example
This example takes as its basis the umbrella world in Russel & Norvig 2003 pp. 540 in which we would like to infer the weather given observation of a man either carrying or not carrying an umbrella. We assume two possible states for the weather: state 1 = rain, state 2 = no rain. We assume that the weather has a 70% chance of staying the same each day and a 30% chance of changing. The transition probabilities are then:We also assume each state generates 2 events: event 1 = umbrella, event 2 = no umbrella. The conditional probabilities for these occurring in each state are given by the probability matrix:
We then observe the following sequence of events: {umbrella, umbrella, no umbrella, umbrella, umbrella} which we will represent in our calculations as:
Note that
differs from the others because of the "no umbrella" observation.In computing the forward probabilities we begin with:
which is our prior state vector indicating that we don't know which state the weather is in before our observations. While a state vector should be given as a row vector, we will use the transpose of the matrix so that the calculations below are easier to read. Our calculations are then written in the form:
instead of:
Notice that the transformation matrix is also transposed, but in our example the transpose is equal to the original matrix. Performing these calculations and normalizing the results provides:
For the backward probabilities we start with:
We are then able to compute (using the observations in reverse order and normalizing with different constants):
Finally, we will compute the smoothed probability values. These result also must be scaled so that its entries sum to 1 because we did not scale the backward probabilities with the
's found earlier. The backward probability vectors above thus actually represent the likelihood of each state at time t given the future observations. Because these vectors are proportional to the actual backward probabilities, the result has to be scaled an additional time.Notice that the value of
is equal to and that is equal to . This follows naturally because both and begin with uniform priors over the initial and final state vectors (respectively) and take into account all of the observations. However, will only be equal to when our initial state vector represents a uniform prior (i.e. all entries are equal). When this is not the case needs to be combined with the initial state vector to find the most likely initial state. We thus find that the forward probabilities by themselves are sufficient to calculate the most likely final state. Similarly, the backward probabilities can be combined with the initial state vector to provide the most probable initial state given the observations. The forward and backward probabilities need only be combined to infer the most probable states between the initial and final points.The calculations above reveal that the most probable weather state on every day except for the third one was "rain." They tell us more than this, however, as they now provide a way to quantify the probabilities of each state at different times. Perhaps most importantly, our value at
quantifies our knowledge of the state vector at the end of the observation sequence. We can then use this to predict the probability of the various weather states tomorrow as well as the probability of observing an umbrella.Performance
The brute-force procedure for the solution of this problem is the generation of all possible state sequences and calculating the joint probability of each state sequence with the observed series of events. This approach has time complexityTime complexity
In computer science, the time complexity of an algorithm quantifies the amount of time taken by an algorithm to run as a function of the size of the input to the problem. The time complexity of an algorithm is commonly expressed using big O notation, which suppresses multiplicative constants and...
, where is the length of sequences and is the number of symbols in the state alphabet. This is intractable for realistic problems, as the number of possible hidden node sequences typically is extremely high. However, the forward–backward algorithm has time complexity .Several enhancements are known to the general forward–backward algorithm which allow for computations to take place in constant space. In addition, as it grows, algorithms have been developed to compute
efficiently through online smoothing such as the fixed-lag smoothing (FLS) algorithm Russel & Norvig 2003 pp. 552.Pseudocode
ForwardBackward(guessState, sequenceIndex):
if sequenceIndex is past the end of the sequence, return 1
if (guessState, sequenceIndex) has been seen before, return saved result
result = 0
for each neighboring state n:
result = result + (transition probability from guessState to n given observation element at sequenceIndex)
* ForwardBackward(n, sequenceIndex+1)
save result for (guessState, sequenceIndex)
return result
External links
- An interactive spreadsheet for teaching the forward–backward algorithm (spreadsheet and article with step-by-step walk-through)
- Tutorial of hidden Markov models including the forward–backward algorithm
- Collection of AI algorithms implemented in Java (including HMM and the forward–backward algorithm)