#### DMCA

## Policy Recognition in the Abstract Hidden Markov Model (2002)

### Cached

### Download Links

- [www.computing.edu.au]
- [www.cs.washington.edu]
- [www.cs.cmu.edu]
- [www.jair.org]
- [www.cs.curtin.edu.au]
- DBLP

### Other Repositories/Bibliography

Venue: | Journal of Artificial Intelligence Research |

Citations: | 161 - 25 self |

### Citations

11966 | Maximum likelihood from incomplete data via the em algorithm
- Dempster, Laird, et al.
- 1977
(Show Context)
Citation Context ...ith no clear cut temporal boundary between the policies, the problem becomes a type of parameter estimation for DBN with hidden variables, and techniques for dealing with hidden variables such as EM (=-=Dempster, Laird, & Rubin, 1977-=-) can be applied. Extensions can be made to the AHMM to make the model more expressive and suitable for representing more complex agents' plans. For example, a more expressive plan execution model suc... |

8904 |
Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference
- Pearl
- 1988
(Show Context)
Citation Context ...ial structures of the dynamic model. Later on, Rao-Blackwellisation will be used as our key computational technique for performing policy recognition. 2.1 Bayesian Networks The Bayesian network (BN) (=-=Pearl, 1988-=-; Jensen, 1996; Castillo, Gutierrez, & Hadi, 1997) (also known as probabilistic network or belief network) is a well-established framework for dealing with uncertainty. It provides a graphical and com... |

5889 | A tutorial on hidden Markov models and selected applications in speech recognition
- Rabiner
- 1989
(Show Context)
Citation Context ...Model (AMM). The noisy observation about the environment state (e.g., the eects of action) can then be modelled by making the state \hidden", similar to the hidden state in the Hidden Markov Models (=-=Rabiner, 1989-=-). The result is an interesting and novel stochastic process which we term the Abstract Hidden Markov Model. Intuitively, the 1. Also known as options, policies of Abstract Markov Decision Processes, ... |

1524 |
Local computations with probabilities on graphical structures and their applications to expert systems (with discussion),
- Lauritzen, Spiegelhalter
- 1988
(Show Context)
Citation Context ...he conditional probability of a set of variables given the values of another set of variables (the evidence). There are two types of computation techniques for doing this. Exact inference algorithms (=-=Lauritzen & Spiegelhalter, 1988-=-; Jensen, Lauritzen, & Olesen, 1990; D'Ambrosio, 1993) compute the exact value of the conditional probability required based on analytical transformation that exploits the conditional independence rel... |

1153 |
An Introduction to Bayesian Networks
- Jensen
- 1996
(Show Context)
Citation Context ...s of the dynamic model. Later on, Rao-Blackwellisation will be used as our key computational technique for performing policy recognition. 2.1 Bayesian Networks The Bayesian network (BN) (Pearl, 1988; =-=Jensen, 1996-=-; Castillo, Gutierrez, & Hadi, 1997) (also known as probabilistic network or belief network) is a well-established framework for dealing with uncertainty. It provides a graphical and compact represent... |

1128 | introduction to variational methods for graphical models
- Jordan, Jaakkola, et al.
(Show Context)
Citation Context ...t probabilities. The \heads" are thus chosen deterministically rather than randomly as in sampling-based methods. FHMM (Ghahramani & Jordan, 1997; Jordan et al., 1997) uses variational approximation (=-=Jordan, Ghahramani, Jaakkola, & Saul, 1999-=-) which approximates the full FHMM structure by a sparsied tractable structure. This idea is similar to the structured approximation method in (Boyen & Koller, 1998). Our AHMM can be viewed as a type... |

1051 | On sequential Monte Carlo sampling methods for Bayesian filtering
- Doucet, Godsill, et al.
- 2000
(Show Context)
Citation Context ...oximate 455 Bui, Venkatesh & West distribution that can be represented compactly (Boyen & Koller, 1998), or in the form of a set of weighted samples as in the Sequential Monte-Carlo Sampling methods (=-=Doucet, Godsill, & Andrieu, 2000-=-b; Kanazawa, Koller, & Russell, 1995; Liu & Chen, 1998). The most simple case of the DBN where, in each time-slice, there is only a single state variable and an observation node, is the well-known Hid... |

1026 |
Intention is choice with commitment.
- Levesque
- 1990
(Show Context)
Citation Context ...hat the goal has been achieved, or the attempt to achieve the goal using the current policy has failed. This interpretation of the persistence of a policysts into the persistence model of intentions (=-=Cohen & Levesque, 1990-=-): when an intention ends, there is no guarantee that the intended goal has been achieved. Thus, conceptually, there are two types of destination states: one corresponds to the intended goal states, a... |

663 | Sequential Monte Carlo Methods for Dynamic Systems
- Liu, Chen
- 1998
(Show Context)
Citation Context ... compactly (Boyen & Koller, 1998), or in the form of a set of weighted samples as in the Sequential Monte-Carlo Sampling methods (Doucet, Godsill, & Andrieu, 2000b; Kanazawa, Koller, & Russell, 1995; =-=Liu & Chen, 1998-=-). The most simple case of the DBN where, in each time-slice, there is only a single state variable and an observation node, is the well-known Hidden Markov Model (HMM) (Rabiner, 1989). Filtering in t... |

637 | M.I.J.: Factorial hidden markov models.
- Ghahramani
- 1997
(Show Context)
Citation Context ... recently, extensions of the HMM with multiple hidden interacting chains such as the Coupled Hidden Markov Models (CHMM) and the Factorial Hidden Markov Models (FHMM) have been proposed (Brand, 1997; =-=Ghahramani & Jordan, 1997-=-; Jordan, Ghahramani, & Saul, 1997). In these models, the size of the belief state is exponential in the number of hidden chains. Therefore, the inference and parameter estimation problems become intr... |

568 | Between MDPs and semi-MDPs: A framework for temporal abstraction in reinforcement learning
- Sutton, Precup, et al.
- 1999
(Show Context)
Citation Context ...model the three sources of uncertainty involved. The model for planning with a hierarchy of abstraction under uncertainty has been developed recently by the abstract probabilistic planning community (=-=Sutton, Precup, & Singh, 1999-=-; Parr & Russell, 1997; Forestier & Varaiya, 1978; Hauskrecht, Meuleau, Kaelbling, Dean, & Boutilier, 1998; Dean & Lin, 1995). To our advantage, we adopt their basic model, known as the abstract Marko... |

549 |
A Model for Reasoning About Persistence and Causation,
- Dean, Kanazawa
- 1989
(Show Context)
Citation Context ....e. to determine a set of policies that can explain the sequence of observations at hand. We shall refer to this problem as policy recognition. Viewing the AHMM as a type of dynamic Bayesian network (=-=Dean & Kanazawa, 1989-=-; Nicholson & Brady, 1992), it is known that the complexity of this kind of inferencing in the DBN depends on the size of the representation of the so-called belief state, the conditional joint distri... |

534 |
Planning in a hierarchy of abstraction spaces.
- Sacerdoti
- 1974
(Show Context)
Citation Context ...ate space. It has been noted that a hierarchical organisation of policies can help reduce the complexity of MDP-based planning, similar to the role played by the plan hierarchy in classical planning (=-=Sacerdoti, 1974-=-). In comparison with a classical plan hierarchy, a policy hierarchy can model dierent sources of uncertainty in the planning process such as stochastic actions, uncertain action outcomes, and stocha... |

495 | Bayesian Inference in Econometric Models Using Monte Carlo Integration
- Geweke
- 1989
(Show Context)
Citation Context ...n approximation of the required probability, usually obtained either through \forward" sampling (Henrion, 1988; Fung & Chang, 1989; Shachter & Peot, 1989) (a variance of Bayesian Importance Sampling (=-=Geweke, 1989-=-)), or through Gibbs (Monte-Carlo Markov-Chain) sampling (Pearl, 1987; York, 1992). These algorithms have the advantages of simple implementation, can be applied to all types of network, and can trade... |

360 | A Bayesian model of plan recognition
- Charniak, Goldman
- 1993
(Show Context)
Citation Context ...stulate a set of possible plans for the actor, but is unable to determine which plan is more probable. Since then, the important role of uncertainty reasoning in plan recognition has been recognised (=-=Charniak & Goldman, 1993-=-; Bauer, 1994; van Beek, 1996), and Bayesian probability has been argued as the appropriate model (Charniak & Goldman, 1993; van Beek, 1996). The dynamic, \on-line" aspect of plan recognition has only... |

300 | Tractable inference for complex stochastic processes
- Boyen, Koller
- 1998
(Show Context)
Citation Context ...ng in the DBN depends on the size of the representation of the so-called belief state, the conditional joint distribution of the variables in the DBN at time t given the observation sequence up to t (=-=Boyen & Koller, 1998-=-). Thus we can ask the following question: how does the policy hierarchy aect the size of the belief state representation of the corresponding AHMM? Generally, for a policy hierarchy with K levels, t... |

291 | Approximating probabilistic inference in Bayesian belief networks is NP-hard.
- Dagum, Luby
- 1993
(Show Context)
Citation Context ... NP-hard with respect to the network size (Cooper, 1990), while approximate inference, although scales well with the network size, is NP-hard with respect to the hard-bound accuracy of the estimates (=-=Dagum & Luby, 1993-=-). In the light of these theoretical results, approximate inference can be useful in large networks when exact computation is intractable, but a certain degree of error in the probability estimate can... |

285 | Reinforcement Learning with Hierarchies of Machines
- Parr, Russell
(Show Context)
Citation Context ...rtainty involved. The model for planning with a hierarchy of abstraction under uncertainty has been developed recently by the abstract probabilistic planning community (Sutton, Precup, & Singh, 1999; =-=Parr & Russell, 1997-=-; Forestier & Varaiya, 1978; Hauskrecht, Meuleau, Kaelbling, Dean, & Boutilier, 1998; Dean & Lin, 1995). To our advantage, we adopt their basic model, known as the abstract Markov policies (AMP) 1 as ... |

279 |
Generalized Plan Recognition.
- Kautz, Allen
- 1986
(Show Context)
Citation Context ...ral problem as on-line plan recognition under uncertainty. cs2002 AI Access Foundation and Morgan Kaufmann Publishers. All rights reserved. Bui, Venkatesh & West The seminal work in plan recognition (=-=Kautz & Allen, 1986-=-) considers a plan hierarchy, but does not deal with the uncertainty aspects of the problem. As a result, the approach can only postulate a set of possible plans for the actor, but is unable to determ... |

225 |
Expert Systems and Probabilistic Network Models,
- Castillo, Gutierrez, et al.
- 1997
(Show Context)
Citation Context ...ic model. Later on, Rao-Blackwellisation will be used as our key computational technique for performing policy recognition. 2.1 Bayesian Networks The Bayesian network (BN) (Pearl, 1988; Jensen, 1996; =-=Castillo, Gutierrez, & Hadi, 1997-=-) (also known as probabilistic network or belief network) is a well-established framework for dealing with uncertainty. It provides a graphical and compact representation of the joint probability dist... |

211 |
Propagating Uncertainty in Bayesian Networks by Logic Sampling.
- Henrion
- 1988
(Show Context)
Citation Context ...its the conditional independence relationships of the variables in the network. 454 Policy recognition in the Abstract Hidden Markov Model Approximative inference algorithms (Pearl, 1987; York, 1992; =-=Henrion, 1988-=-; Fung & Chang, 1989; Shachter & Peot, 1989) compute only an approximation of the required probability, usually obtained either through \forward" sampling (Henrion, 1988; Fung & Chang, 1989; Shachter ... |

192 |
Rao-Blackwellisation of sampling schemes,”
- Casella, Robert
- 1996
(Show Context)
Citation Context ...ctly, it can be approximated eÆciently by a collection of compact special belief states. This makes the inference problem in the AHMM particularly amenable to a technique called Rao-Blackwellisation (=-=Casella & Robert, 1996-=-) which allows us to construct hybrid inference methods that combine both exact inference and approximate sampling-based inference for greater eÆciency. The application of Rao-Blackwellisation to the ... |

189 | Stochastic dynamic programming with factored representations.
- Boutilier, Dearden, et al.
- 2000
(Show Context)
Citation Context ...e, even when the set of all policies to be chosen from can be large. In addition, the specication of the selection function and the stopping probabilities can make use of factored representations (=-=Boutilier, Dearden, & Goldszmidt, 2000-=-) in the case where the state space is the composite of a set of relatively independent variables. This ensures that we still have a compact specication 463 Bui, Venkatesh & West ’ s S S’ D d’ pi pi ... |

179 |
Simulation Approaches to General Probabilistic Inference on Belief Networks.
- Shacter, Peot
- 1989
(Show Context)
Citation Context ...lationships of the variables in the network. 454 Policy recognition in the Abstract Hidden Markov Model Approximative inference algorithms (Pearl, 1987; York, 1992; Henrion, 1988; Fung & Chang, 1989; =-=Shachter & Peot, 1989-=-) compute only an approximation of the required probability, usually obtained either through \forward" sampling (Henrion, 1988; Fung & Chang, 1989; Shachter & Peot, 1989) (a variance of Bayesian Impor... |

176 | Stochastic simulation algorithms for dynamic probabilistic networks.
- Kanazawa, Koller, et al.
- 1995
(Show Context)
Citation Context ...istribution that can be represented compactly (Boyen & Koller, 1998), or in the form of a set of weighted samples as in the Sequential Monte-Carlo Sampling methods (Doucet, Godsill, & Andrieu, 2000b; =-=Kanazawa, Koller, & Russell, 1995-=-; Liu & Chen, 1998). The most simple case of the DBN where, in each time-slice, there is only a single state variable and an observation node, is the well-known Hidden Markov Model (HMM) (Rabiner, 198... |

163 | Bayesian map learning in dynamic environments
- Murphy
- 1999
(Show Context)
Citation Context ...ief state tractable, the context variables in the framework of context-specic independence (Boutilier, Friedman, Goldszmidt, & Koller, 1996) can be used conveniently as Rao-Blackwellising variables (=-=Murphy, 2000-=-). Indeed, since the context variable acts as a mixing gate for dierent Bayesian network structures, conditioning on these variables would simplify the structure of the remaining vari460 Policy recog... |

153 |
Bayesian Updating in Recursive Graphical Models by Local Computations,
- Jensen
- 1989
(Show Context)
Citation Context ...et of variables given the values of another set of variables (the evidence). There are two types of computation techniques for doing this. Exact inference algorithms (Lauritzen & Spiegelhalter, 1988; =-=Jensen, Lauritzen, & Olesen, 1990-=-; D'Ambrosio, 1993) compute the exact value of the conditional probability required based on analytical transformation that exploits the conditional independence relationships of the variables in the ... |

141 | Hierarchical Solution of Markov Decision Processes using Macroactions.
- Hauskrecht, Meuleau, et al.
- 1998
(Show Context)
Citation Context ...ierarchy of abstraction under uncertainty has been developed recently by the abstract probabilistic planning community (Sutton, Precup, & Singh, 1999; Parr & Russell, 1997; Forestier & Varaiya, 1978; =-=Hauskrecht, Meuleau, Kaelbling, Dean, & Boutilier, 1998-=-; Dean & Lin, 1995). To our advantage, we adopt their basic model, known as the abstract Markov policies (AMP) 1 as our model for plan execution. The AMP is an extension of a policy in Markov Decision... |

131 | Bayesian models for keyhole plan recognition in an adventure game. User Modelling and User-adapted Interaction 8(1–2):5– 47.
- Albrecht, Zukerman, et al.
- 1998
(Show Context)
Citation Context ...3; van Beek, 1996). The dynamic, \on-line" aspect of plan recognition has only been recently considered (Pynadath & Wellman, 1995, 2000; Goldman, Geib, & Miller, 1999; Huber, Durfee, & Wellman, 1994; =-=Albrecht, Zukerman, & Nicholson, 1998-=-). All of this recent work shares the view that online plan recognition is largely a problem of probabilistic inference in a stochastic process that models the execution of the actor's plan. While thi... |

124 | Basic methods of probabilistic context-free grammars, in: - Jelinek, Lafferty, et al. - 1992 |

122 | Hierarchical Control and Learning for Markov Decision Processes,
- Parr
- 1998
(Show Context)
Citation Context ...d the new plan is part of a new higher level goal. A more expressive language for describing abstract probabilistic plan is the Hierarchical Abstract Machines (HAM) proposed in (Parr & Russell, 1997; =-=Parr, 1998-=-). In a HAM, the abstract policy is replaced by a stochasticsnite automaton, which can call other machines at the lower level. Our abstract policies can be written down as machines of this type. Such ... |

118 | Decomposition techniques for planning in stochastic domains.
- Dean, Lin
- 1995
(Show Context)
Citation Context ...ped recently by the abstract probabilistic planning community (Sutton, Precup, & Singh, 1999; Parr & Russell, 1997; Forestier & Varaiya, 1978; Hauskrecht, Meuleau, Kaelbling, Dean, & Boutilier, 1998; =-=Dean & Lin, 1995-=-). To our advantage, we adopt their basic model, known as the abstract Markov policies (AMP) 1 as our model for plan execution. The AMP is an extension of a policy in Markov Decision Processes (MDP) t... |

115 |
Weighting and integrating evidence for stochastic simulation in Bayesian networks
- Fung, Chag
- 1990
(Show Context)
Citation Context ...onal independence relationships of the variables in the network. 454 Policy recognition in the Abstract Hidden Markov Model Approximative inference algorithms (Pearl, 1987; York, 1992; Henrion, 1988; =-=Fung & Chang, 1989-=-; Shachter & Peot, 1989) compute only an approximation of the required probability, usually obtained either through \forward" sampling (Henrion, 1988; Fung & Chang, 1989; Shachter & Peot, 1989) (a var... |

112 | Evidential reasoning using stochastic simulation of causal models,”
- Pearl
- 1987
(Show Context)
Citation Context ...transformation that exploits the conditional independence relationships of the variables in the network. 454 Policy recognition in the Abstract Hidden Markov Model Approximative inference algorithms (=-=Pearl, 1987-=-; York, 1992; Henrion, 1988; Fung & Chang, 1989; Shachter & Peot, 1989) compute only an approximation of the required probability, usually obtained either through \forward" sampling (Henrion, 1988; Fu... |

83 | Probabilistic statedependent grammars for plan recognition.
- Pynadath, Wellman
- 2000
(Show Context)
Citation Context ...N structure directly. The AMM is also closely related to a model for probabilistic plan recognition called the Probabilistic State-Dependent Grammar (PSDG), independently proposed in (Pynadath, 1999; =-=Pynadath & Wellman, 2000-=-). The PSDG can be described as the Probabilistic Context Free Grammar (PCFG) (Jelinek, Laerty, & Mercer, 1992), augmented with a state space, and a state transition probability table for each termin... |

82 | Coupled hidden Markov models for modeling interacting processes
- Brand
- 1997
(Show Context)
Citation Context ..., 1960). More recently, extensions of the HMM with multiple hidden interacting chains such as the Coupled Hidden Markov Models (CHMM) and the Factorial Hidden Markov Models (FHMM) have been proposed (=-=Brand, 1997-=-; Ghahramani & Jordan, 1997; Jordan, Ghahramani, & Saul, 1997). In these models, the size of the belief state is exponential in the number of hidden chains. Therefore, the inference and parameter esti... |

77 | The automated mapping of plans for plan recognition. - Huber, Durfee, et al. - 1994 |

73 | Accounting for context in plan recognition, with application to traffic monitoring.
- Pynadath, Wellman
- 1995
(Show Context)
Citation Context ...96), and Bayesian probability has been argued as the appropriate model (Charniak & Goldman, 1993; van Beek, 1996). The dynamic, \on-line" aspect of plan recognition has only been recently considered (=-=Pynadath & Wellman, 1995-=-, 2000; Goldman, Geib, & Miller, 1999; Huber, Durfee, & Wellman, 1994; Albrecht, Zukerman, & Nicholson, 1998). All of this recent work shares the view that online plan recognition is largely a problem... |

71 |
Dynamic network models for forecasting, in:
- Dagum, Galper, et al.
- 1992
(Show Context)
Citation Context ...n be tolerated by the application. 2.2 Dynamic Bayesian Networks To model the temporal dynamics of the environment, the Dynamic Bayesian Network (DBN) (Dean & Kanazawa, 1989; Nicholson & Brady, 1992; =-=Dagum, Galper, & Horvitz, 1992-=-) is a special Bayesian network architecture for representing the evolution of the domain variables over time. A DBN consists of a sequence of time-slices where each time-slice contains a set of varia... |

70 | TD models: Modeling the world at a mixture of time scales.
- Sutton
- 1995
(Show Context)
Citation Context ...stop after one time-step, D a S a ands(d) = 18d 2 D a (Sutton et al., 1999). The idea that policies with suitable stopping condition can be viewed just as primitive actions issrst made explicit in (=-=Sutton, 1995-=-), which also introduces thesmodel for representing the stopping probabilities. Their subsequent work (Sutton et al., 1999) introduces the abstract policy concept under the name options. The execution... |

52 | Hidden Markov decision trees
- Ghahramani, Saul
- 1996
(Show Context)
Citation Context ...e HMM with multiple hidden interacting chains such as the Coupled Hidden Markov Models (CHMM) and the Factorial Hidden Markov Models (FHMM) have been proposed (Brand, 1997; Ghahramani & Jordan, 1997; =-=Jordan, Ghahramani, & Saul, 1997-=-). In these models, the size of the belief state is exponential in the number of hidden chains. Therefore, the inference and parameter estimation problems become intractable if the number of hidden ch... |

48 |
Incremental Probabilistic Inference.
- D’Ambrosio
- 1993
(Show Context)
Citation Context ... another set of variables (the evidence). There are two types of computation techniques for doing this. Exact inference algorithms (Lauritzen & Spiegelhalter, 1988; Jensen, Lauritzen, & Olesen, 1990; =-=D'Ambrosio, 1993-=-) compute the exact value of the conditional probability required based on analytical transformation that exploits the conditional independence relationships of the variables in the network. 454 Polic... |

46 |
A new approach to linear and prediction problems
- Kalman
- 1960
(Show Context)
Citation Context ...den Markov Model (HMM) (Rabiner, 1989). Filtering in this simple structure can be solved using dynamic programming in the discrete HMM (Rabiner, 1989), or Kalmansltering in the linear Gaussian model (=-=Kalman, 1960-=-). More recently, extensions of the HMM with multiple hidden interacting chains such as the Coupled Hidden Markov Models (CHMM) and the Factorial Hidden Markov Models (FHMM) have been proposed (Brand,... |

35 |
Evaluating in diagrams
- Shachter
- 1986
(Show Context)
Citation Context ...root at level k. The root of the chain can be moved from k to another level k 0 simply by reversing the links lying between k and k 0 using the standard link-reversal operation for Bayesian networks (=-=Shachter, 1986-=-). Each node in the belief chain also has a manageable size. In principle, the domain of k t is k , the set of all policies at level k, and the domain of s t is S, the set of all possible states. ... |

29 |
Multilayer control of large Markov chains.
- Forestier, Varaiya
- 1978
(Show Context)
Citation Context ...model for planning with a hierarchy of abstraction under uncertainty has been developed recently by the abstract probabilistic planning community (Sutton, Precup, & Singh, 1999; Parr & Russell, 1997; =-=Forestier & Varaiya, 1978-=-; Hauskrecht, Meuleau, Kaelbling, Dean, & Boutilier, 1998; Dean & Lin, 1995). To our advantage, we adopt their basic model, known as the abstract Markov policies (AMP) 1 as our model for plan executio... |

29 | Probabilistic Grammars for Plan Recognition
- Pynadath
- 1999
(Show Context)
Citation Context ...rence on this DBN structure directly. The AMM is also closely related to a model for probabilistic plan recognition called the Probabilistic State-Dependent Grammar (PSDG), independently proposed in (=-=Pynadath, 1999-=-; Pynadath & Wellman, 2000). The PSDG can be described as the Probabilistic Context Free Grammar (PCFG) (Jelinek, Laerty, & Mercer, 1992), augmented with a state space, and a state transition probabi... |

19 | Hybrid propagation in junction trees. - Dawid, Kjærulff, et al. - 1995 |

18 |
Context-speci independence in Bayesian networks
- Boutilier, Friedman, et al.
- 1996
(Show Context)
Citation Context ...ere the RB belief state can be maintained by a Kalmanslter. Since we have to make the Rao-Blackwellised belief state tractable, the context variables in the framework of context-specic independence (=-=Boutilier, Friedman, Goldszmidt, & Koller, 1996-=-) can be used conveniently as Rao-Blackwellising variables (Murphy, 2000). Indeed, since the context variable acts as a mixing gate for dierent Bayesian network structures, conditioning on these vari... |

17 | On the recognition of abstract Markov policies. - Bui, Venkatesh, et al. - 2000 |

14 | Integrating probabilistic reasoning into plan recognition.
- Bauer
- 1994
(Show Context)
Citation Context ...plans for the actor, but is unable to determine which plan is more probable. Since then, the important role of uncertainty reasoning in plan recognition has been recognised (Charniak & Goldman, 1993; =-=Bauer, 1994-=-; van Beek, 1996), and Bayesian probability has been argued as the appropriate model (Charniak & Goldman, 1993; van Beek, 1996). The dynamic, \on-line" aspect of plan recognition has only been recentl... |

13 | A computational scheme for reasoning in dynamic probabilistic networks - Kjaerul - 1992 |

9 | Rao-Blackwellised particle for dynamic Bayesian networks - Doucet, Freitas, et al. - 2000 |

4 |
Layered dynamic Bayesian networks for spatio-temporal modelling. Intelli- gent Data Analysis
- Bui, Venkatesh, et al.
- 1999
(Show Context)
Citation Context ... cope with this complexity, an approximation scheme such as sequential importance sampling (SIS) (Doucet et al., 2000b; Liu & Chen, 1998; Kanazawa et al., 1995) can be employed. In our previous work (=-=Bui, Venkatesh, & West, 1999-=-), we have applied an SIS method known as the likelihoodweighting with evidence reversal (LW-ER) (Kanazawa et al., 1995) to an AHMM-like network structure. However the SIS method needs to sample in th... |

4 | An investigation of probabilistic interpretations of heuristics in plan recognition. - Beek - 1996 |

3 | Rao-blackwellised particle ltering for dynamic bayesian networks - Doucet, Freitas, et al. - 2000 |

3 | dHugin: A computational system for dynamic time-sliced Bayesian networks - Kjrul - 1995 |

3 | Rao-blackwellised particle for dynamic Bayesian networks - Murphy, Russell - 2001 |

3 | Coordination of multiple cameras to track multiple people
- Nguyen, Venkatesh, et al.
- 2002
(Show Context)
Citation Context ...iable as the cameras have to deal with noisy video frames and occlusion of objects in the scene. For more information on how low-level tracking is done with multiple cameras, readers are referred to (=-=Nguyen, Venkatesh, West, & Bui, 2002-=-). We assume that the observation of a state can only be in the area surrounding it, thus the observation model is a matrix specifying the observation likelihood for each cell within a neighbourhood o... |

2 |
Particle for partially observed Gaussian state space models
- Andrieu, Doucet
- 2000
(Show Context)
Citation Context ...ctable form. The use of non-Markov RB variables also appears in other special models such as the Bayesian missing data model (Liu & Chen, 1998), and the partially observed Gaussian state space model (=-=Andrieu & Doucet, 2000-=-) where the RB belief state can be maintained by a Kalmanslter. Since we have to make the Rao-Blackwellised belief state tractable, the context variables in the framework of context-specic independen... |

1 |
Policy recognition in the Abstract Hidden Markov Model
- Cooper
- 1990
(Show Context)
Citation Context ..., can be applied to all types of network, and can trade o the accuracy in the estimates for computation resources. It is known that exact inference in BN is NP-hard with respect to the network size (=-=Cooper, 1990-=-), while approximate inference, although scales well with the network size, is NP-hard with respect to the hard-bound accuracy of the estimates (Dagum & Luby, 1993). In the light of these theoretical ... |

1 |
Policy recognition in the Abstract Hidden Markov Model
- Nicholson, Brady
- 1992
(Show Context)
Citation Context ...of policies that can explain the sequence of observations at hand. We shall refer to this problem as policy recognition. Viewing the AHMM as a type of dynamic Bayesian network (Dean & Kanazawa, 1989; =-=Nicholson & Brady, 1992-=-), it is known that the complexity of this kind of inferencing in the DBN depends on the size of the representation of the so-called belief state, the conditional joint distribution of the variables i... |