H ?
«
‹
←
→
›
»
Language peer sets for IBAL:
United States↑
United States/0↑
Genus Bayesian ↑
Specialised Languages ↑
Bayesian↑
Logic-reflexive↑
Close mapping ↑
Bayesian/0↑
Logic-reflexive/0↑
Close mapping/0↑
Bayesian/United States↑
Logic-reflexive/United States↑
Close mapping/United States↑
Specialised Languages ↑
Specialised Languages/0↑
Specialised Languages/us ↑
IBAL(ID:7766/)
alternate simple view
Country: United States
Genus: Bayesian
Sammet category: Specialised Languages
"IBAL (pronounced "eyeball") is a general-purpose language for probabilistic modeling, parameter estimation and decision making. It generalizes Bayesian networks, hidden Markov models, stochastic context free grammars, Markov decision processes, and allows many new possibilities. It also provides a convenient programming-language framework with libraries, automatic type checking and so on."
References:
Pfeffer, A. (2000) Pfeffer, A. "A Bayesian Language for Cumulative Learning" AAAI Workshop on Learning Statistical Models from Relational Data, July 2000
Pfeffer, A. (2001) Pfeffer, A. "IBAL: An Integrated Bayesian Agent Language", IJCAI 2001
Ramsey, N. (2002) Ramsey, N. and Pfeffer, A. Stochastic Lambda Calculus and Monads of Probability Distributions, POPL 2002
Resources
-
IBAL home page
-
IBAL is a general-purpose probabilistic modeling language. It is built on the simple idea that writing a probabilistic model should be as easy as writing a simulator. If you can write a stochastic simulation of your domain, IBAL will apply probabilistic reasoning techniques to compute a probability distribution over the results of the simulation. An IBAL model looks like a computer program with stochastic branches. The program defines the simulation process by which outputs are generated. Rather than simply running the simulation to generate a particular output, IBAL allows you to compute a probability distribution over the program outputs.
The design of IBAL's language is based on the functional programming language model of Lisp, ML and Haskell. IBAL provides many powerful programming-language features for defining models. These include a rich type system, allowing many different kinds of data structures to be defined; functions as first-class citizens in the language; objects with encapsulation; libraries to allow modular construction of large models; and automated type inference.
In addition to describing a model, IBAL allows you to make observations about the outcome of the model. These observations condition the probability distribution over outputs, and can also be used to learn the probabilistic parameters of a model. IBAL also allows you to specify utilities associated with different possible outcomes, and to define decision variables that will be chosen in order to maximize the expected utility.
IBAL's inference engine builds on many sophisticated probabilistic reasoning techniques. These include variable elimination, allowing the engine to exploit conditional independence relationships between variables; structured factors, allowing low-level structural features like context-specific independence to be exploited; object-based reasoning, allowing the large-scale object structure of a domian to be exploited; backward induction for reasoning about sequential decision problems; memoization and dynamic programming, allowing computation to be reused between different objects of the same class, and great speedups to be obtained in recursive models; lazy evaluation, allowing computation to be performed on finite parts of infinite models; and support- and evidence-directed computation, allowing computation to be simplified by taking advantage of knowledge about the possible values of variables. By combining all these techniques, IBAL is able to generalize many existing frameworks, such as Bayesian networks, hidden Markov models, stochastic context free grammars and Markov decision processes, while paving the way for a variety of new ones.
Search in:
Google
Google scholar
World Cat
Yahoo
Overture
DBLP
Monash bib
NZ
IEEE
ACM portal
CiteSeer
CSB
ncstrl
jstor
Bookfinder
|