PIERRE LISON THESIS

And finally, in complement to information refinement, beliefs are also abstracted by constructing high-level, amodal sym- bolic representations from low-level perceptual i. They therefore serve as a representational backbone for a wide range of high-level cognitive capabilities related to reasoning, planning and learning in complex and dynamic environ- ments. After defending his PhD in , he was awarded a research grant from the Research Council of Norway for a postdoctoral project on statistical machine translation, also conducted at the University of Oslo. The joint probability distribution of the Markov network can then be factorised over the cliques of G: The mug is perceived by the robot sensors for instance, by one binocular camera, or by a haptic sensor mounted on a robotic arm.

Markov Logic is a combination of first-order logic and probabilistic graphical models. This situation awareness is generally expressed in some sort of belief models in which various aspects of the external reality are encoded. Key to our approach is the use of a first-order probabilistic language, Markov Logic [24], as a uni- fied representation formalism to construct rich, multi-modal models of context. Given the problem complexity we just outlined, exact inference is unfeasible. Pierre Lison holds a M. The belief i also specifies a belief history h.

New Research Scientists have joined SAMBA | Norsk Regnesentral

Belief models are usually expressed as high level pjerre representations merging and abstracting information over multiple modalities. Click here to sign up. This is expressed as two set of pointers: The probability of a false positive is 0.

pierre lison thesis

Max-margin weight learning for markov logic networks. The core of the binder is its working memory, where beliefs are formed from incoming per- ceptual inputs, and are then iteratively fused, refined and abstracted to yield stable, high-level beliefs.

  CYCLONE SIDR BANGLADESH 2007 CASE STUDY

The beliefs can then be directly exploited by high-level cognitive functions such as planning, cross-modal learning or communication.

pierre lison thesis

Learning words and syntax for a scene description task. Section 6 provides additional details on the attention and filtering systems. First, the newly lieon belief is compared to the piedre existing beliefs for similarity. Markov Logic is a combination of first-order logic and probabilistic graphical models. Generation and evaluation of user tailored responses in multimodal dialogue.

Beliefs also incorporate various contextual information such as spatio-temporal framing, multi-agent epistemic status, and saliency measures. In most cases, weights are learned based on training samples extracted from a relational database. Now, off to a week of well-deserved vacation ;- [] A-magasinet featured this week a three-pages article on Lenny and my research on human-robot interaction through spoken dialogue.

The resulting beliefs can also be easily accessed and retrieved by the other subarchitectures. Such pointers are crucial to capture relational structures between entities. In Perception and Lisoj Technologies: Shared beliefs contain information which is part of the common ground for the group [3]. His thesis developed a new approach for dialogue management which combines expert domain knowledge and statistical models into a unified framework.

A more appropriate solution thesls be found in the use of anytime algorithms combined with various approxima- tion methods.

Information fusion for visual reference resolution in dynamic situated dialogue. Multivariate probability distributions over possible values are used to account for the partial observability of the data, while the first-order expressivity of Markov Logic allows us to consisely describe and reason over complex relational structures.

  DISSERTATION LAUTONOMIE ORIGINELLE DU DROIT ADMINISTRATIF

The components can access sensors, effectors, as well as a blackboard working memory available for the whole subarchitecture.

Pierre Lison Completes Doctoral Degree

They therefore serve as a representational backbone for a wide range of high-level cognitive capabilities related to reasoning, planning and learning in complex and dynamic environ- ments. The belief models must therefore incorporate mecha- nisms for computing and adapting these saliency measures over time. Such perceptual belief i would be formally defined as: For human- robot interaction, the incorporated knowledge might include inter alia: This section explores these two related issues.

After defending his PhD inhe was awarded a research grant from the Research Council of Norway for a postdoctoral project on statistical machine translation, also conducted at the University of Oslo. A-magasinet featured this week a three-pages article on Lenny and my research on human-robot interaction through spoken dialogue. Enter the email address you signed up with and we’ll email you a reset link.

This explicit control of information and processing is crucial to dynamically balance and constrain the computational load among components. Given the probabilistic nature of the framework, the number of beliefs is likely to grow exponentially over time. PhD thesis, MIT,