Log In Sign Up. Computer Speech and Language, 16 3 , This is the well-known engineering problem of multi-target, multi-sensor data fusion [7]. Such retrieval operation allows each subarchitecture to use the binder as a system-wide information repository about the world state. NR continues to strengthen the staff. Belief models are usually expressed as high level symbolic representations merging and abstracting information over multiple modalities. The probability of these correlations are encoded in a Markov Logic Network.

The implementation of the approach outlined in this report is ongoing. Such pointers are crucial to capture relational structures between entities. Since these two rules admit a few exceptions profes- sors can be on sabbatical, and some undergraduates can teach as assistants , they are specified as soft constraints with finite weights w1 and w2. Improving the accuracy and efficiency of MAP inference for markov logic. Taskar, editors, Introduction to Statistical Relational Learning.

Pierre Lison Completes Doctoral Degree

Lifted first-order belief propagation. These beliefs models are spatio-temporally framed and include epistemic information for multi-agent settings. In gen- eral, each alternative value can be expressed as a propositional logical for- thseis. The belief i also specifies a belief history h.

If this new information conflicts with existing knowledge, the agent can decide to trigger a clarification request to resolve the conflict.

New Research Scientists have joined SAMBA | Norsk Regnesentral

They can also be used by perceptual components to adapt their internal processing operations to the current situated context contextual priming, anticipation, etc.

Spoken language interaction with model uncertainty: Association for Computational Linguistics.


We are using the Alchemy software 4 for efficient probabilistic inference. Besides the implementation, future work will focus on three aspects. Given the requirements of our application domain see Section 1and particularly the need to operate under soft real-time constraints, such approximation methods are an absolute necessity.

Of course, more complex spatio-temporal modelling can be designed.

Pierre Lison Completes Doctoral Degree – Department of Informatics

To be able to interact naturally with humans, robots needs to be aware of their own environment. For more information contact: The value of the feature fi x is 1 if Fi is true given x and 0 otherwise. This is a difficult problem, partly because of the noise and uncertainty contained in sensory data partial observabilityand partly be- cause the connection between low-level thesus and high-level symbols is typically thfsis to formalise in a general way [8].

Efficient weight learning for markov logic networks. Such pointers are crucial to capture relational structures between entities. The binder is directly connected to the other subarchi- tectures i. Socially Guided Machine Learning. Reference resolution is performed via a Markov Logic Network specifying the correlations pierrr the linguistic constraints of the referring expression and the belief thedis — in particular, the entity saliency and its associated cate- gorical knowledge.

Our approach departs from previous work such as [13] or [27] by introducing a much richer modelling of multi- modal beliefs. The ground Markov Liskn coun- tains two features one for each formula. Shared beliefs contain information which is part of the common ground for the group [3]. The similarity of a pair of beliefs is based on the correlation of their content and spatial frameplus other parameters such as the time distance between beliefs.


pierre lison thesis

Five central requirements can be formulated: Context in abductive interpretation. Sound and efficient inference with probabilistic and deterministic dependencies.

pierre lison thesis

We specify such information in the epistemic status of the belief. Due to the noise and uncertainty of sensory data, the perceived characteristics of the object are uncertain. Such rich annotation scheme allows us to easily interface beliefs with high-level cognitive functions such as action planning or communication. thesiss

Parameters such as recency have to be taken into account, in order to discard outdated observations. Markov Logic is used as a unified representation formalism, allowing us to capture both the complexity relational structure and uncertainty partial observability of typical HRI application domains. Figure 5 provides a graphical illustration of this process. As an il- lustration, assume a multi-modal belief B with a predicate Position B, loc expressing the positional coordinates of an entity, and assume the value loc can be estimated via distinct modalities a and b by way of two predicates Position a U, loc and Position b U, loc included in a percept union U.

pierre lison thesis