Fundamental concepts of qualitative probabilistic networks
Artificial Intelligence
Lectures on modern convex optimization: analysis, algorithms, and engineering applications
Lectures on modern convex optimization: analysis, algorithms, and engineering applications
Credal Networks under Maximum Entropy
UAI '00 Proceedings of the 16th Conference on Uncertainty in Artificial Intelligence
Comprehensive Database for Facial Expression Analysis
FG '00 Proceedings of the Fourth IEEE International Conference on Automatic Face and Gesture Recognition 2000
Dynamic bayesian networks: representation, inference and learning
Dynamic bayesian networks: representation, inference and learning
Maximum of entropy for credal sets
International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems
Beyond Tracking: Modelling Activity and Understanding Behaviour
International Journal of Computer Vision
Facial Action Unit Recognition by Exploiting Their Dynamic and Semantic Relationships
IEEE Transactions on Pattern Analysis and Machine Intelligence
Learning Reliable Classifiers From Small or Incomplete Data Sets: The Naive Credal Classifier 2
The Journal of Machine Learning Research
Constrained Maximum Likelihood Learning of Bayesian Networks for Facial Action Recognition
ECCV '08 Proceedings of the 10th European Conference on Computer Vision: Part III
A theoretical framework for learning Bayesian networks with parameter inequality constraints
IJCAI'07 Proceedings of the 20th international joint conference on Artifical intelligence
Learning Bayesian network parameters under order constraints
International Journal of Approximate Reasoning
A tree augmented classifier based on Extreme Imprecise Dirichlet Model
International Journal of Approximate Reasoning
Hi-index | 0.00 |
This paper describes an Imprecise Dirichlet Model and the maximum entropy criterion to learn Bayesian network parameters under insufficient and incomplete data. The method is applied to two distinct recognition problems, namely, a facial action unit recognition and an activity recognition in video surveillance sequences. The model treats a wide range of constraints that can be specified by experts, and deals with incomplete data using an ad-hoc expectation-maximization procedure. It is also described how the same idea can be used to learn dynamic Bayesian networks. With synthetic data, we show that our proposal and widely used methods, such as the Bayesian maximum a posteriori, achieve similar accuracy. However, when real data come in place, our method performs better than the others, because it does not rely on a single prior distribution, which might be far from the best one.