Aim and Scope
Autonomous learning approaches aim at the acquisition of knowledge from raw data for analysis, interpretation and to develop reliable autonomous systems that make accurate decisions and predictions for future data. To ensure such reliability of decision based only on the raw data, there is therefore an important need to understand the processes generating the data. This in general leads us to generative learning approaches which will be in the core of this session. Generative model-based approaches are very useful well-established statistical models that explicit the processes generating the data. Such approaches are naturally tailored for an exploratory data analysis through unsupervised learning as they learn the conditional jointly with the prior, the posterior being taken with the Bayes rule. In particular, latent data models, including mixtures and hidden Markov models with the EM algorithms are at the basis of the majority of developed model-based approaches for unsupervised learning. As these well-established models are likely to be very beneficial in many domains, there is namely a growing investigation of adapting them to the context of functional data and in high dimensionality problems, as well as for large-scale data sets through online implementations. Such models have proved their efficiency in many applications domains including signals, text, speech, image, etc.
This session will therefore be dedicated to new theoretical propositions concerning unsupervised generative learning approaches that investigate vectorial high dimensional data, as well as for functional data when the inputs are functions rather than finite size vectors. The frameworks will in particular concern data representation, classification/clustering and dimensionality reduction. Articles will cover these approaches when the data are taken under an independence assumption, as well as in a sequential analysis context.
Topics of the special session include but not limited to:
- Unsupervised Generative Learning for data representation/classification
- Model-based clustering for high dimensional/functional data
- Curve clustering
- Latent data Models (Mixtures, Hidden Markov Models, …)
- (Online) EM algorithms
- Sequential data modelling
- Generative Topographic Mapping
- Temporal Generative Topographic Mapping
- Probabilistic Self Organizing Maps
- Unsupervised Energy-Based Learning:
- Restricted Boltzmann Machines, Deep Belief Networks
- Applications in signal, speech and image processing; handwritten text recognition; human activity recognition; Diagnosis of complex systems; Content based information retrieval; Gene expression data; etc