Machine Learning Applied to Physics and Astronomy
Our research laboratory intends to provide state of the art techniques in the analysis of scientific data, with a particular emphasis in the fields of physics and astronomy. The laboratory comes in response to a time where massive amounts of data are being collected by a great variety of sensor devices (e.g., telescopes and satellites capturing a huge amount of images about our solar system and the entire universe). Such amount of data, readily available for processing and analysis, calls for algorithms that can search for meaningful patterns in an efficient form. Our current projects are the following:
- Domain Adaptation Techniques Applied to Cepheid Variable Star Classification.
- Automatic Classification of Supernovae.
- Cluster Validation Applied to Current Star Taxonomies.
- Learning to Classify Transients Across Astronomical Surveys.
Visit the Pattern Analysis Lab website.
Metalearning: Data Characterization and Adaptive Learning Systems
In this project we try to understand how meta-learning allows machine learning systems to benefit from their repetitive application. If a learning system fails to perform efficiently, one would expect the learning mechanism itself to adapt in case the same task is presented again. Metalearning differs from base-learning in the scope of the level of adaptation; whereas learning at the base-level is focused on accumulating experience on a specific task, learning at the metalevel is concerned with accumulating experience on the performance of multiple applications of a learning system. Briefly stated, the field of metalearning is focused on the relation between tasks or domains, and learning algorithms. Rather than starting afresh on each new task, metalearning facilitates evaluation and comparison of learning algorithms on many different previous tasks, establishes benefits and disadvantages, and then recommends the learning algorithm, or combination of algorithms, that maximizes some utility function on the new task. The utility or usefulness of a given learning algorithm is often determined through a mapping between a characterization of the task and the algorithm estimated performance.