In Machine Learning from Data, the wide variety of methods that exists is still reliant on handcrafting by human decision maker, expert or programmer. For example, even in such methods that are named unsupervised a human made decision is required upfront in regards to: i) feature selection; ii) type of data distribution; iii) number of clusters or thresholds, radii, etc. iv) data (in)dependence; v) type of distance metric, etc. The goal of this competition (and Autonomous Leaning Machines, in general) is to increase the level of autonomy of the learning algorithms, whereby the ideal is to have methods and algorithms that require no human involvement at all.
In this competition, the success will be judged by how close an algorithm and method is to this goal and not by how much the accuracy, purity or other measure of quality is achieved; how much time or complexity of the algorithms require. Of course, these parameters of the quality of the algorithms and methods will also be taken into account, but as a secondary ones, not the main (precisely, the opposite to the current practice).
This competition will consist of two different tracks:
1. New data sets and streams that are suitable demonstrators for the topic of the competition.
2. New methods and algorithms for Autonomous Learning Machines in:
- Anomaly Detecion,
Runner Up: $1000
The proposed methods and algorithms must be described in form of papers and submitted to the ALMA Competition track of IEEE EAIS 2017 and presented at the conference.
The software must be uploaded to the web portal using GNU license.