AutoML concerns the ability of building machine learning (ML) frameworks in completely automated fashion (ie removing the human-in-the-loop) [1]. This is naturally interesting for the Data Science-based industry, where the process of delivering a DS-based product involves dozens of non-ML experiences.

Auto-guided software delivery product pipeline by automizing all the tasks as much as possible. The developed framework wants to be used for time series prediction (eg tool wear prediction for milling processes, stability prediction in NC-milling processes)

Highly motivated and goal-oriented;

Good algorithmic, mathematical and analytical skills;

Domain of the OOP paradigms and basic data structures;

Experience using one of the following scripting languages: R / Python

Background on advanced data analytics is valuable ... but not essential;

This problem concerns multiple sub-problems including:

1)         Automated feature extraction - the art of finding the best possible representation of your feature space for a target variable of interest using a specific supervised learning algorithm for online application [2];

2)         Hyperparameter Tuning - how to choose the best set of hyperparameters to use with a specific learning algorithm [1];

3)         Estimating Generalization Error - how to estimate generalization loss based on a training set for supervised learning problems [3];

4)         Feature Selection - Typically, feature selection procedures have 3 possible types: filters, embedded and wrappers. The goal is to be found in a unified procedure that works independently of the built predictive analytics.

5)         Model clustering and selection: create new approach for clustering ML models based on a designed performance criteria and perform online model selection.

 The student will be given a preference for one or two of these challenging subtopics on very specific contexts.

Thesis Goals:

• To domain the existing state-of-the-art on this specific topic;
• To acquire awareness on the different steps of a Knowledge Discovery process using a real-world industry problem
• To develop a fair understanding on the basic techniques behind this problem;
• To produce a framework able to produce accurate predictions in an automated fashion;

[1] Feurer, M., Klein, A., Eggensperger, K., Springenberg, J., Blum, M., & Hutter, F. (2015). Efficient and robust automated machine learning. In Advances in Neural Information Processing Systems (pp. 2962-2970).

[2] Guyon, I., & Elisseeff, A. (2006). An introduction to feature extraction. Feature extraction, 1-25.

[3] Arlot, S., & Celisse, A. (2010). A survey of cross-validation procedures for model selection. Statistics surveys, 4, 40-79.