Skip to main content

Advanced Machine Learning

2020/2021
Programme:
Financial Mathematics, Second cycle
Year:
1 ali 2 year
Semester:
first or second
Kind:
optional
Group:
O
ECTS:
6
Language:
slovenian, english
Lecturer (contact person):
Hours per week – 1. or 2. semester:
Lectures
2
Seminar
1
Tutorial
2
Lab
0
Prerequisites

There are no prerequisites.

Content (Syllabus outline)

Comparing the performance of machine learning algorithms on multiple data sets: frequentist and Bayesian approach.

Learning from data streams: online learning, evaluating model performance on data streams, change detection mechanisms, composing algorithms for machine learning from data streams.

Meta learning: no-free lunch theorem for machine learning, learning about learning, attribute representation of data sets, parametrization of learning algorithms, optimizing the parameter settings of learning algorithms, surrogate models.

Handling background knowledge in machine learning: equation discovery from data and knowledge, relational learning and surrogate models, hierarchically structured background knowledge (taxonomies), background knowledge and (deep) artificial neural networks.

Selected topics in deep learning: handling different objective functions and back propagation, special topologies of deep neural networks (autoencoders, embeddings of unstructured and semi-structured data), semi-supervised learning.

Readings

Hastie T, Tibshirani R, Friedman J (2009) The Elements of Statistical Learning (2nd edition). New York: Springer-Verlag.

Flach P (2012) Machine learning: the art and science of algorithms that make sense of data. Cambridge: Cambridge University Press.

De Raedt L (2008) Logical and Relational Learning. Berlin: Springer-Verlag.

Predavatelj poleg tega izbere tudi primerne novejše raziskovalne članke iz znanstvenih revij.

Objectives and competences

Students master advanced machine learning methods, such as, machine learning from data streams, meta learning and automatic configuration of learning algorithms, knowledge-intensive learning and learning models of dynamical systems. Students through seminars and homework apply the mastered knowledge on various tasks of upgrading existing algorithms and building predictive models from data and formalized knowledge.

Intended learning outcomes

Knowledge and understanding: Understanding concepts and components of the machine learning algorithms.

Application: Applying existing algorithms and tailoring/upgrading algorithms for solving practical problems in various scientific and engineering fields.

Reflection: Critical insight into the inner workings of the machine learning algorithms and identifying opportunities for their improvement, formal representation of practical problems that allow for solutions based on machine learning.

Transferable skills: Ability to identify, formulate and solve practical problems. Ability to design predictive models with machine learning algorithms. Critical assessment of scientific literature.

Learning and teaching methods

lectures, seminars, excercises, homework and consultations

Assessment

Homework

Oral exam

grading: 5 (fail), 6-10 (pass) (according to the Statute of UL)

Lecturer's references

Ljupčo TODOROVSKI:

KUZMANOVSKI, Vladimir, TODOROVSKI, Ljupčo, DŽEROSKI, Sašo. Extensive evaluation of the generalized relevance network approach to inferring gene regulatory networks. GigaScience, ISSN 2047-217X, [in press] 2018, 21 str., doi: 10.1093/gigascience/giy118.

LUKŠIČ, Žiga, TANEVSKI, Jovan, DŽEROSKI, Sašo, TODOROVSKI, Ljupčo. General meta-model framework for surrogate-based numerical optimization. V: YAMAMOTO, Akihiro (ur.). Discovery science : 20th International Conference, DS 2017, Kyoto, Japan, October 15-17, 2017 : proceedings, (Lecture notes in artificial intelligence, ISSN 0302-9743, LNAI 10558). Cham: Springer. 2017, lNAI 10558, str. 51-66.

ŠEMROV, Darja, MARSETIČ, Rok, ŽURA, Marijan, TODOROVSKI, Ljupčo, SRDIČ, Aleksander. Reinforcement learning approach for train rescheduling on a single-track railway. Transportation research. Part B, Methodological, ISSN 0191-2615. [Print ed.], 2016, letn. 86, št. apr., str. 250-267, ilustr., doi: 10.1016/j.trb.2016.01.004.

SIMIDJIEVSKI, Nikola, TODOROVSKI, Ljupčo, DŽEROSKI, Sašo. Predicting long-term population dynamics with bagging and boosting of process-based models. Expert systems with applications, ISSN 0957-4174. [Print ed.], 2015, vol. 42, no. 22, str. 8484-8496, doi: 10.1016/j.eswa.2015.07.004.

Matija PERTNAR:

– PLOTKIN, Gordon, PRETNAR, Matija. Handling algebraic effects. Logical methods in computer science, ISSN 1860-5974, 2013, vol. 9, iss. 4, paper 23 (str. 1-36) [COBISS-SI-ID 16816729]

– PRETNAR, Matija. Inferring algebraic effects. Logical methods in computer science, ISSN 1860-5974, 2014, vol. 10, iss. 3, paper 21 (str. 1-43) [COBISS-SI-ID 17190745]

– BAUER, Andrej, PRETNAR, Matija. An effect system for algebraic effects and handlers. Logical methods in computer science, ISSN 1860-5974, 2014, vol. 10, iss. 4, paper 9 (str. 1-29). http://arxiv.org/pdf/1306.6316 [COBISS-SI-ID 17191001]