Machine Learning

MACHINE LEARNING IN SUPPORT OF COMPUTATIONAL AND THEORETICAL SCIENCES

The goal of the programme is to strengthen machine learning (ML) collaboration in NITheCS in two ways:

  • ML research: development of new, specialised machine learning techniques.
  • ML as a tool: applying machine learning for scientific modelling applications.
Machine learning research

Collaborating Researchers:

PI – Principal investigator (programme manager) 

Projects

Knowledge Discovery in Time Series Data
Machine learning techniques play an increasing role in assisting scientists and engineers in knowledge discovery: obtaining novel information from large, possibly complex data sets. Many practically important tasks, such as weather prediction, financial forecasting, or speech processing, are modelled using time series information. Knowledge discovery in time series data is an active field of research, with techniques such as feature attribution used to gain new insights into the underlying processes being modelled.

Machine Learning Forum
The programme aims to grow a forum for cross-cutting projects executed in other NITheCS focus areas, where those projects rely on machine learning expertise.

Membership

The programme is open to all NITheCS Associates who currently work in machine learning, or are interested in ML approaches in their research.

Events – 2022

March
The first workshop of the NITheCS Machine Learning programme took place virtually on 10 March 2022. The PIs introduced the project and its two streams, and participants shared their current ML-related work.

September
Stefan Lotz hosted a NITheCS colloquium on 19 September 2022.
Abstract  |  Video  |  Slides

December
NITheCS ML @ SACAIR: A one-day workshop will be hosted at the SACAIR 2022 conference on 6 December 2022: “Knowledge Discovery in Time Series Data”. The workshop will bring together interested researchers in this field, to share their recent findings and views. Work in progress is welcome. For more information about submissions and/or attendance, visit the conference website

Publications

Interpretability of deep learning models
Jacques P. Beukes, Stefan Lotz and Marelie H. Davel, “Pairwise networks for feature ranking of a geomagnetic storm model”. In: South African Computer Journal 32.2 (2020). doi: 10.18489/sacj.v32i2.860.

Coenraad Mouton and Marelie H. Davel, “Exploring layerwise decision making in DNNs”, in Artificial Intelligence Research (SACAIR 2021), Communications in Computer and Information Science, Vol 1551, pp 140-155, January 2022.

Modelling time series data
Walter Heymans, Marelie H. Davel and Charl van Heerden, “Efficient acoustic feature transformation in mismatched environments using a Guided-GAN“, in Speech Communication, Vol 143, pp10-20, September 2022.

Marko Oosthuizen, Alwyn Hoffman and Marelie H. Davel, “A comparative study of graph neural network speed prediction during periods of congestion”, in Proc. 14th Int. Conf. on Neural Computation Theory and Applications (NCTA 2022), October 2022

Thipe Modipa and Marelie H. Davel, “Two Sepedi‑English code‑switched speech corpora“, Language Resources and Evaluation, Vol 56, issue 3, pp703-727, September 2022.

Edrich Fourie, Jaco Versfeld, and Marelie H. Davel. “Neural speech processing for whale call detection”. In Proc. Southern African Conference for AI Research (SACAIR), Dec 2022.

Walter Heymans, Marelie H Davel, and Charl van Heerden. “Multi-style training for South African call centre audio“. In Artificial Intelligence Research, Communications in Computer and Information Science, volume 1551, pages 111–124. Springer, 2022.

Kimara Naicker, Ilya Sinayskiy, and Francesco Petruccione. “Machine learning for excitation energy transfer dynamics“. Physical Review Research, 4(3):033175, 2022.

Kimara Naicker, Ilya Sinayskiy, and Francesco Petruccione. “Statistical and machine learning models for prediction of long-time excitation energy transfer dynamics“. arXiv preprint, 2022.

Other applications of machine learning
Andrew Oosthuizen, Marelie H. Davel and Albert Helberg, “Multi-Layer Perceptron for Channel State Information Estimation: Design Considerations “, in Proc. Southern Africa Telecommunication Networks and Applications Conference (SATNAC 2022), pp 94-99, August 2022.

Marthinus W. Theunissen, Coenraad Mouton and Marelie H. Davel, “The Missing Margin: How Sample Corruption Affects Distance to the Boundary in ANNs”, in Proc. Southern African Conference for AI Research (SACAIR 2022).

Andrew J. Oosthuizen, Albert S.J. Helberg and Marelie H. Davel, “Adversarial Training for Channel State Information Estimation in LTE Multi-Antenna Systems”, in Proc. Southern African Conference for AI Research (SACAIR 2022).