Scientific Support

Machine learning methodsis are nowadayswidely used inacross a wide variety ofmany scientific domains.fields. For example:

• Unsupervised machine learning methods mayare be used as toolsuseful for data exploration and visualisationvisualization (dimensiondimensionality reduction and clustering).
• Supervised machine learning methods may beare used asfor toolsprediction totasks, make predictions: givenwhere an output is inferred from input data point, predict the output (classification or regression).

At the DCSR, we cansupport researchers in several key areas of machine learning:

Training

We help you with the following aspects of machine learning.

Formation

We can help you to understand how a particularspecific machine learning methodmethods workswork and how itto mayapply be usedthem in your research. NoteWe thatalso the DCSR gives a fewoffers short introductory courses on machine learning; see ML courses.

Methodology

We can helpassist you toin chooseselecting and apply theapplying appropriate machine learning methods infor your research.

This may involveinclude:

a

• A pilotepilot phasephase, during whichwhere we collaboratively develop together codes and runtest themcode on your laptop or on the UNIL clusters.
Later
• A on,production phase, where we can also help youscale withand therefine productionyour phase.
  workflow.

More specifically, we cancan:

help

• Identify youexisting tools suited to find existing machine learning programs that would fit the aim of your analysis,analysis
• Help install and run them on your laptop or on the UNIL clusters,clusters
• Explain key parameters and explainsettings
the
• Develop meaning of their underlying option parameters. If no satisfactory machine learning program exists, we can help you to develop newcustom algorithms and writecode codesif thatno wouldsuitable fittools theexist
aim of your analysis.

Infrastructure

We can help you to implement efficiently your machine learning pipeline on the UNIL clusters. More specifically, we can help you to install your codes on the UNIL clusters and to profile them in order to find the optimal setting (in terms of RAM, number of nodes and CPU/GPU).

Foster UNIL collaborations

We can help you to find appropriate experts at the UNIL to discuss with you about specific machine learning related problems. 

A few examples:

1. An experimental scientist would like to learn how to analyse his/her data on his/her laptop or on the UNIL clusters by using machine learning methods. We can help this researcher to find appropriate machine learning tools, to understand how they work and to use them on his/her laptop or on the UNIL clusters.

2. A data scientist would like to implement a machine learning pipeline (on his/her laptop or on the UNIL clusters) but he/she is not sure how to do it properly. We can help this researcher to find and apply appropriate machine learning methods.

3. A data scientist has implemented a machine learning pipeline (on his/her laptop or on the UNIL clusters) and would like to discuss about the methodology he/she used. We can help this researcher to check whether his/her approach is correct and we can also suggest alternative approaches.

4. A data scientist has implemented a machine learning pipeline on his/her laptop and would like to run it on the UNIL clusters (going from laptop to cluster). We can help this researcher to implement efficiently his/her pipeline on the UNIL clusters. This may involve the installation of softwares and the profiling of codes.