Home » Deep Neural Networks for Beginners – 2019

Workshop: “Deep Neural Networks for Beginners”

Dates: 7-9 October 2019

Location: SeRC/Department of Meteorology, Stockholm University, lecture hall TBA

Deadline for applications: 15.08.2019

Target Audience:

PhD students from the SeRC stakeholder organizations: SU, KTH, LU, KI who are interested in applied machine learning (ML), Artificial Intelligence (AI) and interdisciplinary research, and who would like to familiarize themselves with the basic ML/AI concepts and methods. Master students, postdocs and faculty are also welcome to apply although the priority will be given to the PhD students. We expect that course participants from a variety of scientific fields: Mathematics, Computer Science, Geophysics, Climate Science, Chemistry, Biology and Medicine, will find it fruitful to attend the course and learn about the methods and applications of ML and AI which are becoming increasingly common in their research areas, and to apply the newly gained knowledge in practical sessions.

Maximum number of participants: 25

Minimal requirements:

  • Basic programming skills (experience in Python is helpful, for the practical exercises we will organize the students in 2-3 person groups to optimize the programming experience)
  • Basic knowledge in data analysis, statistics, and mathematics

Motivation:

The need for methods to automatically, objectively, and efficiently analyze and interpret data is a common task in many scientific areas. Machine learning is an essential part of this task, where especially deep learning plays a central role. Advances in the field of deep learning have led to a development in machine learning methods such as deep neural networks, which outperform classical methods in many fields. One of the main reasons of their success is the ability to uncover hidden and complex structures in the data, where layered architectures are employed to extract a deep and rich hierarchical feature representation, which is particularly suitable for solving certain tasks. Several applied scientific communities such as the remote sensing community started to use deep learning approaches for their application tasks including the identification of objects and forecasting of bio- and geophysical parameters. This illustrates an increasing demand for interdisciplinary approaches that bridge the gap between machine learning and disciplines such as natural sciences. The global scope of this course is to lay the foundations in machine learning and provide necessary deep learning tools in the context of applied sciences. In detail, it includes lectures about fundamental and advanced concepts in neural networks and deep learning, which will be presented with allocated time for discussions. The gained knowledge will be applied in three hands-on sessions covering various practical aspects. The sessions will cover all necessary aspects of machine learning pipelines that work on real world applications, covering data pre-processing, model learning and testing, as well as quantitative and qualitative evaluation.

The proposed course builds upon a similar course offered at the GEOMAR Helmholtz Centre for Ocean Research Kiel, Germany, in February 2019 which was very well received.

Objectives:

  • Understanding the basics in classification and regression
  • Understanding the basics in neural networks
  • Application of neural networks in hands-on sessions (Python programming sessions)
  • Usage of Keras and Tensorflow to train and build deep learning models for applied sciences
  • Analysis and evaluation of results obtained in hands-one sessions
  • Participation in interactive discussion
  • Presentation of results
  • A course evaluation will be carried out during the week following the course to collect the feedback

Covered topics:

  • Basics in machine learning: classification + regression (learning, testing, evaluation)
  • Challenges in machine learning
  • Classification paradigms and classification tasks
  • Representation learning
  • Basics in neural networks and deep learning
  • Backpropagation
  • Cross-validation
  • Convolutional neural networks, recurrent neural networks, Long-short-term memory networks

Computational setup:

The computational resources will be provided by Hops (https://www.hops.io/), a SeRC platform for ML and Deep Learning.

Time schedule:

Day 1

  • 4 x 45 min: lecture ‘Introduction to neural networks’
  • 3 x 45 min: hands-on ‘Neural network regression’

Day 2

  •  4 x 45 min: lecture ‘Deep neural networks for applied sciences’
  • 3 x 45 min: hands-on ‘Time-series forecasting’

Day 3

  • 2 x 45 min: hands-on ‘Object identification’
  • 3 x 45 min: time for individual discussions of the exercises with course participants

Course coordinator:

Ass. Prof. Inga Koszalka (inga.koszalka@misu.su.se), Department of Meteorology, Stockholm University

Teachers:

Prof. Ribana Roscher, Institute of Computer Science, University of Osnabrueck, Germany

Application: 

Deadline for application: 15.08.2019

Given name(s) *                                              Family name *

University/Affiliation *

Email * - (You must use your university email address, not gmail, yahoo, hotmail etc.)

Supervisor *

Subject of the PhD project

Notes to the organizers