4th Modelling Symposium:
Introducing Deep Neural Networks
Organized by Felix Ball, Nico Marek & Tömme Noesselt
Funded by ...
Overview
We are pleased to announce the 4th Modelling Symposium which provides once more a mix of theoretical and application-oriented analyses. The next symposium will cover Deep Neural Networks (DNNs) including basic introductions into DNNs, common building blocks, design patterns and architectures, best practices, optimization, applications etc. To this end, we welcome a new tutor -- Prof. Dr. Sebastian Stober.
Goal: Please note that DNNs are complex and that this course will help you to get started with DNN analyses. The workshop provides a general introduction into DNNs covering a
wide range of topics. After the 4 days you should have an overview of different DNNs, their strength and weaknesses and which parameters of the model might be important and which ones you might
have to tweak. The course will also help you to make decisions about which information/parameter can be important in steps XY and it also helps you to better understand the DNN literature (e.g.
whether author's omitted important information about the presented models).
When & Where
When
- 27.07.2020 - 31.07.2020
Dinner
- 28.07.2020; Start 7 pm
- We will dine at the Hoflieferant. Please register for the dinner so we can book a table. The dinner is optional and self-paid.
Wednesday off!
- There has to be some time to digest!
Location
- Universitätsplatz campus, Gebäude 28, room 27
Detailed Program (subject to changes)
1st half of week
Monday
(Basics and CNNs)
09.00 - 10.30: General introduction (machine learning basics)
Break
11.00 - 12.30: Convolutional Neural Networks I (Basics)
Break
14.00 - 15.30: Convolutional Neural Networks II (Hands-on)
Break
16.00 - 17.30: Convolutional Neural Networks III (Advanced)
Break
17.40 - 18.30: OPTIONAL - Discussing your data models
Tuesday
(common building blocks, design patterns and architectures)
09.00 - 10.30: Recurrent Neural Networks I (Basics)
Break
11.00 - 12.30: Recurrent Neural Networks I (Hands-on)
Break
14.00 - 15.30: Attention mechanisms
Break
16.00 - 17.30: Transformers
Break
17.40 - 18.30: OPTIONAL - Discussing your data models
2nd half of week
Thursday
(best practices [BP], optimization and introspection)
09.00 - 10.30: Best practices, optimisation and regularization
techniques I (Basics)
Break
11.00 - 12.30: Best practices, optimisation and regularization
techniques II (Hands-on))
Break
14.00 - 15.30: Introspection I (Basics)
Break
16.00 - 17.30: Introspection II (Hands-on)
Break
17.40 - 18.30: OPTIONAL - Discussing your data models
Friday
(Applications, transfer learning and sneak peek)
09.00 - 10.30: Present your data
Break
11.00 - 12.30: Possible applications (EEG and fMRI)
Break
14.00 - 15.30: Model compression and transfer learning
Break
16.00 - 17.30: Sneak peek and summary
Software, Code, Equipment, & Requirements
All information will be regularely updated, so please check for updates!
Software: Hands-on sessions will be based on Python and Tensorflow.
Code & Equipment: The code will be provided during the symposium. We will use a computation cluster so you do not have to worry about software and installation. All you need is a laptop. We will also provide power sockets. In case you are registering for the "data talk sessions", please also bring a VGA and HDMI adapter for your presentation.
Requirements: The hands-on sessions require that you have general coding skills and are not an absolut beginner. You should have already written pieces of code, maybe a data analysis or e.g. an experiment. You should know Python and also Numpy, you should know what loops and conditions are, different types of variables, n-dimensional arrays, what a function is etc. Please note that we do not have time to cover basic programming.
Literature: This course will cover a variety of topics related to DNNs. To enhance your experience and avoid being overwhelmed (e.g. in case, you have never heard about DNNs before), you should consider reading about DNNs beforehand. Here are suggestions for starting with DNNs and computational models (more might follow):
Paper and books
Storrs & Kriegeskorte; Kriegeskorte & Douglas; Cichy & Kaiser; Goodfellow, Bengio & Courville
Videos
TensorFlow and DNNs without PhD
Speaker: Prof. Dr. Sebastian Stober
Foto Quelle: Jana Dünnhaupt / Universität Magdeburg
Sebastian Stober is an interdisciplinary researcher with a PhD in computer science and a background in (applied) machine learning, (music) information retrieval and cognitive neuroscience. He is especially interested in so-called “human-in-the-loop” scenarios, in which both humans and machines learn from each other and together contribute to the solution of a problem. Since October 2018, he is Professor for Artificial Intelligence at the Otto-von-Guericke-University Magdeburg. Before, he was head of a new junior research group on Machine Learning in Cognitive Science at the University of Potsdam and from 2013 to 2015, he was post-doctoral fellow in the labs of Adrian Owen and Jessica Grahn at the Brain and Mind Institute at Western University in London, Ontario.
Registration
Will start end 2019/begin 2020
Restricted number of attendees
As we only have one tutor this year, the number of attendees is limited to 25-30 people.
Costs
To cover the costs for the Symposium, we'll have a mandatory fee per attendee. Although, we provide a 4-day Symposium, we try to keep the costs as minimal as possible. As we are sponsored by the "Neurowissenschaftliche Gesellschaft" (NWG), the fee depends on your affiliation:
NWG member : workshop is free of charge
Non-NWG member : 125 €
Furthermore, we provide the option to have a full 4-day catering during the Symposium. Please indicate your choice in the registration form. In case you are indecisive: all attendees of the last symposia were delighted with our catering choice and the food quality (we chose Yodett as catering company; see also last symposia for images and impressions). The presence of catering is dependent on the number of people deciding for this option.
Catering (optional) : 100 € (4-days)
You also have the option to attend the social event (dinner on Tuesday). Please note that this event is self-paid and requires the transfer of a 30 € security deposit. It will either be re-funded during the dinner or -- in case you are not cancelling in time or spontanously not attending -- used to pay the restaurant for their loss.
Dinner (optional) : self-payed (30 € deposit)
In case you are a registered attendee: Please note that we cannot refund the fee/catering costs. Refund depends on whether we find someone to take your spot (or finding sponsors for catering). Refunds of the dinner deposit depend on the time-point of cancellation.
By registering, you agree to the abovementioned terms.
Present your data
Every year, we provide the option to discuss your data/ideas with our tutor and how the individual methods presented during the respective year could be applied to them. You will have 20-30 min in the evening to discuss with our tutor and on Friday morning you will be presenting the outcome of this discussion to the crowd.This will be a short presentation of 5-8 min maximum, which is followed by a question/discussion round.
Poster presentations are also welcome (again indicate below). We will provide poster boards and also advertise your posters on the first day. Poster will be presented throughout the week (typically during breaks - on demand).
Motivational statement
Please provide a short and concise motivational statement (max. 300 words). This statement will be part of the selection process. We will make a decision by begin/mid of April 2020 and
confirm your attendance by the 15th of April latest.
Register for symposium and dinner
We will add the registration End of 2019/begin of 2020
Download
What
Download
Last update
Contact
events.biopsych(at)ovgu.de