Sissejuhatus arvutuslikku neuroteadusesse 2013/14 kevad

Topics & Deadlines

Aim of the project is to analyse real neuroscientific data in order to prove or disprove some theory about inner workings of the brain, find new interesting patterns, which might lead to a new theory, come up with cool visualizations of the data, apply new method or algorithm, etc.

Work will be done in small (1-3) groups. The final result should include:

• The report with ~10 pages of meaningful text.
• The code to produce the results.

Project gives up to 30 points of your final grade. The main grade component of the project comes from the effort you put in.

1. Your own project

• Team 1: Maksims Ivanovs "Perception of visual illusions"
• Team 2: Tambet Matiisen "Solving 2048-game using Q-learning and RBM"
• Team 3: Kristjan Korjus, Taivo Pungas, Ilya Kuzovkin "Replicating DeepMind"

Come up with an idea of a project related to the field of computational neuroscience and execute it. Before you start working consult with us if your idea is suitable.

• Supervisor: You

2. Image reconstruction from fMRI data

• Team 1: Martin Loginov
• Team 2: Üllar Lindmaa

During the exercise session 4 we have seen fMRI data recorded while test subject were looking at pictures. The task of reconstruction is hard, but fascinating one. The aim of this project would be to read papers, try their algorithms or come up with something on your own and do the reconstruction as good as you can.

• Supervisor: Ilya Kuzovkin
• Resources: https://crcns.org/data-sets/vc/vim-1/about-vim-1
• Difficulty: hard

3. Reproduce MindFlex game using Emotiv EPOC
There is a funny toy, called MindFlex. We have three Emotiv EPOC devices in our lab. The aim of this project is to create a game for two (or three) players which will measure their brain signals and do something depending on that.

• Supervisor: Ilya Kuzovkin
• Resources: http://www.mindflexgames.com, http://www.emotiv.com
• Difficulty: medium (but requires coding skills)

4. Emotiv EPOC
*Team 1: Uku Loskit "Signal quality and data visualizer for Emotiv EPOC" We have small EEG machine in our lab, called Emotiv EPOC. Take the device, propose some hypothesis (for example: the signal will be different when test subject moves his right hand from the signal when he moves his left hand), collect data and perform the analysis.

• Supervisor: Ilya Kuzovkin
• Resources: http://www.emotiv.com
• Difficulty: easy

5. Trial-to-trial variability of monkey spiking data

• Team 1: Marek Oja, Andre Tättar

Monkey is performing a memory task and spikes of lots of neurons are recorded. The goal is to find out trial to trial variability of different neurons and different time phases. The results of the projects will be useful for our research group.

• Supervisor: Kristjan Korjus
• Resources: Few slides to grasp the concept
• Difficulty: medium

6. Model the neurons to produce behaviour as in the practice session 3
One idea of the modelling is to create a model, which can reproduce actual real-life data. Your task is to produce a model, which will generate data similar to what we had in the third session and apply session 3 analysis pipeline to the data you have generated.

• Supervisor: Kristjan Korjus
• Resources: https://courses.cs.ut.ee/2014/neuro/spring/Main/DataAnalysisSpiking
• Difficulty: easy

7. INSTINCT: The IARPA Trustworthiness Challenge
Predict whether person is trustworthy or not based on neurophysiological recording done during while test subjects were playing a trust game.

• Supervisor: Ilya Kuzovkin
• Resources: http://www.iarpa.gov/INSTINCT
• Difficulty: hard

8. Reconstruct the wiring between neurons from fluorescence imaging of neural activity Challenge at kaggle.com to reconstruct wiring between neurons.

• Supervisor: Ilya Kuzovkin
• Resources: http://www.kaggle.com/c/connectomics
• Difficulty: hard

9. Is there a purpose for a synaptic failure
The project consists in formulating a computational project on the subject of "synaptic failure"- the fact that in up to 50% of cases a presynaptic signal is not transmitted to the postsynaptic neuron. The goal of the project is to:
1) propose hypothesis why such failure has been selected for during evolution? Is it in some ways metabolically or computationally more efficient?
2) propose a computational model to study the effect of synaptic failure on information transmission.

• Supervisor: Ardi Tampuu
• Difficulty: hard

10. Essay: Can they read your dreams even if you don't want them to?
Recent research has shown that dreams can be read out directly from the neural activity patterns of human subjects. In this essay you should explain in detail what the group of Kamitani did to decode dreams. You should also discuss the limitations of their present approach - could your own dreams be read out with this methodology? If not, why not? Could you envision ways how the dream-reading methods could be improved in the future?

• Supervisor: Jaan Aru
• Resources: http://zoology.ou.edu/pdf_documents/Neuromunch/Horikawa_et_al_2013.pdf
• Difficulty: easy (take this if you are completely scared of writing some code)

11. Modelling: Flash lag effect

• Team 1: Roberts Mencis

In the flash lag effect a stimulus that is flashed briefly is perceived as lagging a stimulus that is moving although in reality they are exactly at the same spot. What could be a principal neural mechanism behind this perceptual effect? You could put your idea in a simple neural model that tries to capture this effect.

• Supervisor: Jaan Aru
• Resources: http://michaelbach.de/ot/mot-flashLag/
• Difficulty: medium

12. Modelling: The Troxler effect

• Team 1: Anastasia Bolotnikova, Anton Prokopov, Mattias Nurk

In the Troxler effect a peripheral stimulus disappears from our perception during fixation although in reality the stimulus is still on the screen. What leads to the disappearance from your perception? Can you come up with a simple model that explains the phenomenon?

• Supervisor: Jaan Aru
• Resources: http://en.wikipedia.org/wiki/Troxler%27s_fading
• Difficulty: medium

13. Hodgkin-Huxley model simulator
This project will revolve around the simulation of the Hodgkin and Huxley model of neuron under many different conditions. In particular, using the graphical simulator HHsim, the student will complete a list of exercises oriented to explore the effects of stimulation, temperature or ion channels in the dynamics of a neuron.

• Supervisor: Raul Vicente
• Resources: http://www.cs.cmu.edu/%7Edst/HHsim/guide.html
• Difficulty: easy

14. Build an automated EEG preprocessing algorithm

• Team 1: Kristjan-Julius Laak

A tedious phase of working with EEG involves cleaning the recording of different types of non-brain-related activity (i.e. measurement artefacts). This process requires choosing methods to correct or remove them as well as setting parameters for those methods that would assure a desirable mix of false positives and false negatives. In addition, different subjects may need slightly different treatment. Your goal is to build a novel algorithm that could automate some aspects of this phase. You could either concentrate on the big picture and try to combine existing techniques into one automatic pipeline. Or you could build solutions for a specific type of artefacts (e.g. reliable detectors horizontal eye movements). It is advisable you learn how to simulate artifactual data so that you have a benchmark for your algorithm.

• Supervisor: Andero Uusberg
• Difficulty: medium

15. fNIRS and finger tapping
The task is to work with fNIRS data collected during finger tapping experiment from different subjects and analyze the hemodynamic response in motor cortex. Processing stage will involve some data cleaning and removing the artifacts. In theory, fNIRS should be very insensitive to motion and it would be interesting to see, how motion artifacts could be detected and removed from the data. In addition, it would be interesting to apply source localization to the fNIRS signal (across channels), defining the regions of interests etc. This analysis can be done for both oxy-Hb and Deoxy-Hb measures, which could be of conceptual interest of comparison.

• Supervisor: Maria Tamm
• Difficulty: hard