Copyright © Division of Cognitive Neuroscience
Teaching
MACHINE LEARNING MEETS HUMAN LEARNING (Spring 2013)
Gedi Luksys
Overview
Machine learning has recently emerged as one of the most successful and practically applicable fields of artificial intelligence: its algorithms are used in search engines, image and text recognition, medical systems, financial markets, and more recently in cognitive neuroscience and genetic studies. This seminar will cover fundamentals of machine learning and review its most important techniques. The main emphasis will be on learning how to use techniques using standard software packages (such as Matlab) and discussing their applications to cognitive neuroscience and behavioral genetics.
Structure
The seminar will take place on Wednesdays at 16:15-17:45, from February 27 till May 22 in Nebenhaus (64a) Seminarraum 4.
Each session will introduce a theoretical basis and applications for the discussed method. Paper presentations by students will take place on March 27 for the supervised learning topics and as a part of classes for unsupervised and reinforcement learning topics. Please select your chosen paper here.
Tentative schedule, topics, and papers to be presented
February 27. Introduction to the seminar, overview of machine learning approaches, math basics
March 6. Fundamentals of supervised learning, bias-variance dilemma
March 13. Artificial neural networks
March 20. Support vector machines
March 27. Applications of supervised learning. Papers:
- Kamitani & Tong, "Decoding the visual and subjective contents of the human brain", Nature Neuroscience (2005) - (pdf, 12 pts)
- Johnson et al., "Recollection, Familiarity, and Cortical Reinstatement: A Multivoxel Pattern Analysis", Neuron (2009) - (pdf, 14 pts)
- Dosenbach et al., "Prediction of Individual Brain Maturity Using fMRI", Science (2010) - (pdf, 15 pts)
- Knudsen, "Supervised Learning in the Brain", Journal of Neuroscience (1994) - (pdf, 15 pts)
April 3. Introduction to unsupervised learning, k-means clustering, mixture models
Paper: Pulvermüller et al., "Distributed cell assemblies for general lexical and category-specific semantic processing as revealed by fMRI cluster analysis", Human Brain Mapping (2009) - (pdf, 12 pts)
April 10. Probabilistic approaches in machine learning
Paper: Miyawaki et al., "Visual Image Reconstruction from Human Brain Activity using a Combination of Multiscale Local Image Decoders", Neuron (2008) - (pdf, 16 pts)
April 17. Principal component analysis. Papers:
- Clément et al., "Anxiety in Mice: A Principal Component Analysis Study", Neural Plasticity (2007) - (pdf, 10 pts)
- Novembre et al., "Genes mirror geography within Europe", Nature (2008) - (pdf, 10 pts)
April 24. Independent component analysis
Paper: Mueller et al., "Dopamine effects on human error processing depend on catechol-O-methyltransferase VAL158MET genotype", Journal of Neuroscience (2011) - (pdf, 12 pts)
May 15. Introduction to reinforcement learning
Paper: Walsh & Anderson, "Learning from delayed feedback: neural responses in temporal credit assignment", Cognitive, Affective & Behavioral Neuroscience (2011) - (pdf, 14 pts)
May 22. Understanding learning and decision making through model-based analyses
Paper: Schweighofer et al., "Low-Serotonin Levels Increase Delayed Reward Discounting in Humans", Journal of Neuroscience (2008) - (pdf + erratum, 10 pts)
Evaluation
As this seminar has no exam, active participation is essential for a successful completion of the course. This consists of attending classes, presenting one or several papers, and solving several short weekly exercises.
Your performance will be evaluated based on attendance, exercises and paper presentation(s) as follows:
- By presenting a paper you can earn up to 10-16 points, depending on the length & difficulty of the paper and the quality of your presentation.
- For attendance you earn 1 point per class (plus 1 "regular attendance" bonus point, if you miss no more than twice).
- By solving short exercises (due next class) and/or participating actively in class you can earn an additional 1 point each time.
- For 4 credits, you may also earn up to 20-25 points by writing an essay (Referat) about applications of machine learning to a field of your choice (e.g. neuroimaging, neuroprosthetics, behavioral genetics) or by completing a miniproject where you would apply one or more machine learning methods to a problem of your choice (for the latter, some proficiency with Matlab or other package you would use for the analysis is necessary. I could help you with the miniproject, but I would not teach you Matlab/programming basics)
For a pass with 2 credits you need to earn at least 24 pts, which can be achieved by regular attendance + 1 paper presentation.
For 4 credits, you need 38-58 pts. Your grade will then be points / 10, rounded to the nearest half.
THE PREDICTIVE BRAIN: COMPUTATIONAL MODELS OF LEARNING, MEMORY AND DECISION MAKING (Autumn 2012)
Gedi Luksys
Overview
Learning and memory allows humans and animals to become more capable of making decisions that maximize their evolutionary success. During the last 10-15 years it has become popular to interpret neural and behavioural data using computational models that provide more detailed insights into these cognitive processes. In this seminar we will review recent experimental and computational studies of reinforcement learning, associative memory, and decision making. We will also discuss how the influence of genes, emotion, motivation and stress can be modelled.
Structure
The seminar will consist of 12 classes, taking place on Wednesdays at 16-17:30, from October 3 till December 19 in Nebenhaus (64a) Seminarraum 3.
Each class (except the first) will consist of an introductory lecture, followed by a paper presentation with the follow-up discussion.
Tentative schedule, topics, and papers to be presented
September 26. NO CLASS (I am away at a conference).
October 3. Introduction to the seminar, methodology, and overview of brain systems in learning, memory and decision making.
(No papers)
October 10. Hebbian learning and associative memory
Papers:
Bakker et al., "Pattern separation in the human hippocampal CA3 and dentate gyrus", Science (2008) - (pdf, 8 pts)
Mongillo et al., "Synaptic theory of working memory", Science (2008) - (pdf, 12 pts)
October 17. Synaptic plasticity and its computational models
Papers:
Kirkwood et al., "Experience-dependent modification of synaptic plasticity in visual cortex", Nature (1996) - (pdf, 9 pts)
Clopath et al., "Connectivity reflects coding: a model of voltage-based STDP with homeostasis", Nat. Neurosci. (2010) - (pdf, 15 pts)
October 24. Hippocampal coding of spatial and episodic memory
Papers:
Wills et al., "Attractor Dynamics in the Hippocampal Representation of the Local Environment", Science (2005) - (pdf, 9 pts)
MacDonald et al., "Hippocampal 'time cells' bridge the gap in memory for discontiguous events", Neuron (2011) - (pdf, 14 pts)
October 31. Consolidation and stability of long-term memory
Papers:
Pastalkova et al., "Storage of spatial information by the maintenance mechanism of LTP", Science (2006) - (pdf, 9 pts)
Tse et al., "Schemas and memory consolidation", Science (2007) - (pdf, 12 pts)
November 7. Heterosynaptic modulation of plasticity and three-factor learning
Papers:
Seung, "Learning in spiking neural networks by reinforcement of stochastic synaptic transmission", Neuron (2003) - (pdf, 15 pts)
Kim & Yoon, "Stress: metaplastic effects in the hippocampus", Trends Neurosci. (1998) - (pdf, 8 pts)
November 14. Dopamine, reward prediction and reinforcement learning
Papers:
Tobler et al., "Adaptive coding of reward value by dopamine neurons", Science (2005) - (pdf, 9 pts)
Samejima et al., "Representation of action-specific reward values in the striatum", Science (2005) - (pdf, 9 pts)
November 21. Action control: balancing exploration and exploitation
Papers:
Daw et al., "Cortical substrates for exploratory decisions in humans", Nature (2006) - (pdf, 8 pts)
Frank et al., "Prefrontal and striatal dopaminergic genes predict individual differences in exploration and exploitation", Nat. Neurosci. (2009) - (pdf, 14 pts)
November 28. Valuation and future discounting in the brain
Papers:
Tanaka et al., "Prediction of immediate and future rewards differentially recruits cortico-basal ganglia loops", Nat. Neurosci. (2004) - (pdf, 11 pts)
Yang & Shadlen, "Probabilistic reasoning by neurons", Nature (2007) - (pdf, 11 pts)
December 5. Uncertainty and its neural correlates
Papers:
Hsu et al., "Neural systems responding to degrees of uncertainty in human decision-making", Science (2005) - (pdf, 8 pts)
Yu & Dayan, "Uncertainty, neuromodulation, and attention", Neuron (2005) - (pdf, 15 pts)
December 12. Stress, motivation and neuromodulators in learning and memory
Papers:
Niv et al., "A normative perspective on motivation", Trends Cogn. Sc. (2006) - (pdf, 11 pts)
Luksys et al., "Stress, genotype and norepinephrine in the prediction of mouse behavior using reinforcement learning", Nat. Neurosci. (2009) - (pdf, 15 pts)
December 19. Brain-computer interface, patterns and prediction
Papers:
Xue et al., "Greater neural pattern similarity across repetitions is associated with better memory", Science (2010) - (pdf, 9 pts)
Cerf et al., "On-line, voluntary control of human temporal lobe neurons", Nature (2010) - (pdf, 9 pts)
Note: new papers may be added to the list later. If you would like to present another good paper (related to these topics) that is not on the list, please let me know.
Evaluation
As this seminar has no exam, active participation is essential for a successful completion of the course. This consists of attending classes, presenting one or several papers, and participating in the follow-up discussion. Questions regarding the main points of each class will be provided in order to facilitate understanding and encourage critical thinking. After each class students should be able to answer such questions.
Your performance will be evaluated based on attendance, active participation and paper presentation(s) as follows:
- By presenting a paper you can earn up to 8-15 points, depending on the length & difficulty of the paper and the quality of your presentation. The maximum number of points for each paper is provided above.
- For attending each class you earn 1 point.
- By participating actively in the class and the follow-up discussion of the paper you can earn an additional 1 point each time.
- For 4 credits, you may also earn up to 10-20 points by writing an essay (Referat) on a topic of your interest, related to the seminar. However, since our class is small, it's recommended that you rather present more papers than write an essay.
For a pass with 2 KPs you need to earn at least 24 pts, which can be achieved by regular attendance + 1 paper presentation.
For 4 KPs, you need 38-58 pts. Your grade will then be points / 10, rounded to the nearest half.
Dr. Gediminas Luksys, PhD
Postdoctoral Assistant
Computational modelling
