Neuroinformatics, 2016-01-12

Module: 
Neuroinformatics
Examiner: 
Gordon Pipa
Assessor: 
Leon Sütfeld
Date: 
Tue, 2016-01-12

The examination went probably pretty similar to the other 6 recent protocols which are already uploaded, so I will just briefly summarise the questions. Of particular interest might be the questions to my second subject - Leon's "Concepts and Applications of Neural Networks" - which hasn't been topic for a module examination before.

Questions Neuroinformatics:

* Explain Bayes Rule
* Explain Maximum Likelihood
* We usually used the normal distribution as the Likelihood function but we could also use other functions as Bernoulli and Binomial, how are these connected? -> Exponential Family
* In linear regression we use non-linear basis functions, why do wo use linear weights? -> easier to optimise/derive
* If we now fitted different models, how can we decide on which to choose?
* I first explained cross-validation, he then asked what we can do if we can't effort to test each individual model
* I then explained AIC with -logL as biased approximation of KL-divergence to true model. Followup question: The bias term of AIC punishes the parameters, can we also take the sample size into consideration? -> AICc
* Can you Explain Bayesian Inference?

Questions CANN:

* Can you explain the idea behind Reservoir Computing?
* How do you perform a state update (with formula)
* What are SORNs? What are the basic plasticity mechanisms?
* What happens if we use STDP but no Weight normalization or no intrinsic plasticity?