Rozdíly

Zde můžete vidět rozdíly mezi vybranou verzí a aktuální verzí dané stránky.

--- courses:a4m33bia [2013/05/31 16:13]
mnicky
+++ courses:a4m33bia [2025/01/03 18:23] (aktuální)
@@ Řádek 25: / Řádek 25: @@
 Skripta Doc. Snorka
 http://www.uloz.to/xW51TGT/neurony-kniha-zip
+====Test v semestru 2016: ====
+{{:courses:img_3243.png?direct&100|}}
+{{:courses:img_3244.png?direct&100|}}
+{{:courses:img_3245.png?direct&100|}}
 ====Test v semestru 2012: ====
@@ Řádek 42: / Řádek 47: @@
 (na stiahnutie tuna: http://uloz.to/xAgovEPD/test-varianta2012-a-pdf)
 ====Test v semestru 2011: ====
@@ Řádek 64: / Řádek 68: @@
 </code>
 ===== Zkouška =====
@@ Řádek 131: / Řádek 136: @@
 . memory based immigrants scheme
 . linear scaling + zakreslit do grafu
+</code>
+==== 2014, 2. termin ====
+<code>
+B:
+. Sammon's projection - what is it used for, write the equation
+. GMDH MIA - write the architecture, ...
+. Echo State network
+. NEAT - structural mutations, selection, benefits of complexification
+. indirect gene encoding, use in hyperNEAT
+. NSGA2 / SPEA2 - how is the density information used
+. Roulette wheel - solve example (same as above), write expected values and actual value range
+. C-metric - advantages, disadvantages, draw some example
+. Discrete PSO - position, velocity vectors; describe velocity change when Vmax is used
+. Memory-based immigration scheme
+</code>
+==== 2015, 1. 6. ====
+<code>
+varianta A:
+. RBF (4p)
+. Mutation and crossover in neuro-evolution. (3p)
+. What dataset preprocessing approaches do you know? Give a short explanation for each. (3p)
+. HyperNEAT (4p)
+. GMDH MIA (3p)
+. Relation between PCA and autoencoders. How would you learn Stacked Auto-Encoder? (3p)
+. Multi-Objective optimization  (4p)
+  * Domination
+  * Describe two goals of multi-objective optimization. Two desired properties of final set of solutions.
+  * Draw Pareto-optimal set
+  * Draw first 4 non-dominated fronts
+. ACOR  (4p)
+. Linear scaling + draw a graph.  (4p)
+. SPEA2 - describe fitness assignment scheme. (3p)
+  * Strength value
+  * Raw fitness
+  * Density information
+. Dynamic Optimizations - GARB  (3p)
+  * Redundant represenatation
+  * Gene-strength adaption
+.Describe Strongly-Typed Genetic Programming.  (2p)
+</code>
+==== 2016, 2. 6. ====
+<code>
+Explain/define/describe:
+a)	k fold crossvalidation
+b)	BPTT
+c)	NEAT
+d)	CNN
+e)	Domanation in MOO
+f)	Memory-based Approaches: Explicit Memory
+g)	Evolution scheme in NSGA II
+h)	Diffecential evolution – donor vector + crossvalidation
+i)	Velocity in PSO
+j)      Linear scaling (use graph)
+Draw 4 pareto-optimal front into graph (min-max objectives)
+Competing Conventions Problem - i inputs, h hidden neurons, o outputs. # of symetries?
+Only neurons with linear activation function. Write equation for ANN with single hidden layer.
+</code>
+==== 2016, 16. 6. ====
+<code>
+. Self-Organizing Maps (SOM) [3 pts]
+• Explain SOM network architecture and learning.
+• Draw the SOM for 2D input and 5x5 grid of neurons.
+. Recurrent Neural Netowrk (RNN) [3 pts]
+• Describe fully-connected RNN architecture.
+• Describe evaluation and learning of RNN.
+• What is synchronous/asynchronous network evaluation?
+. Neuro-evolution [3 pts]
+• Discuss mutation and crossover in neuro-evolution.
+. Hypercube-based NeuroEvolution of Augmenting Topologies (HyperNEAT) [4 pts]
+• Describe HyperNEAT algorithm.
+• What is the difference between HyperNEAT and standard evolution of ANNs?
+. Overfitting [4 pts]
+• How can you prevent overfitting of artificial neural network?
+. Perceptron vs Radial Basis Function (RBF) type neurons [3 pts]
+• Describe differences between perceptron and RBF neurons.
+• Draw an illustrative picture of how they divide an input space.
+. EAs for Multi-Objective Optimizations [4 pts]
+• Write a definition of domination.
+• Describe two goals of the multi-objective optimization. What are the two desired properties of the final set of solutions?
+• Lets assume a set of all feasible solution in the left figure below. Draw a complete Pareto-optimal set in the figure given the minimization objective o1 and minimization objective o2.
+• Lets assume the set of candidate solutions in the right figure. Draw the first four fronts of non-dominated solutions given the minimization objective o1 and minimization objective o2.
+{ two pictures, first figure - space, second figure - points }
+. Ant Colony Optimization for Continuous Domain (ACOr) [3 pts]
+• Describe the ACOr algorithm
+• Describe the way the Gaussian kernel probabilty density function (PDF) is used to model the pheromone.
+• Describe how the Gaussian kernel PDF is modelled and how its parameters are estimated.
+. Roulette Wheel [4 pts]
+• Describe the roulette wheel selection method.
+• Given a population of 5 individuals with the following fitness value - fitness(A)=1, fitness(B)=2, fitness(C)=3, fitness(D)=5, fitness(E)=9 - determine an expected number of copies that solution A and E recieve among ten solutions samples using a roulette wheel selection method. What is the range of the actual number of copies of A and E out of the ten samples solutions?
+. Dynamic Optimizations [3 pts]
+• Desribe the principles of the GA with Real-coded Binary Representation (GARB).
+• Redundant representaion
+• Gene-strength adaptation
+. Ant Colony Optmimization (ACO) [3 pts]
+• Describe the following formula used for probabilistic decision making in ACO algorithm. How is it used in ACO algorithm?
+• Explain a meaning of the symbols Tau, Eta, alpha, beta, tabuk.
+{ p_ij^k = ... formula }
+. Performance Metrics for Multi-Objective Optimizations [3 pts]
+• Describe the S metric (i.e. the size of the space covered) used to assess a quality of a set of non-dominated solutions.
+• What are its advantages and disadvantages?
+• Illustrate it on an example.
 </code>
 ~~DISCUSSION~~

courses/a4m33bia.1370009606.txt.gz · Poslední úprava: 2025/01/03 18:14 (upraveno mimo DokuWiki)

Nahoru