Exam questions SS21

Question 1

Describe the Chinese room argument

Answer 1

The CRA is a thought experiment which states that a computer cannot think even if it passes the Turing Test / a system that behaves like cognitive systems does not necessarily act with the same principles as a cognitive system.

In the Chinese Room there is someone who translates Chinese symbols to the english language based on rules. To a person outside of the room it appears that the person in the room actually understands the Chinese language although the person does not but only uses rules.

Question 2

2. Explain the purpose of the Chinese room argument

Answer 2

The Chinese Room is a thought experiment which states that the Turing test does not necessarily imply that a machine can have cognition. In the Chinese Room, there is someone who translates Chinese symbols to the english language based on rules. To a person outside of the room it appears that the person in the room actually understands the Chinese language although the person does not but only uses rules. This could be the same case for the machine that passes the Turing Test.

Question 3

3. Describe the Turing test

Answer 3

The Turing Test is a test of a machine’s ability to exhibit intelligent behaviour equivalent to, or indistinguishable from, that of a human.

A human evaluator would judge natural language conversations between a human and a machine designed to generate human-like responses (text only). If the evaluator could not reliably tell the machine from the human, the machine would be said to have passed the test.

Question 4

4. Explain what the robotics paradox is.

Answer 4

The robotics paradox concerns the fact that machines are good (outperform humans) at well-defined, repeatable tasks but can’t outperform humans at supposedly simple tasks. No robot can operate in dynamic real-world environments and carry out everyday tasks. The reason is that simple tasks like shopping are slightly different every time and it is not possible to account for every circumstance when writing a deterministic program. Slight changes create misalignment between worldview and reality. Priory knowledge is required for real-life conditions.

Question 5

5. Explain the difference between cognitivist cognitive systems and emergent cognitive systems

Answer 5

Cognitivist cognitive systems are based on the hypothesis that cognition is a form of computation. Cognitive functions are modelled as working computer programs.
In emergent cognitive models, cognition is a continuous self-organisation process that is driven by the interaction between the agent and its environment.

Question 6

6. Name seven cognitive capabilities that a system must possess in order to be self-reliant and adaptive, and interact with its environment

Answer 6

Self-reliant

• goal-directed
• autonomous
• interact with other agents

Perception & Action

• interpretation
• sensing
• anticipation
• adaption

Adaption

• reaction
• learning
• anomaly detection

Question 7

7. Name the four different lobes in the cerebrum and explain what their main cognitive functions are.

Answer 7

Frontal lobe: short-term memory, action planning, movement control
Parietal lobe: somatic sensation, body image
Temporal lobe: hearing, learning, memory, emotions
Occipital lobe: vision

Question 8

8. Assume that you have a doll company. Your new series features dolls that are more realistic than any toy produced until now. However, the doll is highly unsuccessful and testers describe it as creepy. Name and explain the phenomenon that is taking place

Answer 8

Uncanny valley phenomenon states that a steady increase in human-likeness does not yield a steady increase in the familiarity to humans.

Question 9

9. A company claims that it can make you smarter by unlocking the 90% unused capacity of your brain. Explain whether this is possible or not

Answer 9

This is not possible because if we would use all of our brain at the same time the result would be epilepsy. Humans use their entire brain but just not at once.

Question 10

10. Define what a cognitive system is.

Answer 10

A cognitive system is an autonomous system that can perceive its environment, learn from experience, anticipate the outcome of events, has goal-oriented behaviour and can adapt to changing circumstances.

Question 11

11. Define cognition

Answer 11

Cognition is the process of knowing and includes perception and judgement. It includes all processes of consciousness by which knowledge is accumulated like perceiving, recognizing. It is the experience of knowing and different from feeling or willing. It refers to the mind and the brain.

Question 12

12. Name the components of a cognitive system

Answer 12

Environment (situatedness)
Body (embodiment)
Brain (constraints)
Other Agents (interaction)

Question 13

13. Name two applications that require cognition

Answer 13

Human-machine interaction

Natural Language processing

Question 14

14. Draw the cycle of cognitive processing

Answer 14

perception -> cognition (anticipation, assimilation, adaption) -> action

Question 15

15. Name the two schools of thought which deal with the mind-body problem. For each of them, describe their approach to the problem

Answer 15

Mind-body problem: How is the mental world related to the physical world?

Monism: There is only one mind and body. Mental states are physical states. If two people have the same mental property they share the corresponding physical property.

Substance Dualism: Mind and body are two separate entities. Their separation makes the soul immortal and enables free will.

Question 16

16. Explain What kinds of tasks robots/computers perform better than humans, and which ones they perform worse. Give an example for each of them.

Answer 16

Robots are good at performing well-defined, repeatable tasks in controlled environments like a factory conducting a welding task. Humans are good at performing simple tasks like grocery shopping. These simple tasks are not deterministic and might change.

Question 17

17. Explain why birds sing using an ultimate explanation and using a proximate explanation

Answer 17

Proximate explanation of why birds sing is because certain hormones in their brain trigger the vocal cords to produce songs.
The ultimate explanation is that they do so because they want to mate and reproduce.

Question 18

1. Name four advantages of executing experiments with virtual robots instead of physical ones.

Answer 18

save costs because no physical form is required
save time as you can run virtual experiments quicker and in parallel
less robots breaking down because you can just restart the experiment virtually. Less damage
no sensor imperfections
reset experiment automatically

Question 19

2. Explain what the neurorobotics platform is and explain the advantages of its usage

Answer 19

The neurorobotics platform is a multidisciplinary approach and tries to unify neuroscience, robotics, and AI. You have simulated robots in a virtual environment controlled by NN. The advantage is that you don’t have to have a physical robot and you can thus save time and money.

Question 20

3. Name two software tools that are integrated in the neurorobotics platform

Answer 20

Gazebo: body simulation for robots in virtual environments
OpenSim: modelling and simulation of movement

Question 21

4. Explain the function of buffers in ACT-R

Answer 21

Information about the current state of the module is exposed through named buffers. Every buffer stores one chunk and is assigned to exactly one module. Buffers serve as interfaces between modules. Buffers directly process queries about their contents.

Question 22

5. Explain a module in ACT-R. Name two features of modules in ACT-R.

Answer 22

Modules represent independent parallel processing units that encapsulate specific functions:
every module corresponds to a cognitive function in the brain
information processing is independent from other modules

Question 23

6. Explain what production is in ACT-R

Answer 23

Production is a statement of a particular contingency that control behavior

Question 24

1. Explain whether the following statement is true or false: “Only 10% of the human brain is used”

Answer 24

It is not true. We use our entire brain, just not all of it at the same time.

Question 25

2. Explain what the main purpose of the white matter in the cerebrum is.

Answer 25

White matter essentially functions in affecting learning and brain functions, modulating the distribution of action potential, and coordinating communication between the different brain regions.

Question 26

Name and describe the anatomy of the surface of the cerebral cortex.

Answer 26

Gyri = crests formed by the convoluted surface of the cerebral cortex
Sulci = fissures between two neighbouring gyri
lobes = on each hemisphere there are two major sulci which divide the cerebral surface into the frontal, parietal, temporal and occipital lobe.

Question 27

4. Name two distinctive features of human brain that makes it stand out among other species

Answer 27

neuron morphology: the pyramidal neurons in the human cortex have the most elaborate and spine-rich dendritic trees
Electrical properties of neurons: pyramidal neurons have a low membrane capacity, which enhances signal transmission

Question 28

5. Explain what the Blue Brain Project

Answer 28

Whole brain model that developed a data-driven reconstruction process for cortical microcircuits.
The digital reconstruction is based on experimental data and predictions and aims at reproducing data observed in experiments.

Question 29

6. Provide the names of the three main components of a biological neuron and describe their main tasks.

Answer 29

axon - emerges from the some at the axon hillock and conducts electrical impulses to other neurons
soma - is the cell body and contains the nucleus
dendrites - receive electrical impulses from other neurons through synapses

Question 30

8. Name and describe the two subsystems of the vertebrate nervous system

Answer 30

Central Nervous System: central information processing system formed by the brain and the spinal cord. It collects and distributes data throughout the body
Peripheral Nervous System: transmits signals between sensory organs, muscles, and internal organs and the CNS

Question 31

9. Explain the function of the hippocampus

Answer 31

episodic, long-term memories

Question 32

10. Name two features or properties of the peripheral (autonomic) nervous system

Answer 32

The Peripheral Nervous System is responsible for self-regulation and operates unconsciously. It includes two antagonistic subsystems:

Sympathetic nervous system:
Prepares the organism for stress
Parasympathetic nervous system:
sets the body to resting state and increases digestive functions

Question 33

11. Map the nervous system into the perception-cognition-action diagram

Answer 33

Cognition with sub-functions are part of the brain. Perception are sensory organs and action is spinal cord and muscles.

Question 34

12. Explain what is the main purpose of the brainstem. Name the four sections that compose it.

Answer 34

The brainstem connects the cerebrum with the spinal cord.

Four sections:
Diencephalon
Midbrain
Pons
Medulla Oblongata

Question 35

13. Describe what the encephalization quotient is

Answer 35

The encephalization quotient (EQ) is a measure of relative brain size and is often used to convey how small or large a species brain is compared to that of other species of similar body size.

Question 36

14. Briefly explain the function of the amygdala in the human brain

Answer 36

The amygdala is a group of nuclei related to the analysis of the emotional or motivational significance of a stimulus

Question 37

1. Name three learning rules based on Hebbian learning

Answer 37

Basic Hebb’s rule
Correlation-based rule
Covariance-based rule

Question 38

2. Name and explain two paradigms for encoding information in spiking neural networks

Answer 38

time-to-first-spike: time it takes until the first spike of a neuron
rate based encoding: average spikes during a time interval

Question 39

3. Name five levels of modelling the dynamics of biological neurons

Answer 39

Detailed-compartment model
Reduced-compartment model
Single-compartment model
Cascade Models
Black-box-models

Question 40

4. Explain what the action potential of a biological neuron is

Answer 40

The action potential is an all or nothing event in which one neuron fires to the next one. The action potential is generated by depolarization then the actual potential is reached at a peak, then there is repolarization, hyperpolarization, the refractory period and the resting state.

Question 41

5. Describe the Hodgkin-Huxley Model. You may draw a diagram

Answer 41

The Hodgkin-Huxley model (1952) has a single compartment and is the basis for conductance-based neuron models. It models the passive electrical properties of the cell membrane (Capacity / Resistance).

Its major drawback: computationally expensive, better to use phenomenological models that capture the overall behavior of biological neurons without explicitly emulating the underlying dynamics

Question 42

6. Explain the meaning of the statement “Neurons that fire together, wire together” in the context of learning. Name which type of learning is based on this statement

Answer 42

The synaptic weight between two neurons increases if the two neurons activate simultaneously, and reduces if they activate separately.

Hebbian learning.

Question 43

7. Explain neuroplasticity and why it is relevant for the function of the central nervous system

Answer 43

Ability of neurons to change their morphology, which allows for development and learning

Question 44

8. Explain the refractory period of a neuron. Name the two main phases that it can be split into

Answer 44

The refractory period occurs after repolarization and describes the state after the action potential occurs. There are two sub-phases.

• The absolute refractory period in which no new action potential can be elicited.
• Then there is the relative refractory period in which an action potential can be elicited but only under stronger stimuli than in the resting state.

Question 45

9. Explain the resting membrane potential of a neuron

Answer 45

The resting membrane potential of a neuron is the equilibrium / idle state resulting from the sum of the different ion flows between the inside and the outside of the neuron.

In other words: the resting membrane potential of a neuron is the voltage at the equilibrium point of the charge gradient and the voltage gradient

Generally -75mV

Question 46

10. Name four different activation functions for general analog neuron models

Answer 46

Sigmoid function
Rectified linear unit (ReLu)
Gaussian function
Binary step activation function 
tan(h) activation function

Question 47

11. Explain the main differences between analog neuron models and spiking neuron models

Answer 47

Analog neuron models output real numbers and cannot model the temporal dynamics of biological neurons, they have a low biological realism at a lower computational cost.

Spiking neuron models are able to model temporal dynamics and are more computationally complex. Relatively high biological realism.

Question 48

12. Draw a plot of the membrane potential of the neuron over time and label the different sections.

Answer 48

Resting state -70mV
Depolarization (caused by stimuli)  -70mV ==> 40mV (passes a threshold)
Action potential
Repolarization (-60v), 
refractory period, 
resting state ( -50v)

Question 49

13. Explain the function of the myelin sheath around nerve fibres.

Answer 49

In the myelinated segments of the axon, signals are transmitted through fast electrotonic conduction. The myelin sheath increases the resistance of the membrane. The action potential is regenerated at gaps in the myelin sheath, the nodes of Ranvier.

Question 50

14. Solve the differential equation of a LIF neuron for a given scenario.

Answer 50

Good luck!

Question 51

15. Name two different types of neural plasticity and briefly describe them.

Answer 51

Functional plasticity: ability to alter and adapt the functional properties of neurons → biological processes modelled by STDP and hebbian learning.

Structural plasticity: physical creation and deletion of synapses, healing the network → not modelled so far.

Question 52

1. Name three methods for avoiding the problem of overfitting

Answer 52

Regularisation: add a regularisation term at the end of your objective (loss) function that punishes model complexity (i.e L2 regularisation)
Add more (diverse) training data (preferably that fits more test data)
Data augmentation: when you have not enough data, transform your existing data (i.e in case of an image: crop, shift, flip, noise, salt&pepper)

Question 53

2. In the context of machine learning, describe the concept of Max-Pooling.

Answer 53

Max pooling (complex cells) are filters applied in Convolutional Neural Networks (CNN) that perform subsampling on the output of a layer by computing the maximum value of a small subset of features.

• Reduces the dimensionality of the image patch and the number of synaptic weights in the following layers: we get key features.
• Max-pooling makes the network more robust against translations in the input image (translation equivariant): translating the image by a small amount does not significantly affect the values of most pooled outputs.

Question 54

3. Define “machine learning” with your own words. Provide an example of a robotic task that does not require machine learning.

Answer 54

Machine Learning: field of study that gives computers the ability to learn without being explicitly programmed. A ML system is composed of a Dataset (S), a model (M), an objective (loss or reward) function (L) and an algorithm A that adjusts M based on S and L.

Robotic task without ML: Boston Robotics WildCat uses mostly sensors.

Question 55

4. Explain the advantage of converting the convolutional kernels and inputs of a neural network into the Fourier Space. Explain What the main challenge for this approach is?

Answer 55

Conversion of inputs and kernels into Fourier Space reduces the number of operations required for computing convolutions. The challenge lies in keeping the computational cost for the Fourier Transform low.

Question 56

5. Explain the difference between interpolation and regression.

Answer 56

Interpolation; the approximated function should match exactly with all the dataset points (not used for ML predictive tasks OR overfitting if applied to ML)

Regression: the approximated function should minimise a loss function and does not have to match with all train data points: it should generalise (used for predictive tasks).

Question 57

6. State the problem of using least squares for linear classification. Explain how support vector machines address this problem

Answer 57

Least Square classification is really sensitive to outliers.

Support Vector Machines (SVMs) minimise the generalisation error by computing a hyperplane that maximises the margin of the classifier, i.e. the smallest distance between the decision boundary and the training samples → insensitive to outliers

Question 58

7. State the formula of the output of a general analog neuron model and calculate the output given the parameters.

Answer 58

The output n(x) of a neuron n with activation function A, synaptic weights w, bias b and input x is computed as follows (analog neurons are generalised linear models)

n(x)=A(w1x1+w2x2+ …+wnxn+b)
n(x)=A(W^T . X + b)

Question 59

8. Name three learning paradigms. For each of them, describe the type of data needed for training.

Answer 59

There are 3 main learning paradigms in machine learning:

Supervised learning: data object (with features) and labels associated

Unsupervised learning: data object (with features), no labels needed.
I.e k-means clustering, image segmentation

Reinforcement learning: sequential data and an agent. An agent learns by interactions in the environment (thanks to rewards that are the results of his interactions)

Additional: Semi-supervised learning

Question 60

9. Explain the method of k-fold cross-validation for partitioning a dataset.

Answer 60

Data is very limited in reinforcement learning, so instead of splitting dataset in fixed train set T and validation set V, perform k cross-validation iteratively:

Dataset is partitioned into k subsets and learned in k iterations

In every iteration, a different subset is selected as validation set

Overall performance corresponds to the averaged performance of the k iterations

Question 61

10. Describe the principle of Occam’s Razor and state how it is applied in machine learning.

Answer 61

Given two models M1 and M2 with equal performance, the simplest model (i.e less model features, less layers in a convolutional neural net…etc) is to be preferred (also called law of parsimony).

Question 62

11. Name and describe the role of the two streams of visual processing

Answer 62

The dorsal stream of visual processing determines an object’s spatial location.
It determines how motor actions are carried out.

The ventral stream determines an object’s identity

Question 63

12. Name and describe the three conceptual processing layers in the human visual system.

Answer 63

Low-level-vision → basic image analysis
Mid-level-vision → objects, surfaces
High-level-vision → memory, intention

Question 64

13. Name a problem that can not be solved by a single perceptron with a linear activation function. State briefly or draw the reason for that.

Answer 64

XOR problem: the Perceptron learning rule only converges for linearly separable data sets → It therefore cannot classify the XOR dataset correctly.

Question 65

14. State the difference between regression and classification.

Answer 65

Regression has a continuous output space while classification has a discrete output.

Question 66

1. Explain the difference between on-policy and off-policy learning

Answer 66

The agent is learning on-policy when the behaviour policy and target policy are the same.

The agent is learning off-policy when the behaviour policy differs from the target policy.

Question 67

2. Briefly explain what “bootstrapping value” means in the context of reinforcement learning.

Answer 67

Propagating a value between consecutive states by iteratively exploiting the recursive relationship that is formulated by the Bellman Equation is denoted as bootstrapping.

Question 68

3. Name and describe the two alternating steps that are performed iteratively under the framework of Generalised Policy Iteration.

Answer 68

During Generalised Policy Iteration, we iteratively apply policy evaluation and policy improvement.

Thus, both processes stabilize only when a policy has been found that is greedy with respect to its own evaluation function.

Question 69

4. Describe and state the differences between classic conditioning and operant conditioning in the context of associative learning processes

Answer 69

Classical conditioning is when a subject learns the relationship between an initially neutral conditioned stimulus and an unconditioned stimulus that reflexively produce a conditioned response. CR will be produced even without the unconditioned stimulus.

In operant conditioning a subject learns the relationship between a stimulus and its behavior. A stimulus is only presented in response to a certain action of the subject and serves as a reinforcer that increases or decreases the probability of that action.

Question 70

5. State the reason behind using a linear activation function in the output layer of a Deep Q-Network

Answer 70

The goal of Deep Q-Network is to estimate the action value function which is Q value. Therefore the output of DQN shouldn’t be constrained by any activation function → it is just a linear layer.

Question 71

1. Name the two main types of modern neuroimaging. Name one example for each of them.

Answer 71

Structural and functional neuroimaging. Structural is Magnetic Resonance Imaging (MRI) and Electroencephalography (EEG). Functional is functional MRI, Positron Emission Tomography (PET) and then there is Single Positron Emission Computational Tomography (SPECT).

Question 72

2. Explain the difference between MRI and fMRI, and how they help to better understand neural activity

Answer 72

MRI is structural whereas fMRI is functional. fMRI is an extension of MRI that visualises changes of the blood oxygen level caused by brain activity elicited through cognitive processing.