Untitled Deck Flashcards
What is the primary biological inspiration for artificial neural networks?
The brain, which consists of approximately (10^{11}) neurons, (10^{14}) synapses, and has a response time of 1–10 ms.
What is the main challenge of traditional solutions in robotics?
They require exhaustive knowledge of the mechanical characteristics of the robot and careful calibration.
Who proposed the McCulloch-Pitts model?
McCulloch and Pitts in 1943.
What are the key components of artificial neurons?
Synaptic weights, threshold, synaptic potential, activation function, and state.
What are the main types of neuron states?
Discrete (binary or bipolar) and continuous.
Name three types of activation functions.
Step, ramp, and exponential functions.
What is a perceptron primarily used for?
Recognition and classification tasks.
Who introduced the concept of a single-layer perceptron?
Frank Rosenblatt in 1958.
What is a significant limitation of single-layer perceptrons?
They can only classify linearly separable sets.
How did Minsky and Papert demonstrate the limitation of perceptrons in 1969?
By showing they cannot implement the XOR function.
What is the primary solution to overcome the limitations of single-layer perceptrons?
Adding more layers to create a multilayer perceptron.
What is the purpose of the backpropagation algorithm in multilayer perceptrons?
To find synaptic weights that minimize the error.
What are the main types of layers in a neural network?
Input, hidden, and output layers.
What are the types of synapses in artificial neural networks?
Forward, backward, and lateral synapses.
Define feed-forward neural network architecture.
An architecture where connections between the nodes do not form a cycle.
What is a recurrent neural network?
A network where connections can form cycles, allowing it to retain memory over time.
What are radial basis functions (RBF) typically used for?
Function approximation, time series prediction, and classification.
Who introduced RBF networks?
Broomhead and Lowe in 1988.
What is the most commonly used distance metric in RBF networks?
Euclidean distance.
Give an example of a popular radial basis function.
The Gaussian function: (phi(r) = exp(-(epsilon r)^2)).
What are the two main phases in training RBF networks?
Selection of RBF centers and computation of synaptic weights.
What is a critical step during the training of RBF networks?
Selecting RBF centers to match the distribution of training data.
Name one method used to compute synaptic weights in RBF networks.
Gradient descent.
What is the advantage of multilayer perceptrons in solving inverse kinematics problems?
They do not require precise mechanical knowledge and adapt to variations.
What does the activation function do in an artificial neuron?
It determines the output state based on the input synaptic potential.
What is the role of the threshold in an artificial neuron?
It sets a value above which the neuron activates.
What is meant by ‘adaptive synapses’ in artificial neural networks?
Synapses that adjust their weights during learning.
Why are neural networks considered robust and fault-tolerant?
Because information is distributed and they can continue functioning despite partial damage.
What is online learning in the context of neural networks?
A learning approach where the model updates continuously as new data arrives.
What is the significance of batch learning in neural networks?
The model updates after processing a batch of training data, reducing the frequency of updates.
