Is like a Turing Machine

Functions using neural networks

Works through artificial intelligence models

Does not compute

Only symbolic data

Only non-computable functions

Computable functions

Artificial neural networks

They require explicit rule-based programs

They can be trained to recognize patterns

They manipulate symbols directly

They use complex algorithms for pattern recognition

Static

Dynamic

Non-existent

Symbolic

Fixed

Weighted

Symbolic

None of the above

Information processed sequentially

Information processed in parallel across a network

One neuron processes all information

Information stored in a single memory unit

A) A threshold of inhibitory input

B) A sum of excitatory and inhibitory inputs

C) Only excitatory input

D) A single neurotransmitter input

A) Its weight is negative

B) The sum of its inputs exceeds a threshold

C) It receives an input equal to 0

D) It has an output of 0

Feedforward networks have no hidden layers

Recurrent networks include connections between nodes in the same layer or prior units

A) Activity is spread across multiple units

B) Activity of single units

C) Task-specific rules applied

D) Information is stored in complex symbolic structures

It fires only when one input fires

It fires when both inputs fire

It requires multiple layers to fire

It fires if no input fires

To learn patterns

To manipulate symbol structures

To recognize objects

To adjust neuron connections

A) They must be programmed with explicit rules

B) They can learn from feedback

A) Feedforward networks are only for pattern recognition.

B) Recurrent networks include connections between nodes in the same layer or prior units

They require manual programming for each task

They can learn from feedback

They follow predefined algorithms

They are programmed once and never change

To manually adjust weights

To compute the sum of input values

To adjust weights based on errors and improve the model

To measure the final output of the network

Perception Learning

Hebbian Learning

Natural Language Learning

Classical Conditioning

A) "Neurons that fire together wire together"

B) "Neurons that fire independently create new connections"

C) "Neurons that are inactive wire together"

D) "Connections are made randomly"

A) They can compute any computable function

B) They require backpropagation

C) They can only compute linearly separable functions

D) They do not require learning

A) They only have one layer

B) They can compute any computable function

C) They use only supervised learning

D) They do not require backpropagation

It is not suitable for supervised learning

It may take a lot of data and time to train

It simplifies the problem too much

It doesn't improve the network’s accuracy

A) They excel at recognizing patterns

B) They can be trained without feedback

C) They are transparent and rule-based

D) They can learn from data automatically

Through a single, linear path

By summing all input values

By passing activations from one layer to the next

By processing information sequentially

Designing an explicit rule for the network

Finding suitable data for training

Ensuring the model is mathematically correct

Discerning the rules the ANN learned

The process involves clear, understandable rules

The rules are well-documented

It is often impossible to determine exactly how the ANN learned its solution

All outputs are deterministic

Minimize the training data

Minimize the cost function by adjusting weights

Ensure that the network does not overfit

Verify that the input is correct

They only manipulate symbols

They can adapt to both rule-based and pattern recognition tasks

They are completely transparent and explainable

They only focus on pattern recognition

The challenge of connecting abstract symbols to real-world meanings.

The difficulty in creating symbols for AI systems.

The problem of grounding AI systems in ethical considerations.

The issue of AI systems understanding human emotions.

The difficulty in converting physical symbols into computer code

The challenge of assigning meaning to symbols in AI systems

The challenge of making systems process natural language

The problem of finding appropriate symbols for patterns

The network adjusts based on its own observations

The network learns based on labeled data and feedback

The network trains without any labeled data

The network randomly guesses until it finds a solution

It processes inputs in one direction, from input to output

It uses memory units to retain previous states and processes sequential data

It is incapable of processing sequential data

It is used only in supervised learning

They process data without any layer architecture

They are particularly effective for image processing and recognition

They use very few data inputs

They excel in recognizing sequential data

The model does not perform well on new, unseen data

The model performs exceptionally well on new, unseen data

A) A function used to evaluate the accuracy of the model

B) A mathematical equation used to adjust neuron weights

C) A metric that measures the computational cost of training

D) A measure of how well the model is performing during training

They control the speed of learning

They process the input and produce an output based on specific rules

They regulate the computational cost of training

They transform input into a non-linear output, helping the model generalize

A) A technique to prevent overfitting by randomly disabling neurons during training

B) A method used to increase the number of neurons in a network

C) A way to improve the interpretability of a network's decision-making

D) A technique for fine-tuning the model's cost function

Image classification

Natural language processing

DNN has multiple hidden layers, whereas a shallow network has only one

A shallow network uses more data

DNN operates without activation functions

Shallow networks are faster to train than DNNs

Learning patterns from labeled data

Modifying the weights based on the results of actions taken

Training a model with minimal feedback

Ensuring a consistent output for any given input

The model learns without labeled data

The model requires feedback from a teacher

It involves predicting the output from given inputs

The model is only effective in supervised settings

A) Image denoising

B) Image generation from random noise

Learning from multiple sources simultaneously

Using knowledge gained from one task to improve performance on another related task

Training a model from scratch on a completely new dataset

Transferring the entire neural network from one platform to another

Combining strengths of multiple methods for better performance

Incorporating both pattern recognition and rule-based reasoning

Reducing the transparency of how the model works

Achieving more flexible problem-solving strategies

A) A type of loss function used in training

B) A hidden layer used to learn data patterns

C) An additional parameter added to a neuron to shift the output

D) The process of adjusting weights to minimize error

A) A technique to divide data into training and test sets

B) A method to adjust weights by propagating error gradients backward

C) A process to increase the complexity of a neural network

D) A way to reduce the size of a model

A) A method to adjust weights by propagating error gradients backward

B) A method to adjust weights by propagating error gradients backward

A) Classification involves predicting a continuous output, while regression predicts discrete labels

B) Classification involves predicting discrete labels, while regression predicts a continuous output

C) Classification only works with labeled data, while regression works with unlabeled data

D) Regression is faster than classification

The computational cost of training

The weights in a neural network

The loss function or error between the predicted and actual values

The number of layers in a neural network

Linear regression

k-means clustering

Decision trees

Support vector machines

The model is too simple

The model memorizes training data instead of learning general patterns

The model performs well on training data but poorly on new data

The model memorizes training data instead of learning general patterns

The model generalizes too well to new data

The model uses too few layers

Creating an image from a description

Translating text from one language to another

Recognizing objects in an image

Identifying patterns in time-series data

Increases the dataset size by creating new data from the original dataset through transformations

Reduces the data size to prevent overfitting

Adds noise to the dataset to make training faster

Normalizes the data to ensure better performance

Using multiple machine learning models together to improve performance

A method of training a single model on multiple datasets

A process of reducing the number of features in the model

A method for pre-processing data before feeding it to a neural network

To validate the model on new data only

To divide the training data into different subsets to evaluate the model's generalization

A) A process that normalizes the weights

B) A mathematical function applied to the output of a neuron to introduce non-linearity

C) A method for splitting data into training and testing sets

D) A procedure that adjusts the learning rate

To predict the output

To process input data and extract features

To split the data into training and validation sets

To calculate the weights for neurons

Convolutional Neural Networks (CNNs)

Recurrent Neural Networks (RNNs)

Multilayer Perceptrons (MLPs)

Random Forests

Overfitting

Underfitting

The vanishing gradient problem

The exploding gradient problem

When a model performs well on training data but poorly on unseen data

When a model performs well on both training and unseen data

When a model performs poorly on both training and unseen data

When a model performs poorly on training data but well on unseen data

A) The learning rate used in training

B) The number of times the model sees the entire training dataset

C) The number of layers in a neural network

D) The specific features selected for the model

When the gradients become too small, making learning slow or stagnant

When the gradients grow excessively, leading to model instability

When the data does not match the problem's assumptions

When the model cannot handle large datasets

A) A technique to make a model less complex to improve generalization

B) A method of increasing the size of the training set

C) A process of speeding up the training process

D) A way to modify the loss function to ensure fast convergence

Supervised learning

Semi-supervised learning

Unsupervised learning

Reinforcement learning

To calculate the loss function

To evaluate the performance of classification algorithms

To store the results of training on each epoch

To split the data into different subsets for cross-validation

To evaluate the accuracy of a model

To adjust the weights of a neural network

To provide feedback on the agent's actions and guide future decisions

To reduce the size of the training set

Batch gradient descent updates weights after each training sample, while stochastic updates after each mini-batch.

Batch gradient descent processes the entire dataset at once, while stochastic gradient descent updates weights more frequently based on each sample.

Stochastic gradient descent requires more data, while batch gradient descent uses less data for training.

Batch gradient descent uses a fixed learning rate, while stochastic gradient descent adapts the learning rate automatically.

The model performs well on both the training and testing datasets.

The model performs poorly on both the training and testing datasets.

The model performs well on the training dataset but poorly on the testing dataset.

The model performs moderately on both the training and testing datasets.

To split the data into training and testing sets.

To handle missing data and normalize the dataset for better model performance.

To automatically select the most important features of the dataset.

To train the model on the dataset before evaluating its performance.

K-fold cross-validation

Feature scaling

Data augmentation

Data normalization

The process of selecting the best hyperparameters for a model

A method for minimizing the loss function by adjusting the weights of the network

A type of activation function used in training

The procedure for splitting data into training and testing sets

To reduce the size of the dataset

To identify the most important features in the dataset

To optimize the settings (hyperparameters) that control the learning process

To reduce the complexity of the neural network architecture

When the gradients become excessively large, causing the model to become unstable during training

When the model overfits the training data due to excessive feature selection

When the data distribution is so skewed that it negatively impacts training

When the weights of the network become too small and ineffective

The model's objective function

The strategy or decision rule used by the agent to determine its actions

The loss function used to evaluate the agent's performance

The sequence of rewards the agent receives during training

Automatically select the best features from the dataset

Apply knowledge learned in one domain to a different, but related, domain

Regularize the model to prevent overfitting

Reduce the amount of training data needed for a model

Supervised learning

Unsupervised learning

Reinforcement learning

Semi-supervised learning

A technique that uses multiple learning algorithms to obtain better predictive performance

A method for data preprocessing

A single machine learning algorithm

A type of neural network

Combining multiple weak models to create a stronger model

Using only one model to make predictions

Dividing the dataset into smaller subsets for parallel processing

Automatically tuning hyperparameters during training

To predict a label for each sample in the dataset

To cluster data into meaningful groups or reduce its dimensionality

To generate new data points based on existing data

To apply a supervised learning model to an unsupervised task

K-Nearest Neighbors

Decision Trees

Support Vector Machines

Naive Bayes

The balance between underfitting and overfitting a model

The relationship between the size of the dataset and the complexity of the model

The number of features used in a model versus its performance

The choice of using a linear or non-linear model

To speed up the training process by reducing the number of epochs

To randomly "drop" units from the network during training to prevent overfitting

To increase the number of neurons in the network

To normalize the output from the network

To reduce the size of the dataset before training

To prevent gradients from becoming too large and destabilizing the training process

To improve the model's performance on testing data

To regularize the model by reducing the number of training epochs

To extract higher-level features by reducing the spatial size of the input

To combine features from different layers into a single output

To adjust the learning rate during training

To introduce non-linearity into the model

Softmax

Exponential Decay

Leaky ReLU

Sigmoid

SVM models are computationally inexpensive and easy to train.

SVMs can handle both linear and non-linear decision boundaries using kernels.

SVMs are easily interpretable, making them ideal for feature selection.

SVMs require very little data for training.

The difference between predicted probabilities and true class labels

The variance in predictions across multiple models

The computational cost of training a model

The number of features used in the training process