Machine Learning Exam Questions R22
Uploaded by
mbsailajanawinMachine Learning Exam Questions R22
Uploaded by
mbsailajanawinCommon questions
Powered by AIThe primary types of machine learning models include supervised learning, unsupervised learning, and reinforcement learning. Supervised learning involves training a model on labeled data, which means that each training example is paired with an output label. This type is useful for tasks like classification and regression . Unsupervised learning, on the other hand, deals with unlabeled data and is used to identify hidden patterns or intrinsic structures in input data, commonly used in clustering and association . Reinforcement learning is a type of learning where an agent interacts with an environment, taking actions to maximize cumulative reward; it is widely used in scenarios where decision making is sequential .
Support Vector Machines (SVM) are powerful classification tools that function by finding the optimal hyperplane which best separates the data classes in high-dimensional space. They work by maximizing the margin between different classes, with the support vectors being the data points that are closest to the decision boundary. This method is effective in linearly separable cases, and can be extended to non-linear data distributions using kernel tricks to transform the data into higher dimensions .
Bayesian networks are a type of probabilistic graphical model that use directed acyclic graphs to represent a set of variables and their conditional dependencies. The components of Bayesian networks include nodes representing the variables and edges capturing the probabilistic dependencies. Their significance lies in their ability to model complex distributions and perform efficient probabilistic inference, making them valuable in tasks like prediction, diagnostics, and decision making under uncertainty .
Backpropagation is critical in training feedforward networks as it provides a method for updating the weights of the network to reduce the error in predictions. It involves calculating the gradient of the loss function with respect to each weight by the chain rule, iteratively adjusting weights to minimize the loss. This effectively propagates errors back through the layers of the network, allowing for network optimization by gradient descent .
The Candidate-Elimination algorithm helps identify hypotheses by maintaining a boundary or a set of hypotheses which are consistent with the observed training examples. It uses two sets, G and S, which represent the most general and the most specific hypotheses respectively. As new training examples are presented, the algorithm refines these sets, eliminating hypotheses that do not match the data, until it finds the specific target concept that fits the observed examples .
The Expectation-Maximization (EM) algorithm is important in clustering because it provides a robust framework for finding the maximum likelihood estimates of parameters in probabilistic models with latent variables. It consists of two main steps: the Expectation step, which computes the expected value of the latent variables given the current estimates of the parameters, and the Maximization step, which updates the parameters to maximize the expected log-likelihood found in the Expectation step. This algorithm is iterative and converges to local maxima, making it useful for model optimization in contexts like Gaussian mixture models .
The key difference between Q-learning and SARSA algorithms in reinforcement learning lies in their approach to updating policies. Q-learning is an off-policy algorithm, where the action selection and value learning are based on different policies, potentially leading to more greedily optimized solutions. In contrast, SARSA is an on-policy learning algorithm and updates its Q-values using the actions derived from its current policy, often resulting in more stable learning but potentially slower convergence. Both techniques are popular for tasks requiring sequential decision-making .
Unsupervised learning differs from supervised learning in that it deals with unlabeled datasets, aiming to identify structure or patterns in the data without predefined labels, while supervised learning is conducted on labeled datasets. Common techniques for unsupervised learning include clustering methods like k-means and hierarchical clustering, and dimensionality reduction techniques like PCA. In contrast, supervised learning techniques include regression analysis, decision trees, and support vector machines .
Linear Discriminant Analysis (LDA) facilitates dimensionality reduction by finding the linear combinations of features that best separate the classes. It maximizes the ratio of the difference between classes (between-class variance) to the variance within each class (within-class variance). Applications of LDA include face recognition and text classification, where reducing dimensionality improves computational efficiency without compromising significant discriminative information .
Radial Basis Function (RBF) Networks address interpolation problems by using neurons with radial basis functions as activation functions, which are capable of capturing the local topological structure of the data. The basis functions commonly used are Gaussian functions, which enable the network to smoothly interpolate the data between points. This allows for effectively solving interpolation by adjusting the width and centers of these Gaussian functions to minimize error while fitting the dataset .
