Jawaharlal Nehru Technological University Hyderabad B.Tech III Artificial Neural Networks Previous Question Paper (August - 2022 - R16)

An artificial neuron, often referred to as a Processing Element (PE), is the fundamental building block of an artificial neural network. Its model is inspired by the biological neuron and consists of the following key elements:

• Input Signals (x_i): These represent the external data or signals received by the neuron from other neurons or the environment.
• Weights (w_i): Each input connection is associated with a weight that represents the strength or importance of that specific input. Weights can be positive (excitatory) or negative (inhibitory).
• Summation Function (Net Input): This component calculates the weighted sum of all incoming signals. It combines the inputs and their corresponding weights, often including a bias term.
net = Σ (xᵢ * wᵢ) + b
• Bias (b): The bias is an additional parameter added to the weighted sum. It acts as an adjustable threshold that shifts the activation function, allowing the neuron to better fit the data.
• Activation Function (f): This function determines the output of the neuron by mapping the net input to a specific range. Common functions include the Step function, Sigmoid function, ReLU, and Tanh.
y = f(net)
• Output (y): The final result produced by the activation function, which may be sent to other neurons or serve as the network's final prediction.

Learning in artificial neural networks is the process of adjusting the weights and biases to improve performance on a specific task. The primary forms of learning include:

• Supervised Learning: The network is trained using a labeled dataset where the desired output for every input is known. The network compares its prediction with the target output, calculates the error, and adjusts weights to minimize this error. Examples include backpropagation and pattern classification.
• Unsupervised Learning: The network is provided with input data without corresponding target outputs. It attempts to discover hidden patterns, clusters, or structures within the data based on statistical properties. A common example is Hebbian learning or competitive learning (e.g., Self-Organizing Maps).
• Reinforcement Learning: The network learns through interaction with an environment. It receives a reward or penalty signal based on its actions. The goal is to maximize the cumulative reward over time by learning an optimal policy.
• Competitive Learning: A form of unsupervised learning where neurons in a layer compete for the right to respond to a given input. Only the neuron with the highest activation (the winner) is allowed to update its weights, a process often called Winner-Take-All.