JNTUH B.Tech IV NEURAL NETWORKS AND DEEP LEARNING Previous Question Paper (July/August - 2022 - R18)

A biological neuron is a complex living cell that processes information through electrochemical signals, while an artificial neuron is a simplified mathematical model used in computational systems. Key differences include:

• Structure: Biological neurons consist of a soma, dendrites, an axon, and synapses. Artificial neurons consist of weighted inputs, a summation function, and an activation function.
• Signal Processing: Biological neurons communicate via discrete electrical pulses (action potentials) and chemical neurotransmitters. Artificial neurons process continuous numerical values.
• Learning: Biological learning occurs through synaptic plasticity (strengthening or weakening of connections over time). Artificial learning involves adjusting weights and biases using optimization algorithms like backpropagation.
• Speed and Power: Biological neurons operate at relatively slow speeds (milliseconds) but exhibit massive parallelism and extreme energy efficiency. Artificial neurons operate at high speeds (nanoseconds) but require significant electrical power.

Neuron connection architectures define how neurons are organized and how information flows through a network. The primary architectures include:

• Single-Layer Feedforward Network: The simplest architecture where input units are directly connected to output units. It is limited to solving linearly separable problems.
• Multi-Layer Feedforward Network: Contains one or more hidden layers between the input and output layers. This architecture allows the network to learn non-linear relationships and complex patterns by extracting hierarchical features.
• Recurrent Neural Networks (RNNs): Feature feedback loops where the output of a neuron is fed back into itself or previous layers. This creates an internal memory, making them suitable for sequential data like time series or natural language.
• Mesh or Fully Connected Networks: Every neuron in one layer is connected to every neuron in the next layer. While powerful, these networks are computationally expensive due to the high number of parameters.
• Competitive Networks: Neurons compete with each other for the right to respond to a given input, often used in unsupervised learning tasks like clustering and feature map generation (e.g., Kohonen maps).