🤖 Top AI Interview Questions: Part-2
Continue your AI interview prep with these advanced concepts and neural network questions.
A perceptron is the simplest type of neural network unit. It takes inputs, multiplies them with weights, adds a bias, and passes the result through an activation function to produce output.
Formula: output = activation(w₁x₁ + w₂x₂ + ... + b)
Gradient Descent is an optimization algorithm used to minimize the loss function in machine learning models.
It updates model weights iteratively in the opposite direction of the gradient to reduce prediction error.
- Variants: Batch, Stochastic, Mini-batch
- Learning rate controls step size
Backpropagation is the algorithm used in training neural networks.
It calculates the gradient of the loss function with respect to each weight using the chain rule, then updates weights with gradient descent.
- Forward pass (prediction)
- Backward pass (error correction)
CNNs are deep learning models specifically designed for image and spatial data.
- Use convolutional layers to detect features (edges, shapes)
- Pooling layers reduce dimensions
- Fully connected layers make predictions
- Used in: face recognition, image classification, object detection
RNNs are neural networks designed for sequential data like time series or text.
- They use memory (hidden state) to store previous inputs
- Struggle with long-term dependencies
- Variants like LSTM and GRU solve this issue
Transfer learning involves reusing a pre-trained model on a new but similar task.
Example: Use a model trained on ImageNet and fine-tune it for medical image classification.
Saves time and resources, especially with limited data.
- Parametric: Assume a fixed number of parameters (e.g., Linear Regression)
- Non-parametric: Don’t assume a specific form; grows with more data (e.g., KNN, Decision Trees)
- Non-parametric models are more flexible but need more data
- ANI (Narrow AI): Performs one task (e.g., Siri, ChatGPT)
- AGI (General AI): Human-like reasoning across domains (theoretical)
- ASI (Super AI): Exceeds human intelligence (future concept)
Reinforcement Learning (RL) is where an agent interacts with an environment and learns to make decisions via rewards/penalties.
Used in: Game playing (Chess, Go), robotics, autonomous driving
MDP provides a framework for modeling RL problems.
Includes:
- States
- Actions
- Transition probabilities
- Rewards
Comments (0)
No comments yet
Be the first to share your thoughts!
Leave a Comment