✅ Deep Learning Interview Questions & Answers 🤖🧠

1️⃣ What is Deep Learning and how is it different from Machine Learning?
Deep learning is a subset of machine learning that uses multi-layered neural networks to automatically learn hierarchical features from raw data (e.g., images, audio, text). Traditional ML often requires manual feature engineering. Deep learning typically needs large datasets and computational power, whereas many ML methods work well with less data. ML models can be more interpretable; deep nets often appear as “black boxes”.

2️⃣ What is a Neural Network and how does it work?
A neural network consists of layers of interconnected nodes (“neurons”). Each neuron computes a weighted sum of inputs plus bias, applies an activation function, and passes the result forward. The input layer receives raw data, hidden layers learn features, and the output layer produces predictions. Weights and biases are adapted during training via backpropagation to minimize the loss function.

3️⃣ What are activation functions and why are they important?
Activation functions introduce non-linearity into the network, allowing it to learn complex patterns. Without them, the network would be equivalent to a linear model. Common examples: ReLU (outputs zero for negative inputs), Sigmoid and Tanh (map to bounded ranges), and Softmax (used in output layer for multi-class classification).

4️⃣ What is backpropagation and the cost (loss) function?
A cost (loss) function measures how well the model’s predictions match the true targets (e.g., mean squared error for regression, cross-entropy for classification). Backpropagation computes gradients of the loss with respect to weights and biases, and updates them (via gradient descent) to minimize the loss. This process is repeated over many epochs to train the network.

5️⃣ What is overfitting, and how can you address it in deep learning?
Overfitting occurs when a model learns the training data too well, including noise, leading to poor generalization on unseen data. Common techniques to avoid overfitting include regularization (L1, L2), dropout (randomly dropping neurons during training), early stopping, data augmentation, and simplifying the model architecture.

6️⃣ Explain convolutional neural networks (CNNs) and their key components.
CNNs are designed for spatial data like images by using local connectivity and parameter sharing. Key components include convolutional layers (filters slide over input to detect features), pooling layers (reduce spatial size and parameters), and fully connected layers (for classification). CNNs automatically learn features such as edges and textures without manual feature engineering.

7️⃣ What are recurrent neural networks (RNNs) and LSTMs?
RNNs are neural networks for sequential or time-series data, where connections loop back to allow the network to maintain a memory of previous inputs. LSTMs (Long Short-Term Memory) are a type of RNN that address the vanishing-gradient problem, enabling learning of long-term dependencies. They are used in language modeling, machine translation, and speech recognition.

8️⃣ What is a Transformer architecture and what problems does it solve?
Transformers use the attention mechanism to relate different positions in a sequence, allowing parallel processing of sequence data and better modeling of long-range dependencies. This overcomes limitations of RNNs and CNNs in sequence tasks. Transformers are widely used in NLP models like BERT and GPT, and also in vision applications.

9️⃣ What is transfer learning and when should we use it?
Transfer learning reuses a pre-trained model on a large dataset as a base for a new, related task, which is useful when limited labeled data is available. For example, using an ImageNet-trained CNN as a backbone for medical image classification by fine-tuning on the new data.