Neural Network Neural Network Applications and Frameworks

Question 1

What is PyTorch's dynamic computational graph and how does it differ from TensorFlow 1.x's static graph?

Accepted Answer

PyTorch builds the graph incrementally during execution; TensorFlow 1.x required defining the full graph before running

Answer

PyTorch uses less memory; TensorFlow is faster at inference

Answer

PyTorch only supports CPU; TensorFlow supports GPU

Answer

PyTorch uses symbolic differentiation; TensorFlow uses numerical

Question 2

What is the purpose of the `nn.Module` class in PyTorch?

Accepted Answer

To serve as the base class for all neural network modules, managing parameters and forward computation

Answer

To load pretrained model weights from disk

Answer

To define the training loop and optimizer steps

Answer

To handle GPU memory allocation for tensors

Question 3

In TensorFlow/Keras, what does model.compile() configure?

Accepted Answer

The optimizer, loss function, and evaluation metrics for training

Answer

The model's layers and architecture

Answer

The input and output tensor shapes

Answer

The hardware device (CPU/GPU) for training

Question 4

What is ONNX (Open Neural Network Exchange) used for?

Accepted Answer

A standard format for representing deep learning models enabling interoperability between frameworks

Answer

An open-source framework for training neural networks

Answer

A tool for visualizing neural network architectures

Answer

A dataset format for neural network benchmarking

Question 5

What is TensorBoard primarily used for?

Accepted Answer

Visualizing training metrics, model graphs, embeddings, and other diagnostics during training

Answer

Deploying models to production servers

Answer

Compiling model code to native binaries

Answer

Managing distributed training across multiple GPUs

Question 6

What is the difference between model inference and model training in the context of neural networks?

Accepted Answer

Training updates weights via backpropagation; inference runs a forward pass only to produce predictions

Answer

Inference uses different hardware than training always

Answer

Inference requires labeled data; training does not

Answer

Training uses smaller batch sizes than inference