Neural Network Neural Network Applications and Frameworks 2

Question 1

What is model quantization in the context of deploying neural networks?

Accepted Answer

Reducing numerical precision of weights (e.g., from FP32 to INT8) to decrease model size and inference latency

Answer

Quantization reduces weight and activation precision from 32-bit float to lower bits (e.g., INT8), dramatically reducing model size and speeding up inference with minimal accuracy loss.

Question 2

What is neural architecture search (NAS)?

Accepted Answer

Automated methods to discover optimal architecture configurations using search algorithms

Answer

NAS uses search strategies (evolutionary, reinforcement learning, gradient-based) to automatically discover neural architecture designs that optimize performance on a target task.

Question 3

What is the purpose of mixed precision training?

Accepted Answer

Using FP16 for most computations while keeping FP32 for critical operations, speeding training while maintaining accuracy

Answer

Mixed precision training uses 16-bit floats (FP16) for forward/backward passes and 32-bit floats (FP32) for weight updates, achieving ~2-3x speedup with minimal accuracy loss on modern GPUs.

Question 4

What is federated learning in the context of neural network training?

Accepted Answer

Training models across decentralized devices while keeping data local and only sharing model updates

Answer

Federated learning trains a shared model across many devices (e.g., smartphones) without centralizing data — each device trains locally and only shares weight updates with a central server.

Question 5

What is model pruning in neural network deployment?

Accepted Answer

Removing weights or neurons that contribute little to predictions, reducing model size

Answer

Model pruning identifies and removes parameters (weights, neurons, filters) that have minimal impact on model output, creating sparser, smaller models with faster inference.

Neural Network Practice Test

Neural Network Practice Test

Neural Network Neural Network Applications and Frameworks 2