🚀 AI Interview Questions with Answers (Part 7)
61. What is batch size, and how does it affect training?
Batch size is the number of training samples processed before the model updates its weights.
Effects of batch size:
- Small batch size: Uses less memory, updates weights more frequently, but training can be noisier. - Large batch size: Faster training on GPUs and more stable updates, but requires more memory.
Common batch sizes are 32, 64, 128, and 256.
---
62. What are epochs, and how many are typically required?
An epoch is one complete pass of the entire training dataset through the model.
For example, if a dataset has 10,000 samples, one epoch means the model has processed all 10,000 samples once.
The required number of epochs depends on the dataset and model complexity. Too few epochs may cause underfitting, while too many may lead to overfitting.
---
63. What is dropout, and how does it prevent overfitting?
Dropout is a regularization technique where a random percentage of neurons is temporarily deactivated during training.
Benefits:
- Prevents overfitting - Reduces dependency on specific neurons - Improves model generalization
Typical dropout values range from 0.2 to 0.5.
---
64. What is batch normalization, and why is it used?
Batch normalization normalizes the inputs of each layer during training.
Advantages:
- Faster convergence - More stable training - Allows higher learning rates - Reduces the chances of vanishing or exploding gradients
It is commonly used in deep neural networks and CNNs.
---
65. What are Convolutional Neural Networks (CNNs), and where are they used?
Convolutional Neural Networks (CNNs) are deep learning models specifically designed for processing image and visual data.
They automatically learn features such as edges, textures, and shapes using convolutional layers.
Applications:
- Image classification - Face recognition - Medical image analysis - Object detection - Autonomous vehicles
---
66. What are the common applications of CNNs?
CNNs are widely used in computer vision tasks, including:
- Facial recognition - Image classification - Object detection - Medical diagnosis from X-rays and MRI scans - OCR (Optical Character Recognition) - Video analysis - Self-driving cars - Industrial quality inspection
---
67. What are Recurrent Neural Networks (RNNs), and how do they work?
Recurrent Neural Networks (RNNs) are neural networks designed to process sequential data by maintaining information from previous inputs.
Unlike traditional neural networks, RNNs have memory, making them suitable for tasks where the order of data matters.
Applications:
- Language translation - Speech recognition - Time-series forecasting - Text generation
---
68. What are Long Short-Term Memory (LSTM) networks?
LSTM is an advanced type of Recurrent Neural Network designed to overcome the vanishing gradient problem.
It uses memory cells and gates (input, forget, and output gates) to retain important information over long sequences.
Applications:
- Machine translation - Chatbots - Stock price prediction - Speech recognition - Text generation
---
69. What are Gated Recurrent Units (GRUs), and how do they differ from LSTMs?
GRUs are a simplified version of LSTMs that use fewer gates, making them computationally more efficient.
GRU vs LSTM:
- GRUs have 2 gates, while LSTMs have 3 gates. - GRUs train faster and require fewer parameters. - LSTMs generally perform better on very long sequences.
Both are commonly used for sequence modeling tasks.
---
70. What are Transformer models, and why are they important?
Transformer models are deep learning architectures that process input data using a self-attention mechanism instead of sequential processing.
Advantages:
- Process sequences in parallel - Handle long-range dependencies efficiently - Faster training than RNNs - State-of-the-art performance in NLP
Popular Transformer models:
- BERT - GPT - T5 - LLaMA - Gemini
Double Tap ❤️ For Part-8