{"description":"Assess your knowledge of Stochastic Gradient Descent, its efficiency advantages, batch processing, and momentum enhancements.","questions":[{"answer":"SGD updates parameters using a subset of data, while Gradient Descent uses the full dataset","number":1,"options":["SGD uses all data at once, while Gradient Descent uses batches","SGD updates parameters using a subset of data, while Gradient Descent uses the full dataset","SGD requires no learning rate","SGD is only used for classification problems"],"question":"What is the key difference between Stochastic Gradient Descent (SGD) and traditional Gradient Descent?"},{"answer":"To reduce computational cost and improve efficiency","number":2,"options":["To increase the loss","To simplify the loss function","To reduce computational cost and improve efficiency","To avoid using gradients"],"question":"What is the purpose of using batches in SGD?"},{"answer":"It introduces randomness by using randomly selected data batches","number":3,"options":["It uses a constant step size","It updates parameters without using gradients","It introduces randomness by using randomly selected data batches","It avoids calculating the loss"],"question":"Why is SGD considered 'stochastic'?"},{"answer":"Introduces noise but may help escape local minima","number":4,"options":["Decreases noise in parameter updates","Increases training time","Introduces noise but may help escape local minima","Requires fewer epochs"],"question":"What is the effect of using a small batch size in SGD?"},{"answer":"Stabilizes convergence and dampens oscillations","number":5,"options":["Reduces computation speed","Enhances model overfitting","Stabilizes convergence and dampens oscillations","Avoids the need for loss function calculations"],"question":"What does SGD with Momentum (SGDM) help to solve?"},{"answer":"It determines how much of the previous gradient is retained","number":6,"options":["It increases the batch size","It determines how much of the previous gradient is retained","It adjusts the learning rate","It scales the input features"],"question":"What role does the momentum coefficient (typically 0.9) play in SGDM?"},{"answer":"SGDM helps smooth parameter updates by remembering previous gradients","number":7,"options":["SGDM uses less memory","SGDM helps smooth parameter updates by remembering previous gradients","SGDM does not require a learning rate","SGDM only works with small datasets"],"question":"Which of the following is TRUE about SGDM vs SGD?"},{"answer":"Model parameters (e.g., weights)","number":8,"options":["Feature scaling","The loss function definition","Model parameters (e.g., weights)","Dataset values"],"question":"In SGD, what is typically updated at each batch iteration?"},{"answer":"It performs frequent updates using smaller data batches","number":9,"options":["It uses complex models","It avoids loss calculations","It performs frequent updates using smaller data batches","It skips derivative calculations"],"question":"Why can SGD converge faster than traditional Gradient Descent?"},{"answer":"Effective learning and rapid convergence","number":10,"options":["Poor model performance","Ineffective learning rate","Effective learning and rapid convergence","Excessive overfitting"],"question":"What does a sharp decrease in the loss function over initial epochs indicate?"}],"title":"Stochastic Gradient Descent"}