10+ Bhatia Insurance Services Interview Questions and Answers

Small change in one dimension causing drastic difference in model output. Explanation and solution.

• This is known as sensitivity to input

• It can be caused by non-linearities in the model or overfitting

• Regularization techniques can be used to reduce sensitivity

• Cross-validation can help identify overfitting

• Ensemble methods can help reduce sensitivity

• It is generally a bad thing as it indicates instability in the model

Optimizers are used to improve the efficiency and accuracy of the training process in machine learning models.

• Optimizers help in finding the optimal set of weights for a given model by minimizing the loss function.

• They use various techniques like momentum, learning rate decay, and adaptive learning rates to speed up the training process.

• Optimizers also prevent the model from getting stuck in local minima and help in generalizing the model to unseen data.

• Examples of optimizers...read more

LSTMs are better than RNNs due to their ability to handle long-term dependencies.

• LSTMs have a memory cell that can store information for long periods of time.

• They have gates that control the flow of information into and out of the cell.

• This allows them to selectively remember or forget information.

• Vanilla RNNs suffer from the vanishing gradient problem, which limits their ability to handle long-term dependencies.

• LSTMs can be used in applications such as speech recognition, la...read more

P-values are a statistical measure that helps determine the likelihood of obtaining a result by chance.

• P-values range from 0 to 1, with a smaller value indicating stronger evidence against the null hypothesis.

• A p-value of 0.05 or less is typically considered statistically significant.

• P-values are commonly used in hypothesis testing to determine if a result is statistically significant or not.

A logarithm is a mathematical function that measures the relationship between two quantities.

• Logarithms are used to simplify complex calculations involving large numbers.

• They are used in linear algebra to transform multiplicative relationships into additive ones.

• Logarithms are also used in data analysis to transform skewed data into a more normal distribution.

• Common logarithms use base 10, while natural logarithms use base e (approximately 2.718).

Gradients are the changes in values of a function with respect to its variables.

• Gradients are used in calculus to measure the rate of change of a function.

• They are represented as vectors and indicate the direction of steepest ascent.

• Gradients are used in optimization problems to find the minimum or maximum value of a function.

• They are also used in physics to calculate the force acting on a particle.

• Gradients can be calculated using partial derivatives.

Boosting and bagging are ensemble learning techniques used to improve model performance.

• Bagging involves training multiple models on different subsets of the data and averaging their predictions.

• Boosting involves training multiple models sequentially, with each model focusing on the errors of the previous model.

• Bagging reduces variance and overfitting, while boosting reduces bias and underfitting.

• Examples of bagging algorithms include Random Forest and Bootstrap Aggregating, ...read more

Slope and gradient are both measures of the steepness of a line, but slope is a ratio while gradient is a vector.

• Slope is the ratio of the change in y to the change in x on a line.

• Gradient is the rate of change of a function with respect to its variables.

• Slope is a scalar value, while gradient is a vector.

• Slope is used to describe the steepness of a line, while gradient is used to describe the direction and magnitude of the change in a function.

• Example: The slope of a line wi...read more

Pooling in CNNs has learning but reduces spatial resolution.

• Pooling helps in reducing overfitting by summarizing the features learned in a region.

• Max pooling retains the strongest feature in a region while average pooling takes the average.

• Pooling reduces the spatial resolution of the feature maps.

• Pooling can also help in translation invariance.

• However, too much pooling can lead to loss of important information.

KNN during training stores all the data points and their corresponding labels to use for prediction.

• KNN algorithm stores all the training data points and their corresponding labels.

• It calculates the distance between the new data point and all the stored data points.

• It selects the k-nearest neighbors based on the calculated distance.

• It assigns the label of the majority of the k-nearest neighbors to the new data point.

Correlation is a statistical measure that shows how two variables are related to each other.

• Correlation measures the strength and direction of the relationship between two variables.

• It ranges from -1 to 1, where -1 indicates a perfect negative correlation, 0 indicates no correlation, and 1 indicates a perfect positive correlation.

• Correlation does not imply causation, meaning that just because two variables are correlated, it does not necessarily mean that one causes the other...read more