Question Bank
Machine Learning interview questions
Naming a model or a metric is only the start of an interview answer. Explain why it fits the task, and connect training, regularization, validation and evaluation to a concrete example. Start with your own explanation, then state its assumptions and limitations.
Showing 20 of 746 questions
In what cases are different tools and methods applied?
How are sensitive data stored during LLM processing with masking?
Is only PDF content checked for sensitive data or metadata as well?
Where does the golden dataset for testing come from?
How is the selection of system operation modes configured (block, alert, mask)?
How does the visual module for generating car image embeddings work?
Imagine you have images of two different cars. You compute their embeddings and measure the distance between them. What method would you use to determine the preferred car based on these embeddings?
Tell me more about the CatBoost model
What other embeddings are used, for example for color? Is there a separate module?
Were augmentations used for training the visual module?
Give an example of your idea/modification in model architecture that improved results
Give an example of your architectural modification idea that boosted performance
И дальше, что можно делать при сильном дисбалансе классов?
Как выглядит рукоук кривая? 0.5 получается, все рандомный классификатор. Да. А рукоук кривая как выглядит?
Насколько важно, чтобы эти две выборки для этой модели, которая уже обучена, насколько важно, чтобы у них было одинаковое распределение?
Как нам уменьшать дисперсию и смещение за счет особенностей этих архитектур?
What are ensemble methods in machine learning and how can they be used to improve model performance?
Practice in botWhat are the main steps for analyzing requirements and constraints in a machine learning project?
Practice in botPrepare for your next interview with Vibe Interview.
Download Vibe Interview