Machine Learning / AI
What is Dropout and how does it behave during training and inference?
Which algorithm complexity depends on n or k and what does it depend on?
What is gradient boosting? Where does the gradient appear in it?
What is CTC loss in OCR?
What are TPR and FPR? How are they related to precision and recall?
How does a random forest reduce the influence of individual trees on the final prediction?
What are alternatives to CLIP (SigLIP, EVA-CLIP)?
What is the difference between self-attention and cross-attention, and what is each used for?
Why did you decide to change your job?
How to obtain embeddings from text that will be used as input to a transformer?
Tell about multi-head attention in detail: what are Q/K/V, how are attention weights calculated, why are there multiple heads?
How to evaluate the quality of a retriever? Where to get a collection of documents for metric evaluation?
How to use BERT in a SQuAD-style QA task?
What are the problems with the Sigmoid activation function? Where is it used and how is it interpreted?
How do you determine if a model is underfitted or overfitted?
What is an OOV token and how does subword tokenization handle it?
Which metric is more important: precision or recall depending on the task?
What are the requirements of the EU AI Act for high-risk systems?
What is a p-value?