What is an OOV token and how does subword tokenization handle it?
Machine Learning / AI
How does stemming differ from lemmatization? In which tasks should each be used?
How to profile inference of LLM on GPU (nsight, torch profiler)?
What risks does AutoML pose in terms of interpretability?
What are multispectral and hyperspectral images?
What is redaction and tokenization of PII in datasets?
What are multi-armed bandits and where are they used instead of A/B tests?
Friday, 2.5 hours left until the end of the workday. A business client comes and says there is a meeting at 9:30 on Monday and the task needs to be completed, which takes 5-6 hours. What will you do?
Candidate's questions about work format and team.
What are Medusa and multi-token prediction heads?
What is MIG (Multi-Instance GPU) on A100/H100?
What is a token and how do character-level, word-level, and subword tokenization differ?
What is the mean-field approximation?
Was the data stored in a flat table? Did you develop features yourself or did someone do it for you?
What is targeted vs untargeted attack?
What are the problems when forecasting unstable series (cold start of products)?
What is a confounder and how to account for it?
What is video object segmentation and mask tracking (XMem, SAM2)?
How does sparse retrieval (BM25) differ from dense?
Что такое meta-learners (S-learner, T-learner, X-learner, R-learner)?