What is Donut and why is OCR-free approach used?
Machine Learning / AI
Опишите цикл работы агента: thought → action → observation → ...
Where is the amount of data known from?
Tell us more about your end-to-end experience: what stages did you perform from task setting to deploying the model in production?
Можно ли считать новую модель с F1-score 0.81 лучше модели с 0.8 и сразу запускать в продакшен?
What is the difference between NaN and None in Python?
What is bias-variance trade-off?
How to compare floating-point coordinates considering the peculiarities of float in different languages?
How can classical ML be used to improve product recognition?
What is 6D pose estimation of an object?
Where are the positional embeddings located architecturally?
What are the quality assessment metrics for audio generation (FAD, MOS)?
What is the CLIP architecture?
What method of a promise is used to get the result of the first completed request?
Were there any other Data Science/ML specialists at PIK with whom knowledge could be exchanged?
There is gradient boosting and random forest with 1000 trees. What will happen to the quality of each model if the first built tree is removed?
What is ONNX Runtime and what are its advantages?
What are the features of active learning in NLP tasks?
What is shared memory and why use it?
Что такое ERR (Expected Reciprocal Rank)?