Forwarded from di
вот тут хорошо разобрали почему так есть https://habrahabr.ru/company/mailru/blog/331696/
Хабр
Табы, пробелы и ваша зарплата — какая связь?
Пару дней назад Дэвид Робинсон опубликовал на Stack Overflow статью с очень провокационным названием: Разработчики, использующие пробелы, зарабатывают больше ис...
Forwarded from Generative Ai
В ИТМО через 10 дней пройдет летняя школа машинного обучения ЦРТ. Учеба бесплатная, организаторы оплачивают билеты и проживание 20 участникам. http://algorythm.tech/
Focal Loss for Dense Object Detection
Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, Piotr Dollár
The highest accuracy object detectors to date are based on a two-stage approach popularized by R-CNN, where a classifier is applied to a sparse set of candidate object locations. In contrast, one-stage detectors that are applied over a regular, dense sampling of possible object locations have the potential to be faster and simpler, but have trailed the accuracy of two-stage detectors thus far. In this paper, we investigate why this is the case. We discover that the extreme foreground-background class imbalance encountered during training of dense detectors is the central cause. We propose to address this class imbalance by reshaping the standard cross entropy loss such that it down-weights the loss assigned to well-classified examples. Our novel Focal Loss focuses training on a sparse set of hard examples and prevents the vast number of easy negatives from overwhelming the detector during training. To evaluate the effectiveness of our loss, we design and train a simple dense detector we call RetinaNet. Our results show that when trained with the focal loss, RetinaNet is able to match the speed of previous one-stage detectors while surpassing the accuracy of all existing state-of-the-art two-stage detectors.
https://arxiv.org/abs/1708.02002
#deeplearning #paper
Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, Piotr Dollár
The highest accuracy object detectors to date are based on a two-stage approach popularized by R-CNN, where a classifier is applied to a sparse set of candidate object locations. In contrast, one-stage detectors that are applied over a regular, dense sampling of possible object locations have the potential to be faster and simpler, but have trailed the accuracy of two-stage detectors thus far. In this paper, we investigate why this is the case. We discover that the extreme foreground-background class imbalance encountered during training of dense detectors is the central cause. We propose to address this class imbalance by reshaping the standard cross entropy loss such that it down-weights the loss assigned to well-classified examples. Our novel Focal Loss focuses training on a sparse set of hard examples and prevents the vast number of easy negatives from overwhelming the detector during training. To evaluate the effectiveness of our loss, we design and train a simple dense detector we call RetinaNet. Our results show that when trained with the focal loss, RetinaNet is able to match the speed of previous one-stage detectors while surpassing the accuracy of all existing state-of-the-art two-stage detectors.
https://arxiv.org/abs/1708.02002
#deeplearning #paper
Интересное в мире DS/ML (в этот раз негусто):
- Andrew Ng открыл новый курс на Курсере про Deep Learning - ссылка выше (было)
- Facebook полностью перешел на нейросети для перевода - https://goo.gl/VhjA9H
- Pytorch v0.2 - https://goo.gl/99QpB3 - может кому-то актуально
- Люди из Salesforce тренируют LTSM сети для перевода с одного языка на другой, берут получившийся декодер (часть сети) и используют его чтобы скормить его выход другим нейросетям, которые делают более простые вещи - https://goo.gl/6Kt77o
#data_science
- Andrew Ng открыл новый курс на Курсере про Deep Learning - ссылка выше (было)
- Facebook полностью перешел на нейросети для перевода - https://goo.gl/VhjA9H
- Pytorch v0.2 - https://goo.gl/99QpB3 - может кому-то актуально
- Люди из Salesforce тренируют LTSM сети для перевода с одного языка на другой, берут получившийся декодер (часть сети) и используют его чтобы скормить его выход другим нейросетям, которые делают более простые вещи - https://goo.gl/6Kt77o
#data_science
Facebook Code
Transitioning entirely to neural machine translation
More than 4.5 billion translations every day are now powered by neural networks, which provide better accuracy and fluency than previous models.