What are the drawbacks of training a deep neural network with a small amount of labeled data ? O The model does not converge because the training is not stable. O Training error becomes very high and,...

PLEASE ENSURE ANSWERS ARE CORRECT AND DONT JUST GUESS

Extracted text: What are the drawbacks of training a deep neural network with a small amount of labeled data ? O The model does not converge because the training is not stable. O Training error becomes very high and, testing and validation errors become very low due to high bias (under fitting). Because the data set is small and training is very quick, the Inference during test time is also very quick. Training errors may be very low but test and validation errors may be very large due to over-fitting, Consider a deep neural network (fruit network) that has been trained to classify a labelled set of Images of fruits based on their species (say C, categories). You are presented with a tiny labeled dataset of images of vegetables (a few thousand images with C categories). The goal is to train a model to classify images of vegetables,. Which of the below approaches would work well? Apply a feature extractor like SIFT or HoG on the vegetable images and use the features to train a Support Vector Machine classifier. The last layer of the fruit network can be replaced with a classification layer based on the number of vegetable categories and the network can be fine-tuned with the tiny labeled set of images of vegetables. O Retrain the fruit network with images of vegetables. O The fruit network can be used without any modifications to classify images of vegetables since it has already been trained on the labelled images of fruits and fruits look similar to vegetables.