Few-shot learning using pre-training and shots, enriched by pre-trained samples
This is an incremental improvement for few-shot learning in image recognition, specifically on the EMNIST dataset.
The paper tackles few-shot learning on handwritten digits by pre-training a neural network on a subset of digits and then fine-tuning with a method that freezes the first layer and uses a mix of new and known digits per shot, achieving about 90% accuracy after 10 shots.
We use the EMNIST dataset of handwritten digits to test a simple approach for few-shot learning. A fully connected neural network is pre-trained with a subset of the 10 digits and used for few-shot learning with untrained digits. Two basic ideas are introduced: during few-shot learning the learning of the first layer is disabled, and for every shot a previously unknown digit is used together with four previously trained digits for the gradient descend, until a predefined threshold condition is fulfilled. This way we reach about 90% accuracy after 10 shots.