In few-shot learning, simply tuning initial parameters ($$\theta_0$$) using gradient descent on a small dataset can easily lead to overfitting. To mitigate this, regularization techniques are used during fine-tuning. These techniques include early-stopping, selectively updating parameters, updating related parts of $$\theta_0$$ together, and employing a model regression network.

University of Notre Dame

Google

When using $D_{train}$ to refine $\theta_0$, there’re several techniques can be used to improve results:

1)	Fine-Tuning Existing Parameter by Regularization

2)	Aggregating a Set of Parameters.

3)	Fine-Tuning Existing Parameter with New Parameters.


Strategy of Fine Tuning Parameters 

Wang, Y., Yao, Q., Kwok, J. T., & Ni, L. M. (2020). Generalizing from a few examples: A survey on few-shot learning. ACM computing surveys (csur), 53(3), 1-34.

Generalizing from a Few Examples: A Survey on Few-Shot Learning

When a suitable initial parameter set $$\theta_0$$ is not available, but parts of $$\theta_0$$ are accessible from related pre-trained models, these parts can be aggregated to form a suitable $$\theta_0$$. This aggregated parameter set can then be used directly or fine-tuned using the training dataset $$D_{train}$$.

Aggregating a Set of Parameters for Fine-Tuning

Fine-Tuning Existing Parameters by Regularization

In few-shot learning, when the initial parameter set $$\theta_0$$ alone is not enough to encode the input of a new task, new parameters can be added to the model. The model then fine-tunes the existing $$\theta_0$$ while simultaneously learning the new parameters.

Learn Before

Related