Few-Shot Learning Networks: Optimization Techniques and Trends

Most modern machine learning systems require to be trained over a large set of data. This is useful when there is readily large amounts of data. In recent years, few-shot learning has been proposed to circumvent this issue. Instead, the machine is given limited amounts of data and can make accurate predictions. Although the initial data may be limited, data can be artificially developed through data augmentation. Unfortunately, few-shot learning systems are not at the state where they can be utilized reliably. There is possibilities for these systems to be optimized through implementing dropout and hyperparameter tuning. This study is designed to analyze any trends and techniques that may allow for higher performance generally.