Level Up: Defining and Exploiting Transitional Problems for Curriculum Learning
Researchers have developed a new method for curriculum learning that measures problem difficulty based on a series of models with increasing competence. This approach identifies "transitional problems" that become easier as model ability grows. Training models on a curriculum that progresses through these transitional problems from easier to harder instances has shown to be the most efficient way to improve a model to the next level of competence, outperforming other training strategies. AI
IMPACT This new curriculum learning approach could lead to more efficient training of AI models by optimizing the order of training data.