Evaluating alignment between humans and neural network representations in image-based learning tasks