Over time, we expect our models to approach an idealized set of "most accurate" weights. This process of moving towards the optimal performance is known as convergence. In contrast, divergence is the situation where a model veers off course and its performance deteriorates progressively over time.
