Researchers at NeurIPS 2020 introduce new alternatives to backpropagation that aim to solve the problem of catastrophic forgetting in AI models. These proposed continuous learning methods mimic human memory to allow neural networks to learn new information without losing past knowledge.