认知重构。

Abstract: Modern AI language models often "hallucinate," meaning they state false information with high confidence. We believe a core reason lies in their fundamental architecture, which forces the model to be decisive even when it lacks sufficient information. To fix this, we've developed the Credal Transformer. It replaces the standard decision-making component with a new mechanism that allows the model to express uncertainty. Instead of always giving one definite answer, it can represent a range of possible answers, with the size of the range indicating its confidence level. Our experiments show this new model is much better at knowing when it doesn't know the answer, significantly reducing confident mistakes and creating a foundation for safer and more reliable AI.

Abstract: Continual or Lifelong Learning aims to develop models capable of acquiring new knowledge from a sequence of tasks without catastrophically forgetting what has been learned before. Existing approaches often rely on storing samples from previous tasks (experience replay) or employing complex regularization terms to protect learned weights. However, these methods face challenges related to data privacy, storage limitations, and performance degradation when tasks are dissimilar. To address these challenges, we introduce MyGO (Memory Yielding Generative Offline-consolidation), a novel lifelong learning framework inspired by the biological wake-sleep cycle. During the "wake" phase, the system rapidly learns a new task and trains a compact generative model (Generative Memory, G-mem) to capture its data distribution. During the "sleep" phase, the system enters an offline state, using all learned G-mem models to generate pseudo-data ("dreams") and consolidate new and old knowledge into a core feature extractor via knowledge distillation. This approach obviates the need to store any raw data, retaining only compact generative models, which offers significant advantages in privacy and storage efficiency. We evaluate MyGO on computer vision (Split-MNIST) and natural language processing (Split-AG News) benchmarks, comparing it against a sequential fine-tuning baseline. The results demonstrate that MyGO significantly mitigates catastrophic forgetting and maintains high average accuracy across tasks, proving the framework's effectiveness and domain-generality.