Incremental learning (IL) seeks to acquire new knowledge (plasticity) while preserving previously learned information (stability), yet balancing the two remains a major challenge for intelligent systems. The human brain achieves this balance through multiple memory units that jointly encode and store information. Inspired by this mechanism, we introduce Co2In, an IL framework that incorporates knowledge encoding, consolidation, and integration. Co2In adopts a bi-level memory architecture: a working memory for adaptive acquisition of new knowledge and a long-term memory for persistent retention. During learning the two modules cooperate. The working memory first learns from new data, encoding knowledge into model parameters; then a consolidation step extracts underlying patterns from these parameters and re‑expresses them as a compact knowledge representation. This representation together with the identified patterns are transformed into separate network layers and merged into the long-term memory. These designs enable Co2In to accumulate knowledge with high plasticity and stability. We evaluate Co2In on CIFAR‑10, CIFAR‑100 and Tiny‑ImageNet under exemplar‑free class‑IL and task‑IL scenarios. Results show that Co2In attains state‑of‑the‑art performance while using memory efficiently. Review