Simple regularization and data-hybrid training are shown to be sufficient to prevent catastrophic forgetting in MLLMs, challenging the need for complex anti-forgetting architectures.
arXiv · March 17, 2026 · 2603.14493
The Takeaway
The paper demonstrates that 'easier than you think' recipes (like low learning rates or param constraints) outperform specialized continual learning methods. This simplifies the workflow for developers needing to adapt vision-language models to new tasks while maintaining general capabilities.
From the abstract
The paper demonstrate that simple adjustments of the fine-tuning recipes of multimodal large language models (MLLM) are sufficient to mitigate catastrophic forgetting. On visual question answering, we design a 2x2 experimental framework to assess model performance across in-distribution and out-of-distribution image and text inputs. Our results show that appropriate regularization, such as constraining the number of trainable parameters or adopting a low learning rate, effectively prevents forge