Vision-language pretrained models, such as CLIP, have established new benchmarks in multimodal data mining. In such models, few-shot fine-tuning is a major challenge to achieve optimal performance on both in-distribution (ID) and out-of-distribution (OOD) datasets, especially when labeled data is scarce. Most existing fine-tuning approaches rely on first-order gradient-based optimizers, which typically suffer from slow convergence, sensitivity to step-size hyperparameters, and poor generalization in OOD settings. In contrast, second-order methods utilize local curvature information of the loss landscape to adjust the update step size. This is particularly beneficial for CLIP models, whose non-convex loss functions often contain sharp critical points. In such cases, natural gradient direction can offer more substantial and efficient per-iteration updates when fine-tuning with limited data. Natural Gradient Descent (NGD) is obtained by preconditioning the standard gradient with the inverse Fisher Information Matrix (FIM), which is computationally expensive for large models. To address this, we propose a Bayesian approximation of NGD using a Kalman filter for CLIP models. Our method combines the benefits of second-order optimization with Bayesian inference, which enhances generalization while providing uncertainty quantification. Extensive experiments conducted on diverse image classification datasets demonstrate that our algorithm consistently achieves superior-or comparable-ID performance and improved OOD robustness compared to state-of-the-art baselines. To the best of our knowledge, this work represents the first successful application of Kalman filtering to fine-tuning CLIP-based models, which enables more robust and efficient learning in vision-language tasks.

Safe motion control in unknown environments is one of the challenging tasks in robotics, such as autonomous navigation. Control Barrier Function (CBF), as a strong math-ematical tool, has been widely used in many safety-critical systems to satisfy safety requirements. However, there are only a handful of recent studies on safety controllers with perception inputs. Common assumptions in most of the works are that the CBF is already known and obstacles have predefined shapes. In this work, we introduce a novel Vision-based Control Barrier Function (V-CBF), which enables generalization to new environments and obstacles of arbitrary shapes. We then derive CBF safety conditions over RGB-D space and relate those to actual robot control inputs. To train the CBF function, we introduce a method to generate ground truth with desired properties complying with CBF and a method to generate part of the CBF as an image-to-image translation problem. We finally demonstrate the efficacy of V-CBF on the safe control of an autonomous car in CARLA simulator.