The field of image and speech restoration is rapidly advancing with the development of novel models and techniques. A key direction in this field is the use of universal models that can handle multiple types of distortions and degradations, such as additive noise, reverberation, and band limitation. These models have shown superior performance and practicality compared to traditional methods. Another area of focus is the use of adaptive and efficient networks that can dynamically select the suitable approach based on input severity, making them ideal for resource-constrained devices. The integration of explicit edge priors and masked degradation classification has also been shown to significantly boost performance in image restoration tasks. Furthermore, the use of decorrelated backpropagation has been found to accelerate convergence and reduce computational costs in vision transformer pre-training. Noteworthy papers include: Universal Discrete-Domain Speech Enhancement, which proposes a novel model that redefines speech enhancement as a discrete-domain classification task, and Universal Image Restoration Pre-training via Masked Degradation Classification, which introduces a pre-training method that facilitates the classification of degradation types in input images. Additionally, An Adaptive Edge-Guided Dual-Network Framework for Fast QR Code Motion Deblurring and Pruning Overparameterized Multi-Task Networks for Degraded Web Image Restoration have also made significant contributions to the field. Decorrelation Speeds Up Vision Transformers has also shown promising results in reducing training time and energy use while improving downstream performance for large-scale vision transformer pre-training.