This work proposes enhancing image classification performance by averaging weights from models trained via exploratory trajectory learning and local trajectory learning for initialization. Two models are independently trained, and their weights are av...
This work proposes enhancing image classification performance by averaging weights from models trained via exploratory trajectory learning and local trajectory learning for initialization. Two models are independently trained, and their weights are averaged to form a new starting point for re-adaptive learning. This strategy leverages distinct optimization paths, guiding the model toward flatter loss regions that improve test accuracy. Experiments on CIFAR-10, CIFAR-100, and Tiny-ImageNet with ResNet, VGG, AlexNet, and ViT architectures show consistent improvements over baselines, with ResNet34 achieving a 2.92% gain on CIFAR-100. These findings demonstrate that the heterogeneous based weight fusion and re-adaptive learning strategy is an effective method for enhancing model performance without additional inference costs.