I’m working inside a Jupyter Notebook that contains several deep learning models for IMU fusion and activity recognition. The current accuracy is around 60%, but my target is at least 95%. i need to run the notebook on a high-performance GPU environment (such as an NVIDIA A100, V100, or T4) because the dataset is quite large and each model requires about 50 epochs of training, which takes several hours on CPU. task is to: Review the full deep learning pipeline (data preprocessing, normalization, window segmentation, and fusion setup). Tune the architectures if needed. Optimize training parameters (batch size, learning rate, dropout, and scheduler). Ensure class balance and correct label mapping Add early stopping and performance tracking per epoch. Provide comparisons showing accuracy and F1-score before vs. after your modifications. Goal: An optimized, end-to-end notebook that reaches 95%+ accuracy and runs smoothly on a high-GPU setup. Budget : 70$ Deadline: 1 day