Adaptive Noise Level Masked Diffusion Models

# Adaptive Noise Level Masked Diffusion Models ## Description This project investigates the hypothesis that incorporating adaptive noise levels based on Denoising Entropy could improve performance in non-stationary environments for Masked Diffusion Models (MDMs). ## Research Hypothesis Incorporating adaptive noise levels based on Denoising Entropy can enhance MDM performance in dynamic environments by adjusting to changing noise characteristics. ## Implementation Approach We will develop an adaptive mechanism that modulates noise levels within MDMs in response to real-time Denoising Entropy metrics. This involves: - Designing an adaptive noise mechanism - Integrating it into the MDM framework - Conducting experiments under simulated non-stationary conditions ## Setup Instructions 1. Clone the repository: ```bash git clone https://github.com/your_username/adaptive_noise_mdm.git cd adaptive_noise_mdm ``` 2. Install dependencies: ```bash pip install -r requirements.txt ``` ## Usage - To train the model, run: ```bash python src/train.py ``` - To evaluate the model, run: ```bash python src/evaluate.py ``` ## Expected Results We expect the adaptive noise mechanism to demonstrate improved stability and performance in non-stationary environments compared to static noise level models. ## References - [Optimizing Decoding Paths in Masked Diffusion Models by Quantifying Uncertainty](http://arxiv.org/abs/2512.21336v1)