Enhancing Deep Reinforcement Learning: A Tutorial on Generative Diffusion Models in Network Optimization
Generative Diffusion Models (GDMs) have emerged as a transformative force in the realm of Generative Artificial Intelligence (GAI), demonstrating their versatility and efficacy across a variety of applications. The ability to model complex data distributions and generate high-quality samples has made GDMs particularly effective in tasks such as image generation and reinforcement learning. Furthermore, their iterative nature, which involves a series of noise addition and denoising steps, is a powerful and unique approach to learning and generating data.
This repository contains the code accompanying the paper published in IEEE COMST:
"Enhancing Deep Reinforcement Learning: A Tutorial on Generative Diffusion Models in Network Optimization"
Authored by Hongyang Du, Ruichen Zhang, Yinqiu Liu, Jiacheng Wang, Yijing Lin, Zonghang Li, Dusit Niyato, Jiawen Kang, Zehui Xiong, Shuguang Cui, Bo Ai, Haibo Zhou, and Dong In Kim, accepted by IEEE Communications Surveys & Tutorials.
The paper can be found at ArXiv or IEEE.
We initiate our discussion with the foundational knowledge of GDM and the motivation behind their applications in network optimization. This is followed by exploring GDM’s wide applications and fundamental principles and a comprehensive tutorial outlining the steps for using GDM in network optimization. In the context of intelligent networks, we study the impact of GDM on algorithms, e.g., Deep Reinforcement Learning (DRL), and its implications for key scenarios, e.g., incentive mechanism design, Semantic Communications(SemCom), Internet of Vehicles (IoV) networks, channel estimation, error correction coding, and channel denoising. We conclude our tutorial by discussing potential future research directions and summarizing the key contributions.
GDM training approaches with and without an expert dataset. Part A illustrates the GDM training scenario when an expert database is accessible. The process learns from the GDM applications in the image domain: the optimal solution is retrieved from the expert database upon observing an environmental condition, followed by the GDM learning to replicate this optimal solution through forward diffusion and reverse denoising process. Part B presents the scenario where no expert database exists. In this case, GDM, with the assistance of a jointly trained solution evaluation network, learns to generate the optimal solution for a given environmental condition by actively exploring the unknown environment.
In this part, we representatively formulate an optimization problem in a wireless network and show a step-bystep tutorial to solve it by using GDMs.
Consider a wireless communication network where a base station with total power P_T serves a set of users over multiple orthogonal channels. The objective is to maximize the sum rate of all channels by optimally allocating power among the channels. Let g_n denote the channel gain for the n_th channel and p_n denote the power allocated to that channel. The sum rate of all M orthogonal channels is given by the sum of their individual rates. Let the noise level be set as 1 without loss of generality for the analysis. The optimization goal is to find the power allocation scheme {p_1, ..., p_M} that maximizes the sum rate C under the power budget and the non-negativity constraints as:
The dynamic nature of the wireless environment presents a significant challenge, as the values of the channel gains, denoted as {g_1, ..., g_M}, can fluctuate within a range. Therefore, our objective is, given a set of environmental variables as a condition, to use GDM to denoise the Gaussian noise into the corresponding optimal power allocation scheme under this condition.
Here, we consider M= 100. Specifically, the first 50 channels are in good quality and the last channels are in deep fadings.
def state(self):
# Provide the current state to the agent
states1 = np.random.uniform(13, 14, 50)
states2 = np.random.uniform(0, 0.1, 50)
states = np.concatenate([states1, states2])
self._laststate = states
return statesTo create a new conda environment, execute the following command:
conda create --name gdmopt python==3.8
Activate the created environment with:
conda activate gdmoptThe following package can be installed using pip:
pip install tianshou==0.4.11
pip install matplotlib==3.7.3
pip install scipy==1.10.1Run main.py in the file Main to start the program.
For the considered case, in env/utility.py, please set
actions = torch.abs(actions)To use the software version, place the two .py files from the Software folder into the main directory (replacing the current main file)
When is model is training, the following command can be used for checking:
tensorboard --logdir .
After the model is well-trained, the following command can be used for inference:
python main.py --watch --resume-path log/default/diffusion/Jul10-142653/policy.pthA. Note that the power allocation problem we consider here is a highly simplified one. In such cases, the performance of GDM is not always superior to DRL. For more realistic optimization problems (such as decision problems involving state transitions), considering combining GDM with DRL could be worthwhile, as is explored in our D2SAC code and paper:
"Generative AI-aided Optimization for AI-Generated Content (AIGC) Services in Edge Networks."
Here, the total utility of all users, which is designed as the objective function to be maximized, can only be calculated after a long period of the allocation process. As a result, a decision-making process, such as allocating user tasks to desired servers, has to be modeled by forming a Markov chain.
B. The relationship between GDMs and DRL in intelligent network optimization is not just the substitution or competition but rather a compliment and/or supplement of each other that allows for mutual enhancement and learning. In situations where expert strategies are not available for guidance, GDM can leverage a solution evaluation network during the training phase. This is like the Q-network commonly used in DRL. The solution evaluation network estimates the quality of a given solution, e.g., the power allocation scheme in the discussed example, under specific environmental conditions. This quality assessment guides the GDM during its iterative denoising process. Moreover, other advanced techniques from the DRL field can be adopted to make GDM training even more efficient. For example, the double Q-learning technique, which aims at reducing over-estimation in Q-learning, can be adopted. This approach maintains two Q-networks, using the smaller Q-value for updates, thus offering a conservative estimate and mitigating over-optimistic solution assessments. Incorporating such methods can augment GDM training, promoting robustness and efficiency.
C. Based on recent feedback from researchers employing our network optimization framework, 80% of encountered issues—such as actions getting stuck at boundaries or falling into local optima—are related to external networks and environments aiding diffusion method, including inherent structures like those in DDPG. We advise against directly applying our optimization framework as is, instead, consider using an MLP as the decision output within your own decision-making framework (whether it involves various DRL architectures or multi-agent decision systems). Ensure your algorithm converges before incorporating the diffusion module into your project to facilitate the transition from MLP to diffusion architecture, rather than persistently adjusting parameters. Based on the feedback we received, most issues were resolved using this approach, resulting in performance improvements. However, given the inherent energy consumption issues with diffusion, whether to adopt it should be considered based on the specific problem at hand.
Please refer to our tutorial paper for more details.
If our diffusion-based method can be used in your paper, please help cite:
@article{du2023beyond,
title={Enhancing Deep Reinforcement Learning: A Tutorial on Generative Diffusion Models in Network Optimization},
author={Du, Hongyang and Zhang, Ruichen and Liu, Yinqiu and Wang, Jiacheng and Lin, Yijing and Li, Zonghang and Niyato, Dusit and Kang, Jiawen and Xiong, Zehui and Cui, Shuguang and Ai, Bo and Zhou, Haibo and Kim, Dong In},
journal={IEEE Communications Surveys and Tutorials},
year={2024}
}