Diffusion-DICE: In-Sample Diffusion Guidance for Offline Reinforcement Learning

One important property of DIstribution Correction Estimation (DICE) methods is that the solution is the optimal stationary distribution ratio between the optimized and data collection policy. In this work, we show that DICE-based methods can be viewed as a transformation from the behavior distribution to the optimal policy distribution. Based on this, we propose a novel approach, Diffusion-DICE, that directly performs this transformation using diffusion models. We find that the optimal policy's score function can be decomposed into two terms: the behavior policy's score function and the gradient of a guidance term which depends on the optimal distribution ratio. The first term can be obtained from a diffusion model trained on the dataset and we propose an in-sample learning objective to learn the second term. Due to the multi-modality contained in the optimal policy distribution, the transformation in Diffusion-DICE may guide towards those local-optimal modes. We thus generate a few candidate actions and carefully select from them to approach global-optimum. Different from all other diffusion-based offline RL methods, the guide-then-select paradigm in Diffusion-DICE only uses in-sample actions for training and brings minimal error exploitation in the value function. We use a didatic toycase example to show how previous diffusion-based methods fail to generate optimal actions due to leveraging these errors and how Diffusion-DICE successfully avoids that. We then conduct extensive experiments on benchmark datasets to show the strong performance of Diffusion-DICE. Project page at https://ryanxhr.github.io/Diffusion-DICE/.

Paper

References (58)

Scroll for more · 38 remaining

Similar papers

Peer review

Reviewer UuBV8/10 · confidence 4/52024-07-10

Summary

The paper mainly addresses the problem of obtaining the optimal policies in the distribution correction estimation (DICE) setting, which is one popular offline RL approach. Note that the offline RL assumes that we can’t interact with an environment and only have access to a dataset—sets of (state, action, and next state) tuples collected by some behavior policy. For the given history, DICE estimates the optimal value functions $V^*$ and $Q^*$ and obtains the optimal stationary distribution ratio $w^*$, which is the rate between the optimal policy’s state occupancy distribution $d^*$ and the one of the behavior policy $d^D$. Even if it is possible to get the optimal value functions and the ratio via DICE, obtaining the corresponding optimal policy $\pi^*$ from them is still challenging. In this context, the paper proposes a novel training method for diffusion-based models to learn the optimal policy from the DICE-driven optimal value functions and the ratio. Unlike common density modeling, learning the diffusion-based models from the optimal value functions is not straightforward. Such a problem is closer to variational inference than density modeling; specifically, we don’t have samples from the target distribution to diffuse. Moreover, unlike a typical variational inference problem, sampling from the models is not available due to the nature of the offline RLs. To do that the paper, the paper first introduces that the optimal policy can be represented by a product of the optimal stationary distribution ratio $w^*$ and the behavior policy $\pi^D$. Since the product of two distributions is unnormalized, the complete expression includes the normalization constant, which is the expected value of $w^*$ under the behavior policy (Line 155). Next, the paper assumes that the behavior policy can be represented by a diffusion-based model with the forward process, which will also be used to perturb the policy of interest. Then, the authors show that the optimal policy at the perturbation level $t$ can be represented by using the behavior’s perturbed distribution at the same $t$ and the DICE ratio $w^*$ (Equation 6). In particular, this representation includes the logarithm of the expectation of the DICE ratio $w^*$ under the posterior distribution of the clean action $a_0$ for a given perturbed action $a_t$. While this novel representation doesn’t require sampling from the model policy anymore, computing the logarithm of the Monte Carlo estimate of any expectation is biased in general. To circumvent this, the authors propose using the tangent transform, one of the variational approximation methods; thus, the quantity inside the logarithm is represented by an optimization problem, as in Equation 7. Interestingly, the new training objective only requires sampling from the behavior policy, which is suitable for the offline RLs. In summary, the paper introduces a new representation of the diffusion-based optimal policy model by using the diffusion-based behavior policy and time-dependent learnable terms that will be learned by the convex problem described in Equation 7. The authors refer to this approach as In-sample Guidance Learning (IGL). While there have been a few previous approaches to learning diffusion-based optimal policy models, the authors point out that the IGL shows some benefits. For example, IGL only requires a single sample from the behavior policy, which can be obtained from the history, while some previous approaches require more than one sample from the behavior, which is not favorable in the offline RLs. In addition, the authors suggest a few techniques to improve the stability of IGLs, such as using piecewise $f$-divergence for the DICE regularization term. Finally, the paper discusses some failure cases of previous approaches and how the proposed method bypasses them. It also demonstrates the efficacy of the proposed method via experiments on several benchmark datasets. ----------------------------------------------------- Updated the rating from 7 to 8 after the authors' rebuttal.

Strengths

In my understanding, the paper's contributions are clear, and I also consider that the results are essential for several reasons. The paper introduces a novel representation of the diffusion-based optimal policy model and its training method, IGL. In addition, the paper also motivates the solution well. For example, this approach circumvents the drawbacks of previous approaches, which require more than one sample from the behavior policy, and such actions may be out-of-distribution of the environments. Moreover, the authors provide extensive discussions to help potential readers comprehend the characteristics of previous approaches and the proposed method. Finally, the paper demonstrates the effectiveness of the proposed method via various experiments, which further supports the authors' claim.

Weaknesses

Overall, the paper presents a novel method for learning optimal policy in DICE, which is a valuable contribution to the field. However, improvements in the presentation would greatly enhance the clarity and comprehensibility of the manuscript. In particular, several equations within the paper omit the definitions of variables, which can lead to confusion—for example, $a_0$ in Line 119. In addition, some variables overlap while they are independent. For instance, in Line 155, the variable of the integration overlaps with the $a$ in the numerator. I recommend revising the paper to address these issues.

Questions

N/A

Rating

8

Confidence

4

Soundness

4

Presentation

3

Contribution

4

Limitations

N/A

Reviewer uMN36/10 · confidence 3/52024-07-11

Summary

The paper introduces a novel offline reinforcement learning approach that leverages diffusion models integrated with DICE-based methods. The proposed guide-then-select paradigm aims to minimize error exploitation. The resultant algorithm achieves state-of-the-art performance on D4RL benchmark tasks.

Strengths

- Well-motivated and novel integration of diffusion models with DICE-based methods - Well-written theoretical justification for the approach - Strong empirical results that improve upon prior diffusion-policy baselines

Weaknesses

- Missing comparison in Table 1 of more optimal Gaussian policy methods, e.g. EDAC [1] - Selection of D4RL datasets is limited, e.g. what about expert/random datasets? Further environments like Adroit would also be interesting - Discussion of hyperparameter choice in the Appendix is important and should be included in the experimental section. - A comparison of the inference speed of Diffusion-DICE and prior baselines would be valuable. Minor: - Line 70: “M” -> “M=” - Line 73: LP abbreviation not explained - There is concurrent related work [2] which also performs an analogous transformation between the behavior distribution to an online policy with diffusion models for synthetic data generation. [1] Uncertainty-Based Offline Reinforcement Learning with Diversified Q-Ensemble. Gaon An, Seungyong Moon, Jang-Hyun Kim, Hyun Oh Song. NeurIPS, 2021. [2] Policy-Guided Diffusion. Matthew Thomas Jackson, Michael Tryfan Matthews, Cong Lu, Benjamin Ellis, Shimon Whiteson, Jakob Foerster. RL Conference, 2024.

Questions

Please see the above weaknesses.

Rating

6

Confidence

3

Soundness

3

Presentation

4

Contribution

3

Limitations

Limitations are discussed in the conclusion.

Reviewer QJqE6/10 · confidence 3/52024-07-12

Summary

This paper introduces Diffusion-DICE for offline reinforcement learning. Diffusion-DICE motivates from the transformation between the behaviour distribution and the optimal distribution, which inspires the use of generative models for behaviour distribution modelling. Next, Diffusion-DICE decomposes the policy score function into two components, one from the behaviour distribution, another from the guidance, i.e., the transformation. Lastly, Diffusion-DICE employs a guide-then-select paradigm, which uses only in-sample actions for training to avoid out-of-distribution issues. In their experiments, Diffusion-DICE has achieved strong performance compared with baselines on the D4RL benchmark.

Strengths

The proposed idea is novel and interesting. I like the way the authors connect DICE with diffusion policies, decompose the score functions, and make the guidance score tractable. Theoretically, they have provided careful analysis and derivations to support the claims. Empirically, they have conducted both toy experiments for intuitive understanding, and demonstrating the strong performance on the D4RL benchmark.

Weaknesses

There are several weaknesses of the paper I’d like to point out. 1/ The biggest issue is the presentation. Although I do like the idea of the work and recognise its contributions, I found the paper very hard to follow and needed to read through the paper to understand the introduction. The way authors presented the guide-then-select is confusing. I’d suggest authors provide a bit more background and carefully define the “guidance term”, and how it relates to the RL before using it in both abstract and introduction. 2/ Achieving in-sample learning of offline diffusion RL is not new. Efficient Diffusion Policy (EDP) [1] has introduced an IQL-based variant which naturally allows training Q-values using only in-sample data, without querying out-of-distribution actions during policy evaluation. I’d suggest the authors carefully check the claims and avoid over claiming. 3/ The D4RL experiments are only conducted on the locomotion tasks and the antmaze tasks. The commonly tested kitchen and adroit tasks are missing, which weakens the claim of the paper. References: [1] Kang, B., Ma, X., Du, C., Pang, T., & Yan, S. (2024). Efficient diffusion policies for offline reinforcement learning. Advances in Neural Information Processing Systems, 36.

Questions

Could you provide a bit more discussion about the differences of using guidance for inference with the Diffusion-QL style inference, which directly guides the sampling process towards actions with high returns? It would be interesting to understand the pros and cons of these two different paradigms

Rating

6

Confidence

3

Soundness

3

Presentation

2

Contribution

3

Limitations

I think in general this is an interesting work and I don’t see major limitations.

Reviewer QJqE2024-08-09

I do appreciate the authors' efforts and detailed explanations. I feel most of my concerns are addressed. I still feel this is a good paper, although certain efforts are still needed for a better presentation. I'll keep my original score and vote for an acceptance.

Authorsrebuttal2024-08-12

Kind Reminder: Discussion Period Ending

Dear Reviewer uMN3, We sincerely apologize for any inconvenience this reminder may cause. We just wanted to kindly remind you that the discussion period will conclude tomorrow. As the discussion period is nearing its end, we wanted to check if you have any remaining questions or concerns. **We would be more than happy to address further inquiries you may have.** We understand how busy you must be during this time, and we truly appreciate the effort and time you've dedicated to the rebuttal process. Thank you very much, and we look forward to your response. Best regards, The Authors of Paper 16567

Program Chairsdecision2024-09-25

Decision

Accept (poster)

© 2026 NYSGPT2525 LLC