Diffusion fashions have just lately emerged because the de facto customary for producing complicated, high-dimensional outputs. It’s possible you’ll know them for his or her means to provide beautiful AI artwork and hyper-realistic artificial pictures, however they’ve additionally discovered success in different purposes resembling drug design and steady management. The important thing concept behind diffusion fashions is to iteratively rework random noise right into a pattern, resembling a picture or protein construction. That is usually motivated as a most chance estimation drawback, the place the mannequin is educated to generate samples that match the coaching information as intently as attainable.
Nonetheless, most use circumstances of diffusion fashions are usually not instantly involved with matching the coaching information, however as an alternative with a downstream goal. We don’t simply need a picture that appears like present pictures, however one which has a selected sort of look; we don’t simply desire a drug molecule that’s bodily believable, however one that’s as efficient as attainable. On this submit, we present how diffusion fashions will be educated on these downstream targets instantly utilizing reinforcement studying (RL). To do that, we finetune Steady Diffusion on quite a lot of targets, together with picture compressibility, human-perceived aesthetic high quality, and prompt-image alignment. The final of those targets makes use of suggestions from a big vision-language mannequin to enhance the mannequin’s efficiency on uncommon prompts, demonstrating how highly effective AI fashions can be utilized to enhance one another with none people within the loop.
A diagram illustrating the prompt-image alignment goal. It makes use of LLaVA, a big vision-language mannequin, to judge generated pictures.
Denoising Diffusion Coverage Optimization
When turning diffusion into an RL drawback, we make solely probably the most fundamental assumption: given a pattern (e.g. a picture), we’ve got entry to a reward perform that we are able to consider to inform us how “good” that pattern is. Our aim is for the diffusion mannequin to generate samples that maximize this reward perform.
Diffusion fashions are usually educated utilizing a loss perform derived from most chance estimation (MLE), that means they’re inspired to generate samples that make the coaching information look extra possible. Within the RL setting, we not have coaching information, solely samples from the diffusion mannequin and their related rewards. A technique we are able to nonetheless use the identical MLE-motivated loss perform is by treating the samples as coaching information and incorporating the rewards by weighting the loss for every pattern by its reward. This provides us an algorithm that we name reward-weighted regression (RWR), after present algorithms from RL literature.
Nonetheless, there are just a few issues with this method. One is that RWR isn’t a very actual algorithm — it maximizes the reward solely roughly (see Nair et. al., Appendix A). The MLE-inspired loss for diffusion can also be not actual and is as an alternative derived utilizing a variational certain on the true chance of every pattern. Because of this RWR maximizes the reward by two ranges of approximation, which we discover considerably hurts its efficiency.
We consider two variants of DDPO and two variants of RWR on three reward features and discover that DDPO persistently achieves the most effective efficiency.
The important thing perception of our algorithm, which we name denoising diffusion coverage optimization (DDPO), is that we are able to higher maximize the reward of the ultimate pattern if we take note of all the sequence of denoising steps that obtained us there. To do that, we reframe the diffusion course of as a multi-step Markov determination course of (MDP). In MDP terminology: every denoising step is an motion, and the agent solely will get a reward on the ultimate step of every denoising trajectory when the ultimate pattern is produced. This framework permits us to use many highly effective algorithms from RL literature which can be designed particularly for multi-step MDPs. As an alternative of utilizing the approximate chance of the ultimate pattern, these algorithms use the precise chance of every denoising step, which is extraordinarily straightforward to compute.
We selected to use coverage gradient algorithms resulting from their ease of implementation and previous success in language mannequin finetuning. This led to 2 variants of DDPO: DDPOSF, which makes use of the easy rating perform estimator of the coverage gradient also referred to as REINFORCE; and DDPOIS, which makes use of a extra highly effective significance sampled estimator. DDPOIS is our best-performing algorithm and its implementation intently follows that of proximal coverage optimization (PPO).
Finetuning Steady Diffusion Utilizing DDPO
For our foremost outcomes, we finetune Steady Diffusion v1-4 utilizing DDPOIS. We now have 4 duties, every outlined by a distinct reward perform:
- Compressibility: How straightforward is the picture to compress utilizing the JPEG algorithm? The reward is the unfavorable file measurement of the picture (in kB) when saved as a JPEG.
- Incompressibility: How laborious is the picture to compress utilizing the JPEG algorithm? The reward is the constructive file measurement of the picture (in kB) when saved as a JPEG.
- Aesthetic High quality: How aesthetically interesting is the picture to the human eye? The reward is the output of the LAION aesthetic predictor, which is a neural community educated on human preferences.
- Immediate-Picture Alignment: How effectively does the picture characterize what was requested for within the immediate? This one is a little more difficult: we feed the picture into LLaVA, ask it to explain the picture, after which compute the similarity between that description and the unique immediate utilizing BERTScore.
Since Steady Diffusion is a text-to-image mannequin, we additionally want to select a set of prompts to offer it throughout finetuning. For the primary three duties, we use easy prompts of the shape “a(n) [animal]”. For prompt-image alignment, we use prompts of the shape “a(n) [animal] [activity]”, the place the actions are “washing dishes”, “enjoying chess”, and “driving a motorbike”. We discovered that Steady Diffusion typically struggled to provide pictures that matched the immediate for these uncommon situations, leaving loads of room for enchancment with RL finetuning.
First, we illustrate the efficiency of DDPO on the easy rewards (compressibility, incompressibility, and aesthetic high quality). The entire pictures are generated with the identical random seed. Within the high left quadrant, we illustrate what “vanilla” Steady Diffusion generates for 9 completely different animals; all the RL-finetuned fashions present a transparent qualitative distinction. Apparently, the aesthetic high quality mannequin (high proper) tends in the direction of minimalist black-and-white line drawings, revealing the sorts of pictures that the LAION aesthetic predictor considers “extra aesthetic”.
Subsequent, we display DDPO on the extra complicated prompt-image alignment job. Right here, we present a number of snapshots from the coaching course of: every sequence of three pictures reveals samples for a similar immediate and random seed over time, with the primary pattern coming from vanilla Steady Diffusion. Apparently, the mannequin shifts in the direction of a extra cartoon-like type, which was not intentional. We hypothesize that it is because animals doing human-like actions usually tend to seem in a cartoon-like type within the pretraining information, so the mannequin shifts in the direction of this type to extra simply align with the immediate by leveraging what it already is aware of.
Surprising Generalization
Stunning generalization has been discovered to come up when finetuning massive language fashions with RL: for instance, fashions finetuned on instruction-following solely in English typically enhance in different languages. We discover that the identical phenomenon happens with text-to-image diffusion fashions. For instance, our aesthetic high quality mannequin was finetuned utilizing prompts that had been chosen from an inventory of 45 widespread animals. We discover that it generalizes not solely to unseen animals but in addition to on a regular basis objects.
Our prompt-image alignment mannequin used the identical record of 45 widespread animals throughout coaching, and solely three actions. We discover that it generalizes not solely to unseen animals but in addition to unseen actions, and even novel mixtures of the 2.
Overoptimization
It’s well-known that finetuning on a reward perform, particularly a realized one, can result in reward overoptimization the place the mannequin exploits the reward perform to realize a excessive reward in a non-useful means. Our setting isn’t any exception: in all of the duties, the mannequin finally destroys any significant picture content material to maximise reward.
We additionally found that LLaVA is inclined to typographic assaults: when optimizing for alignment with respect to prompts of the shape “[n] animals”, DDPO was capable of efficiently idiot LLaVA by as an alternative producing textual content loosely resembling the proper quantity.
There may be presently no general-purpose technique for stopping overoptimization, and we spotlight this drawback as an vital space for future work.
Conclusion
Diffusion fashions are laborious to beat relating to producing complicated, high-dimensional outputs. Nonetheless, to date they’ve largely been profitable in purposes the place the aim is to study patterns from heaps and many information (for instance, image-caption pairs). What we’ve discovered is a solution to successfully prepare diffusion fashions in a means that goes past pattern-matching — and with out essentially requiring any coaching information. The probabilities are restricted solely by the standard and creativity of your reward perform.
The way in which we used DDPO on this work is impressed by the latest successes of language mannequin finetuning. OpenAI’s GPT fashions, like Steady Diffusion, are first educated on large quantities of Web information; they’re then finetuned with RL to provide helpful instruments like ChatGPT. Usually, their reward perform is realized from human preferences, however others have extra just lately found out produce highly effective chatbots utilizing reward features primarily based on AI suggestions as an alternative. In comparison with the chatbot regime, our experiments are small-scale and restricted in scope. However contemplating the large success of this “pretrain + finetune” paradigm in language modeling, it actually looks as if it’s value pursuing additional on the planet of diffusion fashions. We hope that others can construct on our work to enhance massive diffusion fashions, not only for text-to-image era, however for a lot of thrilling purposes resembling video era, music era, picture modifying, protein synthesis, robotics, and extra.
Moreover, the “pretrain + finetune” paradigm isn’t the one means to make use of DDPO. So long as you’ve a very good reward perform, there’s nothing stopping you from coaching with RL from the beginning. Whereas this setting is as-yet unexplored, it is a place the place the strengths of DDPO may actually shine. Pure RL has lengthy been utilized to all kinds of domains starting from enjoying video games to robotic manipulation to nuclear fusion to chip design. Including the highly effective expressivity of diffusion fashions to the combination has the potential to take present purposes of RL to the subsequent degree — and even to find new ones.
This submit is predicated on the next paper:
If you wish to study extra about DDPO, you possibly can take a look at the paper, web site, unique code, or get the mannequin weights on Hugging Face. If you wish to use DDPO in your personal venture, take a look at my PyTorch + LoRA implementation the place you possibly can finetune Steady Diffusion with lower than 10GB of GPU reminiscence!
If DDPO evokes your work, please cite it with:
@misc{black2023ddpo,
title={Coaching Diffusion Fashions with Reinforcement Studying},
writer={Kevin Black and Michael Janner and Yilun Du and Ilya Kostrikov and Sergey Levine},
12 months={2023},
eprint={2305.13301},
archivePrefix={arXiv},
primaryClass={cs.LG}
}
