Diffusion fashions have just lately emerged because the de facto customary for producing advanced, high-dimensional outputs. You could know them for his or her potential to supply beautiful AI artwork and hyper-realistic artificial photographs, however they’ve additionally discovered success in different functions comparable to drug design and steady management. The important thing thought behind diffusion fashions is to iteratively remodel random noise right into a pattern, comparable to a picture or protein construction. That is usually motivated as a most chance estimation downside, the place the mannequin is skilled to generate samples that match the coaching information as carefully as attainable.
Nevertheless, most use instances of diffusion fashions will not be straight 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 photographs, however one which has a selected kind of look; we don’t simply desire a drug molecule that’s bodily believable, however one that’s as efficient as attainable. On this put up, we present how diffusion fashions will be skilled on these downstream targets straight utilizing reinforcement studying (RL). To do that, we finetune Steady Diffusion on a wide range 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 photographs.
Denoising Diffusion Coverage Optimization
When turning diffusion into an RL downside, we make solely essentially the most fundamental assumption: given a pattern (e.g. a picture), we have now entry to a reward operate 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 operate.
Diffusion fashions are usually skilled utilizing a loss operate derived from most chance estimation (MLE), which means they’re inspired to generate samples that make the coaching information look extra seemingly. 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 operate is by treating the samples as coaching information and incorporating the rewards by weighting the loss for every pattern by its reward. This offers us an algorithm that we name reward-weighted regression (RWR), after present algorithms from RL literature.
Nevertheless, there are just a few issues with this method. One is that RWR shouldn’t be a very precise algorithm — it maximizes the reward solely roughly (see Nair et. al., Appendix A). The MLE-inspired loss for diffusion can be not precise and is as an alternative derived utilizing a variational certain on the true chance of every pattern. Which means RWR maximizes the reward by way of 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 perfect 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 the complete sequence of denoising steps that received 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 might 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 simple to compute.
We selected to use coverage gradient algorithms as a consequence of 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 operate 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 carefully follows that of proximal coverage optimization (PPO).
Finetuning Steady Diffusion Utilizing DDPO
For our primary outcomes, we finetune Steady Diffusion v1-4 utilizing DDPOIS. We have now 4 duties, every outlined by a special reward operate:
- Compressibility: How simple is the picture to compress utilizing the JPEG algorithm? The reward is the adverse file measurement of the picture (in kB) when saved as a JPEG.
- Incompressibility: How exhausting 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 skilled on human preferences.
- Immediate-Picture Alignment: How properly does the picture symbolize what was requested for within the immediate? This one is a little more sophisticated: 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 choose a set of prompts to present 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”, “taking part in chess”, and “using a motorcycle”. We discovered that Steady Diffusion usually struggled to supply photographs that matched the immediate for these uncommon eventualities, 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). All the photographs are generated with the identical random seed. Within the high left quadrant, we illustrate what “vanilla” Steady Diffusion generates for 9 completely different animals; the entire RL-finetuned fashions present a transparent qualitative distinction. Curiously, the aesthetic high quality mannequin (high proper) tends in direction of minimalist black-and-white line drawings, revealing the sorts of photographs that the LAION aesthetic predictor considers “extra aesthetic”.
Subsequent, we display DDPO on the extra advanced prompt-image alignment process. Right here, we present a number of snapshots from the coaching course of: every sequence of three photographs reveals samples for a similar immediate and random seed over time, with the primary pattern coming from vanilla Steady Diffusion. Curiously, the mannequin shifts in direction of a extra cartoon-like fashion, which was not intentional. We hypothesize that it is because animals doing human-like actions usually tend to seem in a cartoon-like fashion within the pretraining information, so the mannequin shifts in direction of this fashion 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 giant language fashions with RL: for instance, fashions finetuned on instruction-following solely in English usually 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 have been chosen from an inventory of 45 widespread animals. We discover that it generalizes not solely to unseen animals but additionally 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 additionally to unseen actions, and even novel combos of the 2.
Overoptimization
It’s well-known that finetuning on a reward operate, particularly a realized one, can result in reward overoptimization the place the mannequin exploits the reward operate to realize a excessive reward in a non-useful means. Our setting is not 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 prone 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 right quantity.
There’s presently no general-purpose methodology for stopping overoptimization, and we spotlight this downside as an essential space for future work.
Conclusion
Diffusion fashions are exhausting to beat in relation to producing advanced, high-dimensional outputs. Nevertheless, to date they’ve largely been profitable in functions the place the aim is to be taught patterns from heaps and many information (for instance, image-caption pairs). What we’ve discovered is a technique 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 operate.
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 skilled on large quantities of Web information; they’re then finetuned with RL to supply helpful instruments like ChatGPT. Sometimes, their reward operate is realized from human preferences, however others have extra just lately discovered methods to 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 definitely looks like it’s price pursuing additional on this planet of diffusion fashions. We hope that others can construct on our work to enhance giant diffusion fashions, not only for text-to-image technology, however for a lot of thrilling functions comparable to video technology, music technology, picture modifying, protein synthesis, robotics, and extra.
Moreover, the “pretrain + finetune” paradigm shouldn’t be the one means to make use of DDPO. So long as you could have reward operate, 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 might actually shine. Pure RL has lengthy been utilized to all kinds of domains starting from taking part in video games to robotic manipulation to nuclear fusion to chip design. Including the highly effective expressivity of diffusion fashions to the combo has the potential to take present functions of RL to the following stage — and even to find new ones.
This put up is predicated on the next paper:
If you wish to be taught extra about DDPO, you may try the paper, web site, unique code, or get the mannequin weights on Hugging Face. If you wish to use DDPO in your individual mission, try my PyTorch + LoRA implementation the place you may 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},
yr={2023},
eprint={2305.13301},
archivePrefix={arXiv},
primaryClass={cs.LG}
}