DALL·E 2 pre-training mitigations

We noticed that our inside predecessors to DALL·E 2 would typically reproduce coaching photos verbatim. This conduct was undesirable, since we wish DALL·E 2 to create authentic, distinctive photos by default and never simply “sew collectively” items of current photos. Moreover, reproducing coaching photos verbatim can elevate authorized questions round copyright infringement, possession, and privateness (if individuals’s photographs had been current in coaching information).

To higher perceive the problem of picture regurgitation, we collected a dataset of prompts that often resulted in duplicated photos. To do that, we used a educated mannequin to pattern photos for 50,000 prompts from our coaching dataset, and sorted the samples by perceptual similarity to the corresponding coaching picture. Lastly, we inspected the highest matches by hand, discovering just a few hundred true duplicate pairs out of the 50k complete prompts. Although the regurgitation fee gave the impression to be lower than 1%, we felt it was essential to push the speed all the way down to 0 for the explanations acknowledged above.

Once we studied our dataset of regurgitated photos, we observed two patterns. First, the pictures had been nearly all easy vector graphics, which had been possible straightforward to memorize because of their low info content material. Second, and extra importantly, the pictures all had many near-duplicates within the coaching dataset. For instance, there is perhaps a vector graphic which seems like a clock displaying the time 1 o’clock—however then we might uncover a coaching pattern containing the identical clock displaying 2 o’clock, after which 3 o’clock, and many others. As soon as we realized this, we used a distributed nearest neighbor search to confirm that, certainly, all the regurgitated photos had perceptually comparable duplicates within the dataset. Other works have noticed the same phenomenon in giant language fashions, discovering that information duplication is strongly linked to memorization.

The above discovering advised that, if we deduplicated our dataset, we’d resolve the regurgitation downside. To attain this, we deliberate to make use of a neural community to establish teams of photos that regarded comparable, after which take away all however one picture from every group.^{[^footnote-2]}

Nonetheless, this may require checking, for every picture, whether or not it’s a duplicate of each different picture within the dataset. Since our complete dataset comprises a whole lot of thousands and thousands of photos, we might naively must examine a whole lot of quadrillions of picture pairs to seek out all of the duplicates. Whereas that is technically inside attain, particularly on a big compute cluster, we discovered a way more environment friendly various that works nearly as nicely at a small fraction of the price.Take into account what occurs if we cluster our dataset earlier than performing deduplication. Since close by samples typically fall into the identical cluster, a lot of the duplicate pairs wouldn’t cross cluster resolution boundaries. We may then deduplicate samples inside every cluster with out checking for duplicates outdoors of the cluster, whereas solely lacking a small fraction of all duplicate pairs. That is a lot sooner than the naive method, since we not need to examine each single pair of photos.^{[^footnote-3]}

Once we examined this method empirically on a small subset of our information, it discovered 85% of all duplicate pairs when utilizingOkay=1024 clusters.To enhance the success fee of the above algorithm, we leveraged one key remark: whenever you cluster completely different random subsets of a dataset, the ensuing cluster resolution boundaries are sometimes fairly completely different. Due to this fact, if a reproduction pair crosses a cluster boundary for one clustering of the information, the identical pair would possibly fall inside a single cluster in a special clustering. The extra clusterings you attempt, the extra possible you might be to find a given duplicate pair. In apply, we settled on utilizing 5 clusterings, which implies that we seek for duplicates of every picture within the union of 5 completely different clusters. In apply, this discovered 97% of all duplicate pairs on a subset of our information.

Surprisingly, nearly 1 / 4 of our dataset was eliminated by deduplication. Once we regarded on the near-duplicate pairs that had been discovered, lots of them included significant adjustments. Recall the clock instance from above: the dataset would possibly embrace many photos of the identical clock at completely different instances of day. Whereas these photos are more likely to make the mannequin memorize this specific clock’s look, they could additionally assist the mannequin study to tell apart between instances of day on a clock. Given how a lot information was eliminated, we had been nervous that eradicating photos like this may need harm the mannequin’s efficiency.

To check the impact of deduplication on our fashions, we educated two fashions with equivalent hyperparameters: one on the complete dataset, and one on the deduplicated model of the dataset. To match the fashions, we used the identical human evaluations we used to judge our authentic GLIDE mannequin. Surprisingly, we discovered that human evaluators barely most popular the mannequin educated on deduplicated information, suggesting that the big quantity of redundant photos within the dataset was really hurting efficiency.

As soon as we had a mannequin educated on deduplicated information, we reran the regurgitation search we had beforehand executed over 50k prompts from the coaching dataset. We discovered that the brand new mannequin by no means regurgitated a coaching picture when given the precise immediate for the picture from the coaching dataset. To take this take a look at one other step additional, we additionally carried out a nearest neighbor search over the complete coaching dataset for every of the 50k generated photos. This manner, we thought we’d catch the mannequin regurgitating a special picture than the one related to a given immediate. Even with this extra thorough examine, we by no means discovered a case of picture regurgitation.