At Scale we built a tool for model debugging in computer vision called Nucleus (scale.com/nucleus) designed exactly for this, which is free try out if you're curious to see where your model predictions are most at odds with your ground truth. Source: over 2 years ago

To address your point about gathering edge cases, which can also be defined as cases of low model fidelity for our use cases, there is active learning and tools such as Aquarium Learning and Scale Nucleus which make it easy to implement into workflows. Source: almost 3 years ago

Try this: https://experiments.withgoogle.com/collection/ai. Source: over 1 year ago

But Google has a whole set of AI writing tools - https://experiments.withgoogle.com/collection/ai So by their own definition they are producing spam? - Source: Hacker News / about 2 years ago

Https://experiments.withgoogle.com/collection/ai might also help (I haven't used this IRL). Source: over 2 years ago

It's hard to imagine you've not seen Google's doodle guessing training (or their other experiments) but it's just another example of how little information you actually need to create a recognizable image, though Canvas also shows this off, but it has the benefit of material information. Source: over 2 years ago

To come back to your original question, as far as I'm aware anyone can publish on arxiv or researchgate. People will just tend to take you less serious. Maybe a better solution for you is something like this https://experiments.withgoogle.com/collection/ai . You already said you think your idea might be industry changing so if it truly is, I'm sure people will start noticing you. Source: about 3 years ago