Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating an interactive image. One of the methods includes accessing an image; detecting a plurality of objects depicted in the image; generating, for at least some of the plurality of objects, one or more rules that define allowed interactions with the respective object; generating, using the image and the one or more rules for the at least some of the plurality of objects, an interactive image; and providing an instruction to cause the device to display the interactive image that depicts the plurality of objects.

Abstract: In one aspect, a generative text model can be used to generate coding editing instructions according to a natural language prompt. Specifically, unique ID line tags may be prepended to lines of the code that are to be edited. The code with unique ID line tags may then be given to a generative model along with a natural language prompt setting forth one or more parameters for the edits. The LLM may then be used to generate the customized editing instructions. The editing instructions may then be received from the LLM and used post-processing to edit the code. After that, the unique ID line tags may be removed and the edited code may be returned to the user.

Abstract: Methods and systems for training a neural network model using domain mixing and multi-teacher knowledge distillation are described. Tokens, including a unique token, are inputted to an encoder of the neural network model. A unique embedding vector encoded from the unique token is inputted to an adaptor network to generate domain probabilities. A domain mixing embedding vector, determined from the unique embedding vector, is inputted to a predictor of the neural network model, to generate a predicted output. A final loss is computed using a domain mixing loss computed from the domain probabilities and a ground-truth domain of the data sample, and using an output prediction loss computed from the predicted output and a ground-truth label of the data sample. Parameters of the neural network model and adaptor network are updated using the final loss.