Qualcomm Incorporated

Roland Memisevic

Mingu Lee

Christopher G Lott

Wonseok Jeon

Reza Pourreza

Sunny Praful Kumar Panchal

Apratim Bhattacharyya

Pulkit Madan

A processor-implemented method for generating grounded rationales for visual reasoning tasks includes receiving, by a first artificial neural network (ANN), an interleaved sequence of images and textual information. The first ANN extracts grid features of the images of the interleaved sequence of the images and the textual information to generate a representation of the interleaved sequence of the images and the textual information based on the grid features. A second ANN maps the grid features to a textual domain. The second ANN extracts visual information of the interleaved sequence of the images and the textual information based on the grid features in the textual domain. The second ANN determines a rationale based on the visual information. The visual information comprises one or more lower-level surrogate tasks.

Using grounded rationales to improve visual reasoning

Certain aspects of the present disclosure provide techniques and apparatus for generating a response to a query input in a generative artificial intelligence model. An example method generally includes receiving a plurality of sets of tokens generated based on an input prompt and a first generative artificial intelligence model, each set of tokens in the plurality of sets of tokens corresponding to a candidate response to the input prompt; selecting, using a second generative artificial intelligence model and recursive adjustment of a target distribution associated with the received plurality of sets of tokens, a set of tokens from the plurality of sets of tokens; and outputting the selected set of tokens as a response to the input prompt.

Speculative decoding in autoregressive generative artificial intelligence models

Growing community of inventors

Toronto, Canada

Roland Memisevic