Abstract: Embodiments described herein provide a mechanism that ensembles trainable soft prompts to transfer knowledge from source tasks under few-shot learning settings. Specifically, given a source task input from a source task training dataset, a set of soft prompts may be trained using a frozen PLM on the large-scale source task training dataset. The set of soft prompts are then prepended to a target task input, based on which the frozen pre-trained language model generates a set of logits for predicting classification of the target task input, respectively. An attention module is used to generate input-logit attention scores, which are used to compute a weighted linear combination of the logits given the attention scores. The weighted linear combination are the final logits to predict the final classification of the target task input.

Abstract: Systems, devices, and techniques are disclosed for automatic product description generation. A first set of features including labels including words may be generated from an image using a first feature extraction model. A second set of features including labels including words may be generated from the image using a second feature extraction model. A text description of a product depicted in the image may be generated by inputting the image and metadata for the image to a description generating model. The text description may include words. Each of the words may be generated by assigning probabilities to candidate words, boosting the assigned probabilities of candidate words that are similar to words of labels of the first set of features or words of labels of the second set of features, and selecting one of the candidate words based on the assigned probabilities after the boosting as a word of the text description.