Amazon Technologies, Inc.

Ismail Baha Tutar

Tarik Arici

Deept Kumar

Luis Antonio Diniz Fernandes De Morais Sarmento

Sachin Birendra Singh

Jeremy Calvert

Roshan Ram Rammohan

Amirhossein Tavanaei

Karim Bouyarmane

Vito Nicola Mandorino

Tal Neiman

Mehmet Saygin Seyfioglu

The present disclosure presents a generative model configured to receive input regarding an item in two different modalities, such as text data and non-text data (including, for example, image or audio data), in order to generate output regarding the item that is determined based on a combination of both modalities' input. Specific relative positional and token type embeddings may be employed in an encoder portion of an encoder-decoder arrangement. An associated decoder may be trained to generate new text corresponding to diverse tasks based on the encoded representation of the two inputs as generated within the encoder. For example, the decoder may be utilized to generate attributes regarding the input item, auto-complete or auto-correct a title or description of the item, among other uses.

Multimodal input to text transformer

Systems and methods for improving training processes for image and text applications are described. A first set of embeddings may be generated based on a text input, and a second set of embeddings may be generated via a convolutional neural network (CNN), based on an input image. The first set of embeddings and the second set of embeddings may be utilized to generate a third set of embeddings including one or more placeholder values to be replaced. The placeholder values may be replaced based on predicted values, to reconstruct the input text and image.

Vision-and-language model training

Methods, systems, and computer-readable media for similarity detection based on token distinctiveness are disclosed. A similarity detection system determines candidate items for a seed item based on a comparison of tokens in textual descriptions of the candidate items to tokens in a textual description of the seed item. The system uses machine learning to determine importance scores for the tokens of the seed item. An importance score is determined based on the frequency of the individual token and the frequency of the most commonly occurring token in the candidate items. Importance scores for the same token differ from the seed item to another seed item. Based on the importance scores, the system determines similarity scores for the candidate items to the seed item. The system selects, from the candidate items, a set of similar items to the seed item based (at least in part) on the similarity scores.

Similarity detection based on token distinctiveness

A plurality of training iterations is conducted for a machine learning task. A given iteration includes generating a version of a stacking model using a portion of a labeled data set. Proposed labels are then obtained in the iteration using the generated version of the stacking model for a set of unlabeled records. The unlabeled records and their proposed labels are then used to generate versions of base models for the iteration. After the training iterations are completed, a trained ensemble of models including the stacking model and the base models is stored.

Artificial intelligence system supporting semi-supervised learning with iterative stacking

Query types for which responses are to be generated with respect to records comprising text attributes are identified, including a text interpretation query type for which records may comprise one or more response-contributor strings. Results of the text interpretation query for a record are based at least partly on an extracted-property class of the records. A machine learning model comprising a first sub-model and a second sub-model is trained to extract results of the text interpretation query. The first sub-model generates an extracted-property class for a record, and the second sub-model predicts positions of response-contributor strings within the record based at least in part on the extracted-property class. A trained version of the model is stored.

Artificial intelligence system with composite models for multiple response-string queries

Similarity detection methods and systems are provided that utilize a convolutional neural network model to jointly learn string matching and semantic textual similarity as an image recognition solution. For example, in some embodiments described herein, the similarity detection system may receive two strings as input, transform the two strings into two separate vectors, generate a high-resolution image and a low-resolution image, apply one or more convolutional operations to each image, and determine string matching and semantic textual similarity based at least partly on the high-resolution image and the low-resolution image.

Similarity detection system

Techniques are provided herein for utilizing a query variations engine. An attribute category for a search query may be identified from a search query history. A search query may be received from a user and a set of search results may be determined. A reduced set of search results may be generated from the set of search results based at least in part on the attribute category identified from the search query history. The user may be provided with the reduced set of search results.

Techniques for dynamic variations of a search query

Technologies are disclosed herein for generating and utilizing machine-learning generated classifiers configured to identify document relationships. Manually-generated data is captured that indicates if documents in a document corpus have a relationship with one another, such as duplicates or variations. A determination may then be made as to whether a classifier is to be generated based on the duplicate decision data. If a classifier is to be generated, machine learning may be performed using training documents from the document corpus and the duplicate decision data to generate a classifier. The machine-learning generated classifier may then be utilized in a production environment to determine whether a new document is a duplicate of documents in the document corpus and/or to identify other relationships between documents in the document corpus, such as documents that are similar or are variations of one another.

Growing community of inventors

Seattle, WA, United States of America

Ismail Baha Tutar