公开(公告)号：US20240386002A1

公开(公告)日：2024-11-21

申请号：US18319748

申请日：2023-05-18

Applicant: Adobe Inc.

Inventor： Raunak Shah ,  Koyel MUKHERJEE ,  Subrata MITRA ,  Dhruv JOSHI ,  Sai KARNAM ,  Shivam Pravin BHOSALE

IPC: G06F16/215 ,  G06F16/28 ,  G06F40/284

Abstract: A dataset comprising tables is received. Embeddings are generated for column titles of a table. Based on the embeddings, similar tables are clustered. The tables are organized into smaller clusters based on statistical similarities. Similarity scores are calculated for tables within the same cluster. A relatedness graph is created based on the similarity scores; similar tables are represented by nodes connected by edges. If the similarity score for a pair of tables exceeds a threshold, a table is deleted.

公开(公告)号：US20240220502A1

公开(公告)日：2024-07-04

申请号：US18092779

申请日：2023-01-03

Applicant: Adobe Inc.

Inventor： Vibhor PORWAL ,  Yeuk-Yin CHAN ,  Vidit BHATIA ,  Subrata MITRA ,  Shaddy GARG ,  Sergey N. KAZARIN ,  Sameeksha ARORA ,  Himanshu PANDAY ,  Gautam Pratap KOWSHIK ,  Fan DU ,  Anup Bandigadi RAO ,  Anil MALKANI

IPC: G06F16/2453 ,  G06F16/2458

CPC classification number: G06F16/24544 ,  G06F16/2462

Abstract: To retrieve information derived from a plurality of separately stored datasets, join structures are identified within the plurality of separately stored datasets. Join structures can include datasets joined by a central dataset, datasets joined by a single key, and datasets joined across a plurality of keys. Each of the join structures corresponds to a query processing schema that defines a sampling technique. When a join query is received as a SQL query, the join query identifies a portion of the plurality of separately stored datasets, from which a join structure is selected and a corresponding query processing schema is identified. The join query is reconstructed to form a reconstructed join query that comprises query processing schema instructions to derive the requested information using the sampling technique defined by the identified query processing schema.

公开(公告)号：US20250005075A1

公开(公告)日：2025-01-02

申请号：US18342474

申请日：2023-06-27

Applicant: Adobe Inc.

Inventor： Sachin Kumar Chauhan ,  Subrata MITRA ,  Sunav CHOUDHARY ,  Ramasuri NARAYANAM ,  Koyel MUKHERJEE ,  Gautam Pratap KOWSHIK

IPC: G06F16/901

Abstract: Tabular data is received. A graph is created based on the tabular data. The graph comprises nodes corresponding to key-value pairs of the tabular data. Weights are assigned to the nodes and to edges that connect the nodes. The node and edge weights are updated using a message-passing neural network (MPNN) framework. The resulting graph is sampled based on the updated weights.

公开(公告)号：US20240273296A1

公开(公告)日：2024-08-15

申请号：US18625884

申请日：2024-04-03

Applicant: Adobe Inc.

Inventor： Sungchul KIM ,  Subrata MITRA ,  Ruiyi Zhang ,  Rui Wang ,  Handong ZHAO ,  Tong YU

IPC: G06F40/295 ,  G06N20/00

CPC classification number: G06F40/295 ,  G06N20/00

Abstract: Embodiments of the technology described herein describe a machine classifier capable of continually learning new classes through a continual few-shot learning approach. A natural language processing (NLP) machine classifier may initially be trained to identify a plurality of other classes through a conventional training process. In order to learn a new class, natural-language training data for a new class is generated. The training data for the new class may be few-shot training data. The training also uses synthetic training data that represents each of the plurality of other classes. The synthetic training data may be generated through a model inversion of the original classifier. The synthetic training data and the natural-language training data are used to retrain the NLP classifier to identify text in the plurality of other classes and the new class using.