View The Document

Accession Number:

AD1151555

Title:

The Human Data Interaction Project

Author Organization(s):

• MASSACHUSETTS INSTITUTE OF TECHNOLOGY

Report Date:

2021-10-27

Abstract:

This project removes costly bottlenecks from an end-to-end data science endeavor, while efficiently redistributing the endeavors most vital resource human effort. To accomplish this, we first apply structure and systematically to the otherwise ad hoc processes of transforming (1) temporal, relational, and transactional data into features and labels, (2) features and labels into trained models, and (3) models into insights. We establish that structure in a generalizable, intuitive way so as to encourage participation from a broad user base. And, in so doing, our system creates opportunities to automate, optimize, and learn across disparate data science endeavors.

Document Type:

Conference:

Journal:

Pages:

18

File Size:

0.60MB

Contracts:

Grants:

Descriptors:

Identifiers:

• DFS(DEEP FEATURE SYNTHESIS)
• ATM SYSTEMS
• BTB SYSTEMS
• HYPERPARAMETERS
• BTB(BAYESIAN TUNING AND BANDITS)
• MLBLOCKS(MACHINE LEARNING BLOCKS)
• D3M(DATA-DRIVEN DISCOVERY OF MODELS)

SubjectCategory:

• Behavioral and Social Sciences

Communities of Interest:

• Autonomy

Distribution Statement:

Approved For Public Release

View The Document