We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.LG

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Machine Learning

Title: Deep Extrapolation for Attribute-Enhanced Generation

Abstract: Attribute extrapolation in sample generation is challenging for deep neural networks operating beyond the training distribution. We formulate a new task for extrapolation in sequence generation, focusing on natural language and proteins, and propose GENhance, a generative framework that enhances attributes through a learned latent space. Trained on movie reviews and a computed protein stability dataset, GENhance can generate strongly-positive text reviews and highly stable protein sequences without being exposed to similar data during training. We release our benchmark tasks and models to contribute to the study of generative modeling extrapolation and data-driven design in biology and chemistry.
Comments: NeurIPS 2021
Subjects: Machine Learning (cs.LG); Computation and Language (cs.CL); Quantitative Methods (q-bio.QM)
Cite as: arXiv:2107.02968 [cs.LG]
  (or arXiv:2107.02968v2 [cs.LG] for this version)

Submission history

From: Ali Madani [view email]
[v1] Wed, 7 Jul 2021 01:30:36 GMT (390kb,D)
[v2] Mon, 25 Oct 2021 23:03:20 GMT (422kb,D)

Link back to: arXiv, form interface, contact.