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


Current browse context:


Change to browse by:

References & Citations

DBLP - CS Bibliography


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

Computer Science > Artificial Intelligence

Title: An Initial Look at Self-Reprogramming Artificial Intelligence

Authors: Alex Sheng
Abstract: Rapid progress in deep learning research has greatly extended the capabilities of artificial intelligence technology. Conventional AI models are constrained to explicit human-designed algorithms, although a growing body of work in meta-learning, neural architecture search, and related approaches have explored algorithms that self-modify to some extent. In this paper, we develop and experimentally validate the first fully self-reprogramming AI system. Applying AI-based computer code generation to AI itself, we implement an algorithm with the ability to continuously modify and rewrite its own neural network source code.
Subjects: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
MSC classes: 68T07
ACM classes: I.2.0
Cite as: arXiv:2205.00167 [cs.AI]
  (or arXiv:2205.00167v1 [cs.AI] for this version)

Submission history

From: Alex Sheng [view email]
[v1] Sat, 30 Apr 2022 05:44:34 GMT (76kb,D)

Link back to: arXiv, form interface, contact.