Carnegie Mellon University: AI, Backpropagation, Robotics

Carnegie Mellon's AI research spans symbolic problem-solving, Hinton's backpropagation work, the Robotics Institute and DARPA autonomous-vehicle competitions.

kindUniversity
founded1900

Updated

From Carnegie Tech to AI research

Carnegie Mellon traces its origins to Andrew Carnegie's 1900 funding of technical schools in Pittsburgh. Carnegie Tech merged with the Mellon Institute in 1967 to form Carnegie Mellon University. Its computing research predates that merger: Herbert Simon, Allen Newell and RAND's Cliff Shaw developed the Logic Theorist, and the General Problem Solver followed in 1957. Newell and Simon shared the 1975 Turing Award for contributions to AI and human cognition.

The Computer Science Department was established in 1965. The university approved its elevation to a school in 1988, with operations beginning in 1989. Geoffrey Hinton served on the department's faculty from 1982 to 1987, the period in which he co-authored the 1986 Nature backpropagation paper with David Rumelhart and Ronald Williams. The work described training neural networks to learn internal representations, a different research approach from the earlier symbolic programs.

Robotics and machine learning as research institutions

The Robotics Institute opened in October 1979 with support from Westinghouse and the Office of Naval Research. Raj Reddy was its founding director from 1979 to 1991 and subsequently dean of the School of Computer Science from 1991 to 1999. The Machine Learning Department came later: the Center for Automated Learning and Discovery became a department in 2006. These dates distinguish the growth of separate research units from any one administrator's tenure.

CMU vehicles finished second and third in the 2005 DARPA Grand Challenge. In the 2007 Urban Challenge, the university's Tartan Racing team included faculty, students, staff and embedded industry engineers. Its modified Chevrolet Tahoe, Boss, won the judged competition and a $2 million prize after navigating roads shared with other vehicles. The result demonstrated autonomous driving in the competition setting, rather than unrestricted public-road operation.

Sources

  1. [1]
    History

    Carnegie Mellon University · May 30, 2025

  2. [2]
    Hinton Shares 2019 Turing Award for Deep Learning

    Carnegie Mellon University · Mar 28, 2019

  3. [3]
    History of the Robotics Institute

    Carnegie Mellon University Robotics Institute · Aug 8, 2016

  4. [4]
  5. [5]
    A history of SCS

    Carnegie Mellon University · retrieved Oct 24, 2026

  6. [6]
    AI: Games & Search Strategies

    Carnegie Mellon University · Nov 16, 2012

  7. [7]
    Learning representations by back-propagating errors

    Geoffrey Hinton, University of Toronto · retrieved Oct 24, 2026

  8. [8]
    Raj Reddy

    Carnegie Mellon University · retrieved Oct 24, 2026

  9. [9]
    Carnegie Mellon Tartan Racing Wins

    Carnegie Mellon University · Nov 4, 2007

  10. [10]

Carnegie Mellon University history

2023

■
Mamba offers a Transformer alternative that carries fixed-size state

With Princeton University

A selective state space model, linear in sequence length, with no attention and no cache that grows as the context does, reported at four to five times a comparable Transformer's inference throughput. What actually shipped afterwards were hybrids that keep a few attention layers for recall.

arXiv
Research

1986

●
Backpropagation reaches Nature

With University of California San Diego

Rumelhart, Hinton and Williams showed that a network could learn useful internal representations by propagating error backwards through its layers, and this is the paper that made the technique widely known. Earlier versions existed, credited in the paper itself, and it took another two decades before hardware caught up to train networks of any real depth.

Nature
Research