Yann LeCun has been obsessed with integrating learning end to end so machines learn good representations for over 30 years.

Judged as of February 2015. 'Over 30 years' means the interest dates to about 1985 or earlier. TRUE: dated records (publications, thesis work, talks) show LeCun working on end-to-end trained learning systems that learn internal representations (e.g., multilayer networks trained by backpropagation-like methods) by about 1985 or earlier. MOSTLY TRUE: the earliest documented work is from about 1986 to 1988, so slightly under 30 years, with the substance otherwise correct. MOSTLY FALSE: the documented focus begins only in the 1990s, or his early work was not about end-to-end representation learning. FALSE: no such work before the 2000s, or the evidence shows his research did not concern this idea.

Set before the research began.

1. E1 Yann LeCun's Research and Contributions Yann LeCun (personal research page, mirrored copy) · 1 January 2004 Own record

   “During my PhD, in 1985, I proposed and published (in French), an early version of the learning algorithm known as error backpropagation [LeCun, 1985]”

2. E2 Deep Learning in Neural Networks: An Overview arXiv (Technical Report IDSIA-03-14, Jürgen Schmidhuber, The Swiss AI Lab IDSIA) · 8 October 2014

   “Related work was published several years later (Parker, 1985; LeCun, 1985, 1988).”

3. E3 Bell Labs and the ‘neural’ network, 1986–1996 Cambridge Core (BJHS Themes) · 4 January 2023

   “who, after joining the firm in 1988, applied a version of the backpropagation algorithm”

Only LeCun's own research page says his 1985 publication was an early version of backpropagation (E1). The independent survey lists that 1985 publication only as "related work" and does not say what it covered (E2). The independent Bell Labs history documents backpropagation work from 1988 onward (E3). The independent evidence therefore does not show work on end-to-end representation learning by about 1985.