ClaimsYann LeCun

“We expect to see many new embedded vision systems based on ConvNets in the next few years.”

Yann LeCun, co-author of paper 2010 Convolutional Networks and Applications in Vision (yann.lecun.com) archived copy

Prediction AI compute, hardware and data

The authors expected to see many new embedded vision systems based on ConvNets in the next few years.

Unresolvable The public record could not settle it.

Prediction: Statements about events outside the person's control, checked once they fell due: did they come true?

The second judge would not resolve this claim; the verdict shown is the first judge's.

The evidence is the person's or their organisation's own records.

What would count as true or false

Judge at the end of 2013. This is a 'many' claim, so it is about degree. Count distinct embedded vision systems based on convolutional networks that appeared (were published, demonstrated or released) between 2010 and the end of 2013. Embedded means running on dedicated, mobile, FPGA, ASIC or on-device hardware rather than on server or desktop GPUs. Both products and published hardware or research systems count (for example ConvNet processors on FPGA or ASIC, or on-camera, robot or phone implementations). True: five or more distinct such systems. Mostly true: three or four. Mostly false: one or two. False: dated surveys or literature from 2013–2014 show no new ConvNet-based embedded vision systems appearing in 2010–2013. Systems that first appeared after 2013 do not count.

Set before the research began.

Evidence (3)

  1. E1 A Runtime Reconfigurable Dataflow Processor for Vision Not stated on page (NeuFlow paper by Farabet, Martini, Culurciello, Akselrod, Corda, LeCun; page text is the ECVW-11 invited paper) Own record

    “while consuming 10 Watts when implemented on a Xilinx Virtex 6 FPGA platform, or about ten times less than a laptop computer”

  2. E2 Deep Learning Hardware: Past, Present, & Future ISSCC (talk slides by Yann LeCun, San Francisco, 2019-02-18) · 18 February 2019 Own record

    “NeuFlow ASIC [Pham 2012] 150GOPS, 0.5 Watts (simulated)”

  3. E3 A Runtime Reconfigurable Dataflow Processor for Vision Not stated on page; the cited work is in the Proceedings of the 37th annual international symposium on Computer architecture (ACM, 2010) Own record

    “A dynamically configurable coprocessor for convolutional neural networks. In Proceedings of the 37th annual international symposium on Computer architecture, pages 247–257. ACM, 2010.”

Reasoning

The evidence shows a NeuFlow vision processor on an FPGA (undated, E1), a simulated NeuFlow ASIC from 2012 (E3), and a reference to a 2010 coprocessor for convolutional neural networks (E4). It does not establish three or more distinct ConvNet-based embedded vision systems from 2010–2013, because E3 is a variant of the same NeuFlow processor and E4 is not shown to be an embedded vision system.

Cite this claim

Yann LeCun, “The authors expected to see many new embedded vision systems based on ConvNets in the next few years.” (said 2010). Verdict: unresolvable. Reliability API, claim c_30d6c0291ef8, method 0.1.0+c876030c, dataset release v1.1.2. https://epistacklabs.com/claims/c_30d6c0291ef8

This claim is in dataset release v1.1.2, with the same verdict.

Reply to this claim

For Yann LeCun, or someone speaking for them: a correction or counter-evidence. Replies are read before anything is shown; an approved reply shows your name, role, date, statement and link. Your email address is never shown and is deleted 30 days after the decision (privacy).

More prediction claims by Yann LeCun

  1. Mostly true Prediction 2019
    “We foresee an increase in the usage of such networks for a wide variety of applications, which are likely to violate the assumptions of current DL hardware.”

    Usage of graph neural networks will increase for a wide variety of applications likely to violate the assumptions of current deep learning hardware.

  2. Mostly true Prediction 2019
    “As DL systems are increasingly used for high-level cognitive tasks, such memory modules will become commonplace and very large, requiring hardware support.”

    As DL is used for high-level cognitive tasks, memory modules will become commonplace and very large, requiring hardware support.

  3. Mostly true Prediction 8 March 2024
    “Increasingly, the search engines that we’re going to use are not going to be search engines, they’re going to be dialogue systems”

    Search engines will increasingly be replaced by dialogue systems that answer questions and perhaps point to appropriate references.

  4. Mostly true Prediction 18 February 2015
    “The next frontier for Deep Learning is natural language understanding.”

    Natural language understanding will be the next frontier for Deep Learning.

  5. Mostly true Prediction 2010
    “one of the major challenges of computer vision, machine learning, and AI in general in the next decade will be to devise methods that can automatically learn good features hierarchies from unlabeled…”

    In the next decade, a major challenge of computer vision, machine learning and AI will be devising methods that learn good feature hierarchies from unlabeled and labeled data in an integrated fashion.

All of them

Verdicts are made automatically; how verdicts are made.