Scores by kind of claim
Each kind of claim is scored on its own, out of 100: the share that held up, with the range it likely falls in. Fewer resolved claims give a wider range.
-
Factual claims
Statements about the present or the past, checked against the record: did they hold up?
78 72 83 -
Predictions
Statements about events outside the person's control, checked once they fell due: did they come true?
64 43 83 -
Promises
Commitments to the person's or organisation's own future action, with something to deliver: was it delivered?
Insufficient record: 0 promises resolved, 5 needed for a score.
All types combined: 77 (likely 71 to 82, high confidence), from 84 claims: 79 factual claims, 5 predictions. It mixes kinds of claim in different proportions for different people, so compare people kind by kind.
Yann LeCun's claims
55 claims
“Convolutional Networks (ConvNets) are a biologicallyinspired trainable architecture that can learn invariant features.”
Yann LeCun, co-author of paper 2010 Convolutional Networks and Applications in Vision (yann.lecun.com)
Convolutional Networks are a biologically inspired trainable architecture that can learn invariant features.
Mostly true True in substance, wrong in a detail.
What would count as true or false
Judged against the literature as of 2010. True if ConvNets are widely described as biologically inspired (for example, by Hubel and Wiesel's visual cortex work or Fukushima's Neocognitron), are trained end to end, and learn features that are invariant, at least partially, to transformations such as small shifts and distortions. Mostly true if all of this holds but the invariance is only approximate or local (for example, to small translations only), as long as the claim is not read as perfect invariance. Mixed if one of the two main parts (biological inspiration, or learnable invariant features) lacks support. Mostly false if ConvNets' invariance is mainly built in by design rather than learned and the biological link is only loose. False if both parts are contradicted.
Evidence
- Convolutional neural network - Wikipedia Wikipedia · 31 August 2013
“Convolutional networks were inspired by biological processes[18][19][20][21] in that the connectivity pattern between neurons resembles the organization of the animal visual cortex.”
- Convolutional neural network - Wikipedia Wikipedia · 31 August 2013
“The convolutional part contained 18 hand-designed kernels and was not trained. Only the fully connected part was trained by backpropagation. By 1989 the group had reported training several CNNs end-to-end by backpropagation.”
- Convolutional neural network - Wikipedia Wikipedia · 31 August 2013
“In addition to reducing the sizes of feature maps, the pooling operation grants a degree of local translational invariance to the features contained therein, allowing the CNN to be more robust to variations in their positions.”
- Convolutional neural network - Wikipedia Wikipedia · 31 August 2013
“Counter-intuitively, most convolutional neural networks are not invariant to translation, due to the downsampling operation they apply to the input.[16]”
Reasoning
E1, an independent encyclopedia, says convolutional networks were inspired by biological processes, namely how the animal visual cortex is organised. E2 reports that by 1989 a research group had trained several CNNs end to end by backpropagation, so the architecture is trainable. E3 says pooling gives the features only a degree of local translational invariance, and E4 says most ConvNets are not truly invariant to translation. The evidence supports biological inspiration [E1] and end-to-end training [E2], and shows that pooling gives the features a degree of local translational invariance [E3], though not full invariance [E4]; the invariance is approximate and local, which the criteria rate as mostly true.
Cite this claim
Yann LeCun, “Convolutional Networks are a biologically inspired trainable architecture that can learn invariant features.” (said 2010). Verdict: mostly true. Reliability API, claim c_b987a16259dc, method 0.1.0+c876030c. https://epistacklabs.com/claims/c_b987a16259dc
Claims without a verdict
- Unresolvable 110
- Not yet due 3
- Not researched 309
- Not checkable 669
Is a verdict here wrong, or do you speak for Yann LeCun? Reply on the claim's own page, with the form at its foot (how replies work). Verdicts are made automatically; how verdicts and scores are made.