Since the 1990s, algorithms exist that learn oriented edge detectors fully unsupervised with the equivalent of a few minutes of real time.
Mostly true True in substance, wrong in a detail.
Factual: Statements about the present or the past, checked against the record: did they hold up?
What would count as true or false
Judge as of 2021. True: unsupervised algorithms from the 1990s (for example, Olshausen & Field 1996 sparse coding, or ICA by Bell & Sejnowski 1997) learned oriented, edge-detector-like (Gabor-like) filters from natural images, using an amount of image data comparable to a few minutes of visual input. Mostly true: such 1990s algorithms existed, but the 'few minutes' data equivalence is roughly right rather than exact, or is not clearly documented. Mixed: the algorithms existed, but they needed far more data than a few minutes' equivalent. Mostly false: such results came only after the 1990s, or needed supervision. False: no unsupervised algorithm learned oriented edge detectors in that era.
Set before the research began.
Evidence (1)
-
E1 The "independent components" of natural scenes are edge filters. | tlooto tlooto (academic paper aggregator) · 1 December 1997
“It is shown that a new unsupervised learning algorithm based on information maximization, a nonlinear "infomax" network, when applied to an ensemble of natural scenes produces sets of visual filters that are localized and oriented.”
Reasoning
E1 is the 1997 paper 'The "independent components" of natural scenes are edge filters'. It describes an unsupervised infomax algorithm that, trained on natural scenes, produces visual filters that are localized and oriented, so an unsupervised method from the 1990s did learn oriented edge detectors. The evidence does not say how much image data was used, so the claim that this took the equivalent of a few minutes of real time is not documented. Under the criteria, that makes the claim mostly true.
Cite this claim
Yann LeCun, “Since the 1990s, algorithms exist that learn oriented edge detectors fully unsupervised with the equivalent of a few minutes of real time.” (said 2021). Verdict: mostly true. Reliability API, claim c_5736a3988234, method 0.1.0+c876030c, dataset release v1.1.2. https://epistacklabs.com/claims/c_5736a3988234
This claim is in dataset release v1.1.2, with the same verdict.
Reply to this claim
More factual claims by Yann LeCun
-
Mostly true
“For starters, transformers are equivariant to permutations.”
Transformers are equivariant to permutations, as ConvNets are equivariant to translations.
-
Mostly true
“If it were the case, we would have AI systems that could teach themselves to drive a car in 20 hours of practice, like any 17 year-old.”
-
Mostly true
“The architecture is good, the architecture of the encoder is good, but the fact that you train the system to reconstruct images does not lead it to produce to long, good generic features of images.”
-
Mostly true
“LLMs can do none of those or they can only do them in a very primitive way and they don’t really understand the physical world.”
-
Mostly true
“No, there’s one thing that autoregressive LLMs or that LLMs in general, not just the autoregressive one, but including the bird style bidirectional ones, are exploiting and its self supervised…”
Verdicts are made automatically; how verdicts are made.