| Unresolvable |
Yann LeCun | Prediction |
2019 |
The demands on DL-specific hardware will undoubtedly increase.
Yann LeCun · Prediction · 2019 |
AI compute, hardware and data |
| Not checkable |
Yann LeCun | Not checkable |
2019 |
There are five use cases with different hardware requirements
Yann LeCun · Not checkable · 2019 |
|
| Not checkable |
Yann LeCun | Not checkable |
2019 |
The best architectures are those that can be saturated with the smallest batch of samples.
Yann LeCun · Not checkable · 2019 |
|
| Not checkable |
Yann LeCun | Not checkable |
2019 |
The answer is a resounding yes.
Yann LeCun · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
it is a central property of fat tails that higher moments are explosive (even "infinite") and count more and more.
Nassim Taleb · Not checkable · 2019 |
|
| Not researched |
Yann LeCun | Factual |
2019 |
Large-scale embedding methods that learn vectors per object and subject are particularly efficient for recommender systems.
Yann LeCun · Factual · 2019 |
AI capabilities |
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
We hold that calibration in frequency (probability) space is an academic exercise (in the bad sense of the word) that mistracks real life…
Nassim Taleb · Not checkable · 2019 |
|
| Not researched |
Yann LeCun | Factual |
2019 |
The performance of spiking neuron circuits seems considerably inferior to that of traditional digital architectures for realistic ConvNet-type networks.
Yann LeCun · Factual · 2019 |
AI compute, hardware and data |
| Not researched |
Nassim Taleb | Citation |
2019 |
Reference [29] shows that a stable model needs to be linear in errors.
Nassim Taleb · Citation · 2019 |
Science and research methods |
| Mostly true |
Nassim Taleb | Factual |
2019 |
Value at Risk is not a coherent risk measure because it violates subadditivity.
Nassim Taleb · Factual · 2019 |
Economics and markets |
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
This is rather a problem with the experiment: people do not necessarily separate probabilities from payoffs.
Nassim Taleb · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
meaning the fatter the tails, the worse the uncertainty about the expected payoff and the larger the difference between I1 and I2.
Nassim Taleb · Not checkable · 2019 |
|
| Not researched |
Yann LeCun | Factual |
2019 |
RL's sample inefficiency is not a problem in easy-to-simulate, fully observable, discrete-state environments such as go or chess.
Yann LeCun · Factual · 2019 |
AI capabilities |
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
This example is particularly potent as it has no verbalistic correspondence but can be understood by option traders and machine learning.
Nassim Taleb · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
unwarranted conclusions that people overestimate tail risk have been chronically made since
Nassim Taleb · Not checkable · 2019 |
|
| Not checkable |
Yann LeCun | Not checkable |
2019 |
The number of trials required is large because the feedback from the environment is information-poor.
Yann LeCun · Not checkable · 2019 |
|
| Mostly true |
Nassim Taleb | Factual |
2019 |
Conditional Value at Risk (CVaR) satisfies the coherence (subadditivity) bound.
Nassim Taleb · Factual · 2019 |
Economics and markets |
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
Succession 2 (fat tails): {0, 0, 0, 0, 0, 0, 5}. Mean variation= 0.71 (same). P/L: g(x) = −18 (bust, really bust, but prediction is…
Nassim Taleb · Not checkable · 2019 |
|
| Mostly false |
Nassim Taleb | Factual |
2019 |
Financial regulators compute the recovery rate as the unconditional expected value of collateral, without conditioning on the default event.
Nassim Taleb · Factual · 2019 |
Economics and markets |
| Unresolvable |
Yann LeCun | Prediction |
2019 |
Chances are that the bulk of computation in future deep learning systems will still consist primarily of convolutions.
Yann LeCun · Prediction · 2019 |
AI compute, hardware and data |
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
Simply, their g(.) function was nonlinear not properly adapted to naive forecasting of the first moment of the random variable.
Nassim Taleb · Not checkable · 2019 |
|
| Not researched |
Yann LeCun | Factual |
2019 |
Pooling makes a ConvNet representation robust to small shifts in the location of distinctive features.
Yann LeCun · Factual · 2019 |
AI capabilities |
| Unresolvable |
Nassim Taleb | Factual |
2019 |
The Brier score is favored because it does not allow arbitrage and requires perfect probabilistic calibration.
Nassim Taleb · Factual · 2019 |
Science and research methods |
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
the probability distribution of probability (sic) is robust –in other words the distribution of the probability of X doesn’t depend on the…
Nassim Taleb · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
making "calibration" mistakes where it doesn’t matter or can be helpful should be encouraged, not penalized.
Nassim Taleb · Not checkable · 2019 |
|
| Unresolvable |
Yann LeCun | Prediction |
2019 |
Programmable resistor technology may become sufficiently compact, reliable, durable and configurable for deep learning applications.
Yann LeCun · Prediction · 2019 |
AI compute, hardware and data |
| Not researched |
Yann LeCun | Factual |
2019 |
By the late 1990s, between 10% and 20% of all bank checks in the US were automatically read by the ConvNet-based system developed by LeCun's team.
Yann LeCun · Factual · 2019 |
Economics and markets |
| Not checkable |
Yann LeCun | Not checkable |
2019 |
This kind of model can be used to predict multiple scenarios of how surrounding cars are going to move, and to plan a driving policy…
Yann LeCun · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
for under Gaussian distributions the effect can be marginal, which may have lulled the psychology literature into the conflation.
Nassim Taleb · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
It is practically impossible to hedge one with another
Nassim Taleb · Not checkable · 2019 |
|
| Not researched |
Yann LeCun | Factual |
2019 |
Almost all practical applications of deep learning use supervised learning, with a tiny minority using reinforcement learning.
Yann LeCun · Factual · 2019 |
AI capabilities |
| Resolved true |
Nassim Taleb | Factual |
2019 |
XIV, a volatility-linked exchange traded note, went bust in February 2018 after a sudden jump in volatility, while having been able to "predict" volatility correctly.
Nassim Taleb · Factual · 2019 |
Economics and markets |
| Not checkable |
Yann LeCun | Not checkable |
2019 |
Video prediction is a subject of wide interest because it may allow future systems, such as robots or self-driving cars, to predict what is…
Yann LeCun · Not checkable · 2019 |
|
| Not researched |
Yann LeCun | Factual |
2019 |
Deep learning systems tend to work better the larger they are made.
Yann LeCun · Factual · 2019 |
AI capabilities |
| Not checkable |
Yann LeCun | Not checkable |
2019 |
new ideas seem limited by the available hardware (and software)
Yann LeCun · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
Some nonlinear payoff functions, while not lending themselves to verbalistic expressions and "forecasts", can be well captured by ML or…
Nassim Taleb · Not checkable · 2019 |
|
| Not checkable |
Yann LeCun | Not checkable |
2019 |
The word “architecture” in the context of DL designates the graph of functional modules, not the architecture of the underlying hardware.
Yann LeCun · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
We start with yj = (f − 1Aj), the difference between a continuous Beta distributed random variable and a discrete Bernoulli one, both…
Nassim Taleb · Not checkable · 2019 |
|
| Not researched |
Yann LeCun | Factual |
2019 |
The upper layers of ConvNets are typically less compute bound but more memory bound.
Yann LeCun · Factual · 2019 |
AI compute, hardware and data |
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
reducing these events to verbalistic "collapse", "disaster" is equivalent to a health insurance payout of a lump sum if one is "very ill"
Nassim Taleb · Not checkable · 2019 |
|
| Not researched |
Yann LeCun | Factual |
2019 |
Variations of GANs have produced stunning results in image generation.
Yann LeCun · Factual · 2019 |
AI capabilities |
| Not checkable |
Yann LeCun | Not checkable |
2019 |
system-level performance is considerably more important than raw speed
Yann LeCun · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
If a mistake doesn’t cost you anything –or helps you survive or improve your outcomes– it is clearly not a mistake.
Nassim Taleb · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
Owing to Jensen’s inequality, the discrepancy (I1 − I2) increases under parameter uncertainty, expressed in higher kurtosis
Nassim Taleb · Not checkable · 2019 |
|
| Resolved true |
Nassim Taleb | Factual |
2019 |
For a continuous random variable X with CDF F_X, the variable U = F_X(X) is uniform between 0 and 1 absent additional information.
Nassim Taleb · Factual · 2019 |
Science and research methods |
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
Many perceived "biases" are shown to be just mischaracterizations by psychologists.
Nassim Taleb · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
prices are not binary bets.
Nassim Taleb · Not checkable · 2019 |
|
| Not checkable |
Yann LeCun | Not checkable |
2019 |
Good results are essential but not sufficient.
Yann LeCun · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
Binary forecasts are not particularly prone to model error; real world payoffs are.
Nassim Taleb · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
which works well in situations where one has to forecast the first moment of the distribution and the dispersion around it.
Nassim Taleb · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
There can be considerable mathematical and statistical differences between the following two items:
Nassim Taleb · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
P/L tautologically corresponds to the real world distribution, with an absorbing barrier at the survival condition.
Nassim Taleb · Not checkable · 2019 |
|
| Unresolvable |
Yann LeCun | Prediction |
2019 |
Augmented/virtual reality and telepresence applications will require extremely low-power ASICs for DL inference.
Yann LeCun · Prediction · 2019 |
AI compute, hardware and data |
| Not checkable |
Yann LeCun | Not checkable |
2019 |
The main difficulty is that predicting the future of a video is not achievable exactly because the world is not entirely predictable.
Yann LeCun · Not checkable · 2019 |
|
| Not checkable |
Yann LeCun | Not checkable |
2019 |
What has accounted for the success of GPGPU for DL is their wide availability, generality, programmability, and well-supported software…
Yann LeCun · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
This insures the predictor meets the constraint of staying above water.
Nassim Taleb · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
We have shown that, in risk management (and while forecasting socio-economic and financial variables), one should never operate in…
Nassim Taleb · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
most businesses have severely skewed payoffs, so being calibrated in probability is meaningless.
Nassim Taleb · Not checkable · 2019 |
|
| Not checkable |
Yann LeCun | Not checkable |
2019 |
One hypothesis is that this missing ingredient is self-supervised learning.
Yann LeCun · Not checkable · 2019 |
|
| Not checkable |
Yann LeCun | Not checkable |
2019 |
But there is no algorithmic reason to batch multiple samples.
Yann LeCun · Not checkable · 2019 |
|
| Not researched |
Yann LeCun | Factual |
2019 |
In practical ConvNet applications, much of the computation is spent on convolutions in the lower layers.
Yann LeCun · Factual · 2019 |
AI compute, hardware and data |
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
given that our exposures in the real world rarely map to I1.
Nassim Taleb · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
Accordingly, Fig. I-5 makes little sense empirically.
Nassim Taleb · Not checkable · 2019 |
|
| Not checkable |
Yann LeCun | Not checkable |
2019 |
non-programmability was a major limitation
Yann LeCun · Not checkable · 2019 |
|
| Not checkable |
Yann LeCun | Not checkable |
2019 |
The use of predictive models that not only predict the evolution of the environment, but also predict the consequences of actions, is key…
Yann LeCun · Not checkable · 2019 |
|
| Mostly true |
Yann LeCun | Factual |
2019 |
The ConvNet architecture was loosely inspired by the ventral pathway in the visual cortex.
Yann LeCun · Factual · 2019 |
AI capabilities |
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
These appear to be at best one order of magnitude fewer in decisions entailing formal predictions.
Nassim Taleb · Not checkable · 2019 |
|
| Mostly true |
Yann LeCun | Factual |
2019 |
ReLU nonlinearities and dropout regularization enabled purely supervised backprop training of very deep networks with a dozen layers.
Yann LeCun · Factual · 2019 |
AI capabilities |
| Unresolvable |
Nassim Taleb | Factual |
2019 |
Taleb states that one can further verify the Brier score has thinner tails than the Gaussian, as its kurtosis is lower than 3.
Nassim Taleb · Factual · 2019 |
Science and research methods |
| Mostly true |
Nassim Taleb | Factual |
2019 |
Most natural and socio-economic variables are continuous and have no compact support, so no exact upper bound is known.
Nassim Taleb · Factual · 2019 |
Science and research methods |
| Not checkable |
Yann LeCun | Not checkable |
2019 |
Learning Applied to Document Recognition
Yann LeCun · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
At the core, calibration metrics such as the Brier score are always thin-tailed, when the variable under measurement is fat-tailed, which…
Nassim Taleb · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
As we can show here, prediction markets may be useful for gamblers, but they cannot hedge economic exposures.
Nassim Taleb · Not checkable · 2019 |
|
| Not checkable |
Yann LeCun | Not checkable |
2019 |
The key advantage of DL is that it alleviates the need to hand-design a feature extractor, as would be required with traditional machine…
Yann LeCun · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
Risk management is about changing the payoff function g(.) rather than making "good forecasts".
Nassim Taleb · Not checkable · 2019 |
|
| Not checkable |
Yann LeCun | Not checkable |
2019 |
If the energyminimizing inference procedure is gradient-based, inference hardware will need to support back-propagation.
Yann LeCun · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
Further, we can extract its probability density function of the Brier score for N = 1
Nassim Taleb · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
Binary forecasts and calibration metrics via the Brier score belong to the thin-tailed class.
Nassim Taleb · Not checkable · 2019 |
|
| Unresolvable |
Yann LeCun | Factual |
2019 |
Dataflow stream processors have become a popular design for ConvNet accelerators, such as the Eyeriss project.
Yann LeCun · Factual · 2019 |
AI compute, hardware and data |
| Unresolvable |
Yann LeCun | Prediction |
2019 |
The ultimate solution to DL power constraints may be exploiting sparse activations, perhaps with event-based computation.
Yann LeCun · Prediction · 2019 |
AI compute, hardware and data |
| Not checkable |
Yann LeCun | Not checkable |
2019 |
The tools at our disposal fashion our thoughts more than we care to admit.
Yann LeCun · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
Clearly we are dealing with 1) f beta distributed
Nassim Taleb · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
probabilistic calibration requires estimations of the zeroth moment while the real world requires all moments
Nassim Taleb · Not checkable · 2019 |
|
| Not researched |
Yann LeCun | Factual |
2019 |
The original 1957 Perceptron was a dedicated analog computer whose weights were implemented with motorized potentiometers.
Yann LeCun · Factual · 2019 |
AI compute, hardware and data |
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
We can proceed to prove directly from there the convergence in distribution for the average
Nassim Taleb · Not checkable · 2019 |
|
| Not researched |
Yann LeCun | Citation |
2019 |
Reference [35] shows that an 8-bit logarithmic number system with a Kulisch accumulator significantly reduces power consumption with negligible accuracy loss for ResNet-50 on ImageNet.
Yann LeCun · Citation · 2019 |
AI compute, hardware and data |
| Unresolvable |
Yann LeCun | Prediction |
2019 |
What the AI hardware and software communities produce over the next few years will shape AI research for the coming decades.
Yann LeCun · Prediction · 2019 |
AI compute, hardware and data |
| Not checkable |
Yann LeCun | Not checkable |
2019 |
Not all layers are represented [37].
Yann LeCun · Not checkable · 2019 |
|
| Unresolvable |
Yann LeCun | Prediction |
2019 |
Self-supervised latent-variable predictive models will be the centerpiece of intelligent systems based on model-predictive control and model-based reinforcement learning, e.g. for robotic manipulation and autonomous driving.
Yann LeCun · Prediction · 2019 |
AI capabilities |
| Not checkable |
Yann LeCun | Not checkable |
2019 |
The ideal architecture is a specialized DL-inference accelerator sitting in a standard data-center
Yann LeCun · Not checkable · 2019 |
|
| Not checkable |
Yann LeCun | Not checkable |
2019 |
The idea of self-supervised learning is to train a machine to predict any subset of its input from other subsets
Yann LeCun · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
Inconsistent opinions therefore would lead to a violation of arbitrage rules
Nassim Taleb · Not checkable · 2019 |
|
| Not checkable |
Yann LeCun | Not checkable |
2019 |
It is very unclear whether analog implementations provide any power dissipation advantages over digital
Yann LeCun · Not checkable · 2019 |
|
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
can be close to I2 only under thin-tailed distributions, not under the fat tailed ones.
Nassim Taleb · Not checkable · 2019 |
|
| Not researched |
Yann LeCun | Factual |
2019 |
The amount of raw video data available to train self-supervised systems is essentially unlimited.
Yann LeCun · Factual · 2019 |
AI compute, hardware and data |
| Not checkable |
Nassim Taleb | Not checkable |
2019 |
This in a nutshell shows how, in the world of finance, "bad" forecasters can make great traders and decision makers, and vice versa.
Nassim Taleb · Not checkable · 2019 |
|
| Mostly false |
Yann LeCun | Factual |
2019 |
Users upload 2 to 3 billion photos on Facebook every day.
Yann LeCun · Factual · 2019 |
AI compute, hardware and data |
| Not checkable |
Yann LeCun | Not checkable |
2019 |
The scores are normalized to sum to one.
Yann LeCun · Not checkable · 2019 |
|
| Not researched |
Yann LeCun | Factual |
2019 |
The SN system inaugurated the idea of neural networks as computational graphs of standard parameterized modules with automatic differentiation.
Yann LeCun · Factual · 2019 |
AI capabilities |
| Mostly false |
Nassim Taleb | Factual |
2019 |
Since the 1960s, practically all economic and informational variables have been shown to belong to the D2 class, or at least the intermediate subexponential class, which includes the lognormal.
Nassim Taleb · Factual · 2019 |
Economics and markets |