BooglyWoo

Also sounds a bit like Thompson's Lamp.

EDIT: Whoops, that's pretty much the same thing as Zeno's paradox.

I don't want to ban anything; I was just explaining how the potential effects of Popper's intolerance paradox cannot easily be solved with legislation.

Because it's hard to unanimously and unequivocally identify those who are intolerant as such. Owing to the beguiling nature of language and rhetoric, the intolerant can thrive and propagate by arguing that they're not in fact intolerant, and appealing to "free speech". This seems to be the MO of this so-called "alt-right" stuff; legal or constitutional barriers are almost impossible to define or enforce.

As another user TheOtherHobbes noted, humans share this characteristic with insects to some extent:

> Wild solo humans are only a little smarter than wolves individually, but being able to share and externalise invention and learning created a massive advantage.

This seems to resonate with a strain of Philosophy of Mind https://plato.stanford.edu/entries/content-externalism/ which deals with our mental content being distributed not only around the brain and body, but on paper, computers and relations with other people.

In this case, the guy has been spearheading a very high performance, innovative methodology for real-time object detection. Companies interested in that task should be falling over themselves to hire him.

Can't recommend the Mandelbrot book highly enough - he was a student of Paul Levy, and he wrote extensively about why Gaussian is not a good choice to model financial time series.

It's a good article in a lot of ways, and provides some warnings that many neural net evangelists should take to heart, but I agree it has some problems.

It's a bit unclear whether Fchollet is asserting that (A) Deep Learning has fundamental theoretical limitations on what it can achieve, or rather (B) that we have yet to discover ways of extracting human-like performance from it.

Certainly I agree with (B) that the current generation of models are little more than 'pattern matching', and the SOTA CNNs are, at best, something like small pieces of visual cortex or insect brains. But rather than deriding this limitation I'm more impressed at the range of tasks "mere" pattern matching is able to do so well - that's my takeaway.

But I also disagree with the distinction he makes between "local" and "extreme" generalization, or at least would contend that it's not a hard, or particularly meaningful, epistemic distinction. It is totally unsurprising that high-level planning and abstract reasoning capabilities are lacking in neural nets because the tasks we set them are so narrowly focused in scope. A neural net doesn't have a childhood, a desire/need to sustain itself, it doesn't grapple with its identity and mortality, set life goals for itself, forge relationships with others, or ponder the cosmos. And these types of quintessentially human activities are what I believe our capacities for high-level planning, reasoning with formal logic etc. arose to service. For this reason it's not obvious to me that a deep-learning-like system (with sufficient conception of causality, scarcity of resources, sanctity of life and so forth) would ALWAYS have to expend 1000s of fruitless trials crashing the rocket into the moon. It's conceivable that a system could know to develop an internal model of celestial mechanics and use it as a kind of staging area to plan trajectories.

I think there's a danger of questionable philosophy of mind assertions creeping into the discussion here (I've already read several poor or irrelevant expositions of Searle's Chinese Room in the comments). The high-level planning, and "true understanding" stuff sounds very much like what was debated for the last 25 years in philosophy of mind circles, under the rubric of "systematicity" in connectionist computational theories of mind. While I don't want to attempt a single-sentence exposition of this complicated debate, I will say that the requirement for "real understanding" (read systematicity) in AI systems, beyond mechanistic manipulation of tokens, is one that has been often criticised as ill-posed and potentially lacking even in human thought; leading to many movements of the goalposts vis-à-vis what "real understanding" actually is.

It's not clear to me that "real understanding" is not, or at least cannot be legitimately conceptualized as, some kind of geometric transformation from inputs to outputs - not least because vector spaces and their morphisms are pretty general mathematical objects.

EDIT: a word

I think Bart's blood revitalised Mr Burns once.

Have you seen hypernom.com for visualizing 4 dimensions?

"Classical" CV and deep-learning CV needn't be opposing one another.

There are several cases in which the classical approach is emulated by deep networks - implementing the same carefully thought-out pipelines but in a way that leverages representations learned from huge datasets (which are undeniably very powerful).

Some examples are:

* Bags of convolutional features for scalable instance search https://arxiv.org/pdf/1604.04653.pdf

This paper treats each 'pixel' of a CNN activation tensor as a local descriptor, clusters them, and describes an image as a bag-of-visual-words histogram.

* Learned Invariant Feature Transform https://arxiv.org/abs/1603.09114v2

This paper very explicitly emulates the entire SIFT pipeline for computing correspondences across pairs of images

* Inverse compositional spatial transformer networks https://arxiv.org/abs/1612.03897v1

This paper emulates Lucas-Kanade approach to computing the transform between 2 images with differentiable (trainable) components.

Also, don't forget that deformable part models are convolutional networks! https://arxiv.org/abs/1409.5403

I think these images are called "Lissajous Figures".

This site has an excellent series of in-depth interviews with Mandelbrot; profiling his early life, his capacity to think mathematically "in pictures", and goal to link seemingly disparate fields such as "social" and "natural" sciences through fractal mathematics.

EDIT: forgot the link godammit https://www.webofstories.com/play/benoit.mandelbrot/6

Naive question maybe: why do news articles not cite sources and have a references section at the end, like academic articles?

> You would think so. But my dad is a varsity Professor with a PhD, my mom has a Masters in English. They believe literally every cock and bull story they see on Facebook.

I would guess that the difference between you and your parents is years of exposure to the tropes, tone, register and other pragmatics of internet content.

Mandelbrot cites his ability to "think in pictures" as fundamental to his process and insights.

He offers some interesting reflections and anecdotes on this subject in this interview describing his classes préparatoires aux grandes écoles. Apparently his teacher considered him a total wildcard case who would either flunk the exams or pass with flying colours, because of his habit of approaching everything through geometric intuition rather than symbolic manipulation.

https://www.webofstories.com/play/benoit.mandelbrot/6

Me too. My reading of the Blue and Brown books led me to believe Wittgenstein's conception of meaning is inextricably tied up with the notion of "learning" and exposure to language and its use. Rather than meaning being contingent on 'hard' logico-mathematical derivations of formal semantics.

This contrast seems somewhat reminiscent of the complementary approaches of hard-coded rule based AI vs machine learning.

I'm not sure that quite addresses the problem here.

After all there is no clear definition of 'what dogs look like' (in the sense of a collection of logical rules), but deep learning models excel at detecting them, when provided with enough positive examples.

If it's possible for humans to agree on whether a given article is clickbait or not, we should be able to put together an adequate dataset for training a system to classify them too. From the linked article I am unable to discern how the training dataset was labelled.

In other words, the fact that 'clickbait' is a nebulous concept shouldn't preclude machine learning from being able to detect it.

I believe this problem is common to many types of machine learning classifiers (not even just neural nets). See the discussion here https://news.ycombinator.com/item?id=8721650

I'm not convinced that a person who's never seen animals before could tell the difference between all future dogs and cats from a single training example. Humans draw upon a lifetime of learning and experience to achieve this 'one shot learning' capability.

If you take a pre-trained convnet (which, by analogy is like a person who has had 'life experience' of looking at objects), and extract activations for unseen object categories, in many cases you CAN one-shot-learn these new object categories. Try feeding them into a SVM or use L2 distance between test images and the one-shot exemplar image.

On top of this, there's a lot of work on memory-augmented nets and meta-learning for learning new categories on the fly.