Renormalization That ClicksBonus ⑤ / Learning is a bet on coarse-grainability

Why is learning possible? The answer lives in the same place as "why is the universe coarse-grainable?"

Learning is a bet on coarse-grainability Without prior assumptions nothing can be learned (no free lunch).
So if learning works, the world must be a special kind of world ── the world of Episode 4.
And in deep learning there is a part where a genuine renormalization group runs, not a metaphorical one.

Tools you'll need: Episode 4's relevant/irrelevant, Bonus ④'s rate–distortion and knee, exponents and logs The heart of this one: learning = finding out what may be discarded

Bonus ③ showed statistics, and ④ information theory, looking at the same thing as the renormalization group. Last comes AI. This is a field flooded with "it's kind of like…", so this episode begins by sorting claims by how confident we can be: rigorous / promising but disputed / metaphor only. The first tier has real content: take a neural network's width to infinity and you get a free field theory; 1/width acts as an interaction; and a renormalization group flow runs along the depth. That is a calculation, not a poem. Then we close the whole series with one claim ── 〈why is learning possible〉 and 〈why is the universe coarse-grainable〉 are the same question.

01Without assumptions, nothing can be learned

Machine learning has a blunt theorem.

The no-free-lunch theorem (Wolpert 1996)

Averaged over all possible target functions, every learning algorithm performs exactly as well as guessing.

In other words ── there is no universally clever learner. If a learner works, it is because the world has some bias and the learner carries assumptions (an inductive bias) matching it.

So what is our world's bias? Episode 4's answer transfers directly ── this world is layered, with very few directions that matter (relevant) and overwhelmingly many that don't (irrelevant). In Bonus ④'s language, the rate–distortion curve has a knee. Learning is possible because it bets on that bias.

02The enemy of learning is the irrelevant directions

"Discard or die" is not a figure of speech. Collect data in a space with \(D\) features and distances between points progressively lose meaning as \(D\) grows (the curse of dimensionality). Even if only 2 features actually matter, a learner that dutifully looks at the other \(D-2\) is weaker by exactly that much.

In the figure below two learners compete. Same data, same algorithm (nearest neighbours). One difference only ── whether the irrelevant features are discarded.

Figure: the answer depends on only 2 features, but D features exist in total. Blue = a learner that looks only at the 2 relevant features (coarse-grains); red = a learner that looks at all D (does not). Grey dashed = answering the mean without learning. Horizontal axis is the number of training points (log)
coarse-graining learner (the 2 relevant features) non-coarse-graining learner (all D features) no learning at all

Push \(D\) upward. The blue one (which discards) barely notices, while the red one (which doesn't) degrades until it approaches "no learning at all." Same data, same algorithm. The only thing making the difference is what it decided not to look at.

That is the meaning of this episode's title ── learning is finding out what may be discarded. Episode 4 said discarding determines the answer; in machine learning it literally determines the performance.

03Tier 1: a neural network can be written as an effective field theory

Now the sorting. First, the things that are not metaphors.

Infinite width = free field, 1/width = interaction, depth = renormalization flow

Let \(n\) be the width of each layer (its number of neurons). By the central limit theorem (Bonus ③!), as \(n\to\infty\) the output distribution becomes Gaussian ── in field-theory language, a free (non-interacting) theory. At finite \(n\), \(1/n\) corrections appear and behave as interactions. And stacking layers (depth) makes the parameters flow as under a renormalization group.
Roberts–Yaida–Hanin's The Principles of Deep Learning Theory (2022) builds a perturbation theory on exactly this. Not "deep learning resembles the renormalization group" but "the tools of the renormalization group actually compute deep learning" ── the direction is reversed, and this one is real.

What is delightful is that the starting point is the central limit theorem. Bonus ③ called the CLT the simplest example of the renormalization group; that simplest example is now the foundation of the theory of our most complicated machines.

04Tier 1: machine learning discovered a renormalization transformation

Another non-metaphor. Koch-Janusz and Ringel (2018) did the following ── they had a machine search for a coarse-graining rule that maximises the mutual information between what is inside a block and what is outside it.

The machine then found, with nobody telling it, the block-spin transformation of the Ising model (the very thing we did in Episode 4). This is Bonus ④'s information bottleneck working as an actual tool ── the single instruction "keep only what helps predict the outside/future" produces a renormalization transformation.

What these two mean together Physics → AI (using renormalization-group tools to solve deep learning) and AI → Physics (having machine learning discover renormalization transformations) are both actually running. Not "they seem connected" but "traffic already moves in both directions."

05Tiers 2 and 3 ── promising, and metaphor only

Meanwhile this area is full of things that merely sound right. Honest sorting:

TierContentStatus
rigorousinfinite width = Gaussian process, 1/width = interaction, flow along depth (Roberts–Yaida–Hanin)computable; functions as a perturbation theory
rigorousdiscovering renormalization transformations by maximising mutual information (Koch-Janusz–Ringel 2018)rediscovered Ising block spins; works as a tool
rigorouswavelet scattering transforms (Mallat)multiscale, with provable invariance and stability
promising, disputeddiffusion models ≈ the renormalization group. Adding noise in stages = coarse-graining; reversing it = generation; the noise schedule corresponds to scalerigorous correspondences are being attempted. Attractive but unsettled
promising, disputedexplaining neural scaling laws (loss falling as a power law) from data-manifold dimension or spectrapower laws smell of criticality, but explanations are only partial
metaphor onlythe general claim "deep learning = the renormalization group" (generalising Mehta–Schwab 2014)a correspondence in a specific setting; strong objections to generalising it
metaphor onlythe information-bottleneck explanation of deep learning (the "compression phase")criticised on reproducibility; unsettled

When Bonus ② said "the correspondence between deep learning and the renormalization group is a suggestive resemblance, not a theorem," it meant tier 3. Tier 1 can be stated without hedging.

06And the close of the series

From here, the opening question gets an answer from a new angle.

The heart of this bonus

By the no-free-lunch theorem, the very possibility of learning is a strong claim about the world. If learners work, the world must be the kind that splits into a few relevant and a great many irrelevant directions.

why learning is possible
=
why the universe is coarse-grainable

One question. Which means, read backwards ── the fact that learning actually works is empirical evidence that the universe is coarse-grainable. Bonus ② said "a universe that cannot be coarse-grained has nobody in it to ask." We can now put it more sharply: the fact that we can learn anything at all is itself proof that this universe has a knee.

◇ ◇ ◇
The honest line ── read the three tiers as confidence levels

Established: the no-free-lunch theorem (Wolpert 1996); the curse of dimensionality (degradation of nearest-neighbour methods in high dimensions); that neural networks become Gaussian processes in the infinite-width limit and that the \(1/n\) expansion can be treated as a perturbative effective theory (Roberts–Yaida–Hanin 2022 and others); learning renormalization transformations via mutual information (Koch-Janusz & Ringel, Nature Physics 2018); and the invariance and stability of the scattering transform (Mallat).

Limited or unsettled: (1) The infinite-width correspondence describes a particular (lazy / perturbative) regime and does not fully explain feature learning itself. (2) The diffusion-model / renormalization-group correspondence and the explanation of scaling laws are active but unsettled. (3) Generalisations of the Mehta–Schwab correspondence and the information-bottleneck explanation of deep learning both face strong objections and remain undecided. (4) "Why learning is possible = why the universe is coarse-grainable" is this series' claim, not a theorem; the practical import of the no-free-lunch theorem is itself debated (the premise of averaging uniformly over all target functions). (5) The figure is an idealised demonstration with a simple nearest-neighbour algorithm and does not represent the performance of any real learner.

Exercises
  1. What does the no-free-lunch theorem imply about the world?
    See the answer
    Averaged over all target functions every learner ties, so the existence of a learner that works means the world has a bias. Learning's success is a credit to the algorithm and simultaneously evidence that the world is layered.
  2. In the figure, why does only the red curve degrade as D increases?
    See the answer
    Because it includes irrelevant features in its distance computation, so the curse of dimensionality robs "nearby" of meaning. Blue discarded them from the start and is unaffected. The only difference is what it decided not to look at.
  3. What is the difference between "deep learning is the renormalization group" and "the renormalization group's tools compute deep learning"?
    See the answer
    The former is a general claim facing strong objections and remains undecided (tier 3). The latter is a concrete perturbation theory that actually works: infinite width = free field, 1/width = interaction, depth = flow (tier 1). The direction and scope of the claim are completely different.
  4. Why are "why is learning possible" and "why is the universe coarse-grainable" the same question?
    See the answer
    Learning is finding the relevant and discarding the irrelevant, which requires the world to split into "a few relevant + very many irrelevant" ── that is coarse-grainability itself. So the fact that learning actually works is empirical evidence that the universe has a knee (Bonus ④).

Bonus ⑤ summaryLearning is a bet that this world is layered

By the no-free-lunch theorem, there is no universally clever learner. If learning works, it is betting on the world's bias ── and that bias is Episode 4's "few relevant, many irrelevant," Bonus ④'s "knee in the rate–distortion curve." As the figure shows, with the same data and the same algorithm, whether or not you discard the irrelevant decides everything. Learning is finding out what may be discarded.

The connection to AI splits into three tiers. Rigorous: infinite width = free field (starting from the central limit theorem!), 1/width = interaction, depth = renormalization flow. And set machine learning to maximise mutual information and it rediscovers the Ising block-spin transformation on its own. Promising but unsettled: diffusion models ≈ renormalization group; scaling laws as criticality. Metaphor only: the general "deep learning = renormalization group" claim, and the information-bottleneck explanation. Traffic already moves in both directions, physics → AI and AI → physics.

And the close of the series. 〈Why is learning possible〉 and 〈why is the universe coarse-grainable〉 are the same question. Bonus ② said "a universe that cannot be coarse-grained has nobody in it to ask." Now we can go one step further ── the fact that we can actually learn anything is itself the evidence that this universe has a knee. The story that began with the neutron's 880 seconds ends here. The world is built so that things can be thrown away, and that is why we are here.

This document is Bonus Episode ⑤ (the finale) of the "Renormalization That Clicks" series, a reading piece for physics-loving high-schoolers and undergraduates. The no-free-lunch theorem (Wolpert 1996); the degradation of nearest-neighbour methods in high dimensions (the curse of dimensionality); the Gaussian-process behaviour of neural networks in the infinite-width limit and the effective-theory treatment of the \(1/n\) expansion (Roberts–Yaida–Hanin, The Principles of Deep Learning Theory, 2022 and others); learning renormalization transformations by maximising mutual information (Koch-Janusz & Ringel, Nature Physics 2018); and the invariance and stability of the scattering transform (Mallat) ── all established results. That the infinite-width correspondence describes a particular regime and does not fully explain feature learning; that the diffusion-model correspondence and the explanation of neural scaling laws are unsettled; that the general claim "deep learning = the renormalization group" and the information-bottleneck explanation of deep learning face strong objections; and that "why learning is possible = why the universe is coarse-grainable" is this series' claim rather than a theorem ── all spelled out in the body's "honest line" and in the three-tier table. The figure is an idealised demonstration using k-nearest neighbours and does not represent the performance of any real learner. ── To print, use your browser's "Print" and "Save as PDF" (in the print version the slider and answers are frozen and hidden). Related: Bonus ④ / Bonus ② / Episode 4 / Contents / sister series Learning That Clicks.

Print / make a PDF: ⌘+P (Ctrl+P on Windows). On screen, raising the total number of features D breaks only the learner that refuses to coarse-grain. "Resample the data" regenerates the random numbers. "See the answer" opens each solution.