Superheroes of Deep Learning Vol 2: Machine Learning for Healthcare

Full PDFs free on GitHub. To support us, visit Patreon.

Cover of Volume 2: Machine Learning for Healthcare, of the Superheroes of Deep Learning comic series
The year is 2021. A self-replicating 0.12 micron protein has taken the world hostage and inconvenienced millions of conservative politicians, who suddenly care about civil liberties. Flush with cash and deep minds, the AI illuminati have convened a great conference to confront their common foe
Meanwhile spurned by the ML elite, DAG-man de-stresses on a California beach with some sunnies and funnies…
[DAG-Man] On beach, lounge chair, sipping beach cocktail with little umbrella & pineapple slice,
Reading DL Superheroes Vol 1. 

[Tweeting out on Blackberry --- “Translation please? I grew up with Popeye and Little Red Riding Hood....” ]
[Somewhere in Westchester county...]

[Giant mansion, inspired by Prof X’s school for the gifted—long trail of mafia town cars lined up outside. At the entrance is a long line and a registration desk, with 1000s of people in line to get their badge and nametag]

[Poster outside says International Conference for ML Superheroes 2021]

[All the factions of the ML community are here, The DL Superheroes, Rigor Police (dressed up like British bobbies), The Algorithmic Justice League, The Causal Conspirators, and the Symbol Slappers ]
Anon char 1: How did the Superheroes afford this place?
Anon char 2: Was the Element AI acquihire more lucrative than we thought?
Anon char 3: ... I heard Captain Convolution got in early on Gamestop
Anon char 4: Shhh… he’s about to speak...

The GodFather: I look around, I look around, 
and I see a lot of familiar faces.
[nods at each]
Don Valiant, Donna Boulamwini, GANfather....
It’s not every day that we gather 
the entire family under one roof.

We unite here today 
And put aside our differences
because a gathering threat 
imperils our common interests
You may already know...
[Tensorial Professor]
The curse of dimensionality?

[Kernel Scholkopf]

[Code Poet]

[The GANfather]
[Enforcer puts GANfather in headlock]

[In background, Enforcer and GANfather being dragged out by security (the rigor police)]

I am talking about the greatest threat 
ever to imperil our noble struggle 
to create intelligence in silico...

Severe acute respiratory syndrome 
coronavirus SARS-CoV-2.
[Anon attendee in the aisle at the QA mic:  
Why ML? We are not epidemiologists? 
Shouldn’t we have invited some experts…?]

[GodFather response]
We didn’t need signal processing experts to solve speech transcription 

[badly simultaneously transcribed on Jumbotron as: 
“We hidden eat pig nail process experts to salve peach prescription”]

We didn’t need linguists to solve machine translation
[simultaneously translated as [in hindi]: “hamen shareer parivartan karne ke liye aashavadee ki chahat nahi hai”

We didn’t need opthamologists to solve computer vision

[Anon attendee on phone posting photo of Godfather to twitter—Automatically captioned as “mafia boss inspires nerds to take collective action”... scratches chin]

(Same) Anon attendee: “… that was actually pretty good”
But just to be safe, and in the spirit of interdisciplinarity, we invited the nation’s top infectious disease expert, Dr. Anthony Fauci.

Listen to me, I know you all are superheroes, but you should all be wearing masks. And you should all go home. This is a superspreader event. I can’t believe I agreed to show up here…. but the Godfather can be very persuasive
[Flash to Fauci’s memory of the Godfather in palpatine robe, dark background neuron activation lightning shooting out of his finger, one eye poking out from hood, evil stare]
I now turn over the microphone to our PC chairs - Code Poet, Benchmark and Captain Convolution. 

[Captain Convolution kisses the Godfather’s ring]

[Benchmark, Captain Convolution, Code Poet take over the mic]
Three months prior to convening this conference, 
We issued a clandestine call for papers
 to the world’s top ML heroes.

Amazingly we received 3,409 submissions
proposing ML schemes to save the world,
a 72% increase since the last global crisis.
Anticipating a lack of poster space,
we instructed the area chairs 
to desk reject 99.9% of papers,

Leaving us with the 4 best proposals.

To locate each talk please 
download the WHOVA app.

We will reconvene at 3pm 
After socializing over soggy omelettes,
non-potable coffee, 
surplus ElementAI swag,
and some aggressive recruiting by our sponsors.

Please enjoy the conference.
Track 1 — Learning Theory
Speaker Les, the Valiant

[Les, the Valiant]

Trots up 
 “Is that his real name?!?”

[Les, the Valiant]
Friends, I've solved it! [gesticulating wildly with his lance]
No methodology gives us a greater chance
to contain the coronavirus 
than probably approximately correct learning!
[Les, the Valiant]
[Furiously scribbles with his lance
then raises his hands like a magician]

“By the combined powers of Hoeffding, Chernoff,
Vapnik, and Chervonenkis,
I hereby pronounce that with probability at least 1 - \delta, 
the region in which coronavirus poses a risk 
shall lie inside a sphere centered at this point 
[thumping lance into ground]  
and whose radius is less than or equal to \epsilon!”  

Hands sustain magic spell pose, face remains intense
 Is it working yet? 

Yes, my child. It is guaranteed to work! 
The bound obtains from the very axioms of mathematics! 

[Anon]  [Looks at phone sees record cases on CNN]
Umm …  why are people still getting sick?

Well, …. in practice ...  the smallest epsilon that we can guarantee with high probability is 10^9 meters?

[Zooms out shows a force-field like bound surrounding the Earth (and including the moon) --- area outside bound labeled “probably no coronavirus”, inside: possibly a lot of coronavirus]
Coffee Break

[Anon to Count Vapnik]
Hey, Count Vapnik, you going to the DeepMind party?

Not invited.

You must be shattered

[Vapnik gives stink eye]
Track 2—Metalearning
Speaker: Iris

It’s not enough to learn to fight this virus.
We must learn to learn to fight this virus! 
Or maybe, we must learn to learn to learn to ….
[Each layer of recursion deeper, a server farm in the background increases in size by an order of magnitude. Eventually, at the final panel, it goes mushroom cloud]
[...continues in tatters, smoke damage on walls from data center explosion, like wile e coyote post explosion]

By the power of metalearning,
We can learn from past pandemics
Transferring that experience
To guide the covid19 response 

[Raises finger as if to ask question]
Umm hello… experienced metalearner, right here.

[Fauci Thought bubble]
...maybe they’d take me more seriously
If I changed my name to DINOSAUR?
Poster Session

[Anon to Brute Force]
Hey Brute Force, are you going to the AnthropicAI party?

[Brute Force]
I am not open to it.
Track 3 — Kernel Methods
Speaker: Kernel Scholkopf


You see, these viruses can be elusive targets,
packed all together in this puny 3D space.

However, after my kernel machine projects the virus
into an infinite dimensional Reproducing Kernel Hilbert Space,
It will be a simple matter to divide and conquer them!
[wields his kernel machine: projects (a few) viruses into Reproducing Kernel Hilbert space, a realm in which the viruses are large targets easy to hit, and  Scholkopf manifests his super powers (turning into Kernel Scholköpf --- where he has the full get up and is also super jacked]
Scholkopf smash!!!!!
[The GANfather] 
Nice job, Kernel! Now we just need to scale your weapon up to kill all 1000000000000000000000… viruses! 
How does its power scale with
greater numbers of instances?

[Kernel shows look of terror, converts back to real world Scholkopf casual style, sits resigned (slumped over,  head on hands) ]
Coffee Break

[MOOC to Uberman]
So, how does a superhero change their name?

Hey! You guys going to the Uber AI party?

Sourfully stares into the distance
Track 4 — Tensor Methods
Speaker: The Tensorial Professor

Nice try, but you made a critical mistake. 
You thought  this virus would be 
easier to handle in higher dimensions.
But this way of thinking is cursed.

Everything is a tensor,
and this virus is no exception.
First, we will decompose.
And then, we will dispose.
stand back!
The Tensorial Professor — Decomposes the virus, sharding it into lower-dimensional factors (seen as 2D frames)
Remarkable! The tensor factorization was too efficient. Each of the factors retained enough information to continue to reproduce as a viable virus!!
[Scene of the apocalypse spreading forth from (wherever they are based)]
[DAGman wearing that suit, on the beach reading DL Superheroes Volume 1]

[Looks up, sees the advancing doom. Immediately recognizes it as the mishaps of those associative learning miscreants. ]
when will these curve fitters finally learn?
[Walks over to his super professor-y office. ]
[Full floor to ceiling bookshelves like library, one full shelf of copies of “Book of Why”—each in a different language, drafting table in center of desk]
When I first conceived of assembling this great edifice of causal abstraction, I knew that this day might come. There’s only one option left—to intervene on the very fabric of reality itself, giving humanity a second chance in a counterfactual universe.
[Rushes about, preparing for the intervention]

This must be how Archimedes felt
when the great epiphany struck,
or Leopold Szilard, upon recognizing
the great power locked in the atom,
or John Coltrane, upon mastering
 the sheets of sound,
or… the chef at Wexler’s, 
when he discovered the perfect ratio
of whitefish, mayo, sour cream, and dill…
[1hr later]

Well, here goes nothing.

[Dives in into the DAG/The Timeline, head first. DAG-Man is rendered into Mario-form on the other side
Inside DAG, pearl still has suit but has super moves - he starts karate chopping edges, flipping directions of edges, and throwing fireballs at nodes
[Some hours later… or earlier, in another possible world]

[Pearl (in suit), Hinton, & Assorted other ML heroes, all lined up in beach chairs in LA) California sun in the background].

Nice to see you again, old friend.

Strange, I never seem to recall planning to come here.
(More explicit; i never seem to recall planning these trips)
In fact, I was just planning a conference...

Fate must have intervened.
Falaah Arif Khan - @FalaahArifKhan
Co-creator, writer, artist

Zachary C Lipton - @zacharylipton
Co-creator, writer

Stay up to date on all the adventures of the DL Superheroes at

Support our work on Patreon:

Cite this volume as:
Falaah Arif Khan and Zachary C Lipton. Superheroes of Deep Learning Vol 2: Machine Learning for Healthcare (2021)
[post-credit scene]
The DL superheroes are celebrating on the sunny California beach. The parchment with The Timeline/Master DAG flies in the wind and into the face of a celebrating Causal Conspirator. He picks up the parchment and looks at it in horror, exclaiming:
"DAG-man, please tell me you remembered to endogenize"

On the horizon a green moon rises and a fire-y red and scarlet tide rises.


Author: Falaah Arif Khan

An Engineer/Scientist by training and an Artist by nature, Falaah is a Research Fellow at the CVIT Lab at IIIT-Hyderabad and an Artist-in-Residence at the Montreal AI Ethics Institute.

5 thoughts on “Superheroes of Deep Learning Vol 2: Machine Learning for Healthcare”

  1. I recognized only some character.
    Mooc is Andrew NG
    Q-Silver is David Silver
    Captain Convolution is Yann Lecun
    DAG-man is Judea Pearl
    Kernel Schölkopf (one of the two Causal Conspirator) is Bernhard Schölkopf
    Enforcer is Jürgen Schmidhuber
    Code Poet (of Algorithmic Justice League) is Joy Buolamwini (also referred as Donna Buolamwini)
    One of the Algorithmic Justice League with dark green costume and red coat is Timnit Gebru
    One of the Algorithmic Justice League with purple costume with yellow smile emoticon is Margaret Mitchell
    The ML Kardashians are Siraj Raval & Lex Fridman
    Ganfather is Ian Goodfellow
    The Godfather is Geoffrey Hinton as speaker in vol. 2
    Les the Valiant (Don Valiant) is Leslie Valiant
    Count Vapnik is Vladimir Vapnik
    Iris with Stanford costume is Chelsea Finn
    Benchmark is ? Fei Fei Li ?

    Then I didn’t recognize:
    The Tensorial Professor ?
    Rigor Police ?
    The Symbol Slappers ?
    The fourth Algorithmic Justic League Component, the one on the right with violet costume ?
    The other causal conspirator that is not Schölkopf ?
    Experienced metalearner ?
    Brute Force ?
    Uber map ?
    The guy who asks people if they are going to party is ?

  2. after some research I found out other possible correspondances:
    Uber Map should be Zoubin Ghahramani
    Tensorial Professor should be Anima Anandkumar
    Rigor police are (left to right) :Nati Srebro, Manfred Warmuth, Shai Ben-David

Leave a Reply

Your email address will not be published. Required fields are marked *