Kawaii LSTMs


I created the anime girl faces with Yanghua Jin et al’s GAN.
Take the link. But it’s… it’s not like I like you or anything. Baka!


Screen Shot 2018-03-13 at 10.41.43 AM
First, you must learn slopes

Slopes are changes in y over changes in x. In calculus, we discover that they are the tangent line to a point on a function.

300px-Tangent_to_a_curve.svgIf you know the inclination of the slope, you know if you are walking up a hill or down a hill, even if the terrain is covered in fog.


Screen Shot 2018-03-13 at 11.08.27 AM
Math is useful in life. *__*

The higher the value of the function, the more error it represents. The lower, the less error.

We wish to know the slope so that we can reduce error. What causes the error function to slither up and down in its error are the parameters. 

If we can’t feel the slope, we don’t know if we should step to the “right” or to the “left” to reduce the error.

Screen Shot 2018-03-13 at 11.07.40 AM
But we ought to make the slope flat right?

Not necessarily. The problem is that we can end up on the tip-top of a hill and also have a flat slope. We want to end up at a minimum. This means that we must follow a procedure: If negative slope, then move right. And if positive slope, move left. Never climb, always slide.

There is a similar procedural mission going on in a neural network except that the sense of error comes from a higher-dimensional slope called a gradient.

Screen Shot 2018-03-13 at 11.08.27 AM
You mean that what you said was wrong?

No, its very similar. The gradient tells you the direction of steepest ascent in a multidimensional terrain. Then, you must step towards the negative gradient.

Oh, I didn’t mention that the terrain was multidimensional? Well it is. There is not a single place where the input goes like in the function I initially showed you.

This means that not only is there fog but that the hills and valleys are beyond human comprehension. We can’t visualize them even if we tried. But like the sense-of-error from slope which guides us down a human-world hill, the gradient guides us down to the bottom in multi-dimensional space.

The neural network is composed of layers. Each layer has a landscape to it, and hence its own w of parameters with its own gradient.

Here is the objective function Q(w) for a single layer:


The goal is to plug in a lucky set of parameters w1 on the neurons of the first layer, the lucky set of parameters w2 on the neurons of the second layer, and so on with the intention of minimizing the function.

We don’t just guess randomly each time, we slide towards the better w based on our sense of error from the gradient. The gradient is revealed at the final layer’s output.

However, we are initially dropped randomly in the function. Our first layer’s w has to be random.

This presents a huge problem. Although all we have to do is calculate the gradient of the error with respect to the parameters w, the weights closer to the end of the network tend to change a lot more than those at the beginning. If the initial w randomly falls on [having the trait of lethargic weight updates], the whole network will barely move.


By the way, weights are a subset of the parameters. Think of each weight/parameter update as an almost magical multidimensional-step in the stroll through the landscape; with every single step determined by the gradient.

The first guide in our multidimensional landscape may happen to have a broken leg, so he cannot explore his environment very well. Yet guide number two and guide number three must receive directions from him. This means that they will also be slower at finding the bottom of the valley.

LSTMs solve this by knowing how to remember. So now let’s look inside an LSTM.

Screen Shot 2018-03-13 at 11.07.40 AM
Now, at last you get to the point?!




Recurrent neural networks are intimately related to sequences and lists. Some RNNs are composed of LSTM units.


Stay tuned for the explanation of what is going on in there!

Deconstructing Paradise’s Qualia-Units

We know that experience has a layered structure. There are many components to a single now. There may be the breath and the field of vision, and a particular feeling tone. These aren’t experienced separately in sequence. They are integrated.

This is one of the axioms mentioned in Giulio Tononi’s Integrated information theory:

Integration: Consciousness is unified: each experience is irreducible to non-interdependent, disjoint subsets of phenomenal distinctions. Thus, I experience a whole visual scene, not the left side of the visual field independent of the right side (and vice versa). For example, the experience of seeing the word “BECAUSE” written in the middle of a blank page is irreducible to an experience of seeing “BE” on the left plus an experience of seeing “CAUSE” on the right. Similarly, seeing a blue book is irreducible to seeing a book without the color blue, plus the color blue without the book.

Some of the other axioms in his theory seem arbitrary or overlapping. You be the judge.

But this particular axiom is a hard one to disagree with. What this suggests is that experience contains “nodes” or “qualia-units.” These come together to create a larger experience – the entirety of now. The entirety of now is also irreducible in this sense.  We do not experience the field of vision and then the music. A slice of now is like “BECAUSE,” inseparable in its components. But yet we can identify that it has components. Sound is not the same as bodily sensation, and yet we can feel them at once. Heat is not the same as suffering, and yet we can feel them at once.

Now imagine that there is a catalogue of all possible nows. A Library of Babel with its inner spacetime filled, not with books, but with each possible multi-sensorial frame of experience. You can select from every “now” that was ever, or could ever be known.

It may turn out that in this library of nows, only 10 are so good as to be indistinguishable in perfection. Out of nostalgia for Homo sapiens sapiens, the bookkeeper calls them Firdaus, Heaven, Nirvana, Moksha, Siddhasila, Shamayim, Omeyocan, Devachan, Omega Point, and Supermind.

If we are willing to grant that such experiences could exist in unexplored regions of mind-configuration space, then the next question is: How can we recognize them?

We must first deconstruct the frame of Firdaus into its individual pixels/qualia-units. In our day-to-day life the equivalent of these qualia-units can be tingling sensations, pressures, thought motion, color, shape, etc. In this frame of Heaven, we cannot know its component units until we know them.

So let’s try to identify proto-Heaven based on the experiences we know are good. The first thing to notice is that different experiences of goodness have different qualities. Take some of the greatest feelings of goodness possible in humans: family love differs from romantic love, food pleasure differs from sexual pleasure, pride of victory differs from deep relaxation, the hedonic indulgences of a masochist differ from those of a neurotypical bacon-eater, which in turn differ from those of an experienced meditator. The pixels on the goodness grid contain all kinds of qualia-units.

Presumably, some selection and arrangement of these pixels, of the micro-pleasures that fall through the net of these words, will yield the greatest experiences of all. Delicious cherry, delicious coolness, and delicious joy, can be sliced into even smaller experiences. Now take each of these experiences and make a grid of them. Every location on the grid represents a qualia-unit and each pixel can light up to some extent or another, based on how present in consciousness it is. Pixels that are off (black) are not a part of present experience.


Each pixel represents a qualia-unit. Some are warmth-like, some compose flavors, some compose dense pleasures, others – sparkly pleasures, others compose the perception of beauty. The entire grid is a now – a whole of experience.

How many pixels exist in the real world? Could it be billions? –Or surprisingly few?

The true Nirvana’s and Moksha’s in the posthuman Library of Babel would be the perfect shapes, those that arise when each qualia-unit dosage is just right. Somewhere in this grid lies a solution to the puzzle of existence, a combination of valences and aesthetics that meticulously fine-tune bliss.

Sadly, we do not know the solution to the grid beforehand. Our minds are empty of this knowledge in the same way that shrimp are empty of the number nine.

(Just because shrimp don’t know about nine doesn’t mean that nine doesn’t exist. I have nine tabs open in the browser. And goddamnit, the universe hinges on those nine tabs really being open.
We cannot say the tabs are real but the number is imaginary. We cannot say that the brain is real but experiences are imaginary.)

I hypothesize that the ultimate shape(s) can be known, but not by humans. When asking to become a mind that is as comfortable knowing the shapes as comfortably as we know numbers, you are asking a shrimp to become the Uber driver. There is no transmutation of souls: for both the shrimp and the Uber driver lack one.

In other words, reconfiguration of matter at such a drastic magnitude entails complete annihilation.

As humans, we can do two things to crack the puzzle. The first is to be told, and to believe. This method has a negative success rate evidenced by the lineage’s attempt to trick itself with holy books. Advanced aliens might be more credible sources if they behave in recognizably benevolent ways. But let’s be real: who would follow their map when it would necessarily entail re-engineering the entire ape brain?

But the second thing we can do is stumble around new regions of mindspace via gradual ascent to transhumanity. Neural mesh here, targeted amygdala calcification there, and so on. If this is the approach, then we can gradually become better at recognizing the paradises.

The strategy is to take what reality gives us: some experience that can be decomposed into its components; catalogue these components in 2-d; for a single slice of now, track which components are at play.

(I am elaborately visualizing someone selling data from his brain by filling his connectome with nanobots that record his neural activity and send it to be analyzed on a far away lab with neon screens that bleep with the qualia-unit grid indicating which experiences are on.)

Then transfer the qualia-units into a linear array. And prepare to learn the way with the power of gradient descent and minimization of the cost function. Many training samples from many people reporting peak experiences.







[Memes -> Genes, Media -> Drugs] Yields Dystopia

Cultures do not exert their effects in isolation of one another, but interact together in complex networks. In the coming years, sophisticated methods will be developed to leverage culture-culture interaction (CCI) network structure to improve several stages of the media discovery process. Network based methods will be applied to predict media targets, media side effects, and new propagandistic indications. Previous network-based characterizations of media effects focused on the small number of known media targets, i.e., direct binding partners of media. However, media affects many more memes than its targets – it can profoundly affect the civilization’s memeplex.

For the first time, we use networks to characterize memes that are differentially regulated by media. We found that media-regulated memes differed from media targets in terms of function, regional localizations, and neural properties. Media targets mainly included receptors on the plasma membrane of civilization (the software interface), down-regulated memes were largely in the nucleus (the older generation) and were enriched for memetic binding, and memes lacking media relationships were enriched in the extracellular region of civilization (the isolated sub-cultures). Network topology analysis indicated several significant graph properties, including high degree and betweenness for the media targets and media-regulated memes, though possibly due to network biases. Topological analysis also showed that cultures of down-regulated memes appear to be frequently involved in memeplexes. Analyzing network distances between regulated memes, we found that memes regulated by structurally similar media were significantly closer than memes regulated by dissimilar media. Finally, network centrality of media’s differentially regulated memes correlated significantly with media toxicity.

The Happiness Battles: Aristotle vs. Nietzsche vs. Buddha vs. Mencius vs. a Rabbi

A boiling red ocean of demented, feisty, irrational, fearful, hateful, ungrateful beings wailed in sorrowful pain. As if this repulsive existence had pitied them, a crack of kind light fissured their stormy sky. From this window into another world, descended an old man, Aristotle.

Aristotle paced back and forth in a stone slab that floated on the tears of the creatures below, his white tunic fluttering about in his cool wind. Pitying them, he decided to say some words: “Behold, ye. I have the solution for your ailment, your unrighteous sorrow and fear.”

Some eyes flickered with anger, others were buried in the sea of blubber and sharp elbows, but a few managed to catch fleeting sights of the man above.

“The solution is to fulfill your nature. And your nature is to reason.”

The sea festered along, its motion untainted by the hollow word. After all, what did it mean to reason? To one who is helplessly buoyed upward by the bodies all around and crashed against these same at the wave’s break, what the hell can reason mean?

Aristotle dug into his robes for thermometers and rulers, and dropped these like fish food on the masses. “Go on, establish facts in accord with experience. Do it repeatedly, observe carefully, and adhere to the rules of logic. This is what you must do.”

The beings fumbled the instruments in their decrepit hands, they were confused and attempted to measure everything: legs, teeth, and tongues were all targeted and pinned down by the most fanatical of the bunch. After a long frenzy they had accumulated information, and had established tough logic enforcement, but yet this did not stop the calamity of their heaving; the sea toiled on. The waves changed their form, but not their fundamental nature.

From the fission in the sky descended an angry mustache-man. “Behold, ye. I am Nietzsche. And I say that your suffering is valuable. There is nothing to fix here, toil on. Crack at the ribs; earn your worth and achievement!” With his head raised in self-entrancing speech, Nietzsche could not see where he was stepping, and fell to the call of gravity. As cutting as his voice had resounded, so cutting was his sudden slip into the hellish torture down below.

Next from on high, a being materialized from the light: the Buddha. He descended cross-legged and haloed. “What a way to eat your words,” he said in a loving and compassionate tone to the poor Nietzsche being digested below. “Behold, ye. You must follow the Noble Eightfold Path. This will lead to the ultimate freedom from suffering.”

At this, Aristotle scoffed forcefully. “You are a fool Buddha,” said Aristotle. “For one thing, they have learned to reason, and cannot simply accept your views.”

“That is fine, they must come to see the truth for themselves,” replied Buddha.

“But how can they know that they won’t be wasting their time?” said Aristotle.

“Look who’s talking. You gave them a massive free-range game to play, with no concise direction.”

Aristotle would have blushed had he not been so senile, and replied in turn, “You call this Eightfold Path of yours concise? This is the epitome of an oxymoron. There are too many things to do and no clear percentage of time that should be devoted to each aspect. The truth is we are both incapable of specifying a coherent function for these poor fools.”

The Buddha sat for a long moment knowing Aristotle was right, and so decided to amend his speech. “Okay listen up, and behold, ye … once again. I say unto you that you must overcome craving in all forms. Only this will achieve ultimate happiness.”

“Overcome craving!?” Aristotle was flabbergasted. “If they took you to your word, the rational behavior would be to commit suicide. That would surely overcome craving in all forms. And remember, you can’t say rebirth anymooore.”

“Aristotle, you have not seen the path in my words clearly enough. I said overcome all forms of craving. Craving to not exist should also be overcome.”

“Now that’s just nonsensical.”

“Aristotle,” the Buddha smiled kindly, “the action I recommend is not one of rational analysis, which would just be the fermentation of thought. No, rather, it is a direct knowing. A direct instruction to not crave in this instance. If they can remember this instruction often enough, and thus press repeat on this behavior of non-craving, then they will be much happier.”
At the sound of the Buddha’s words, the sea grew calmer, almost depressingly so. Aristotle looked down and saw the soup of animas in gentled sorrow. This was certainly not the scary hell he had once descended upon, but neither was it what something in his most inner-being desired it to be. But he could no longer argue, and simply wished it did not end this way. He sat and sat, contemplating the situation with great discomfort as the enlightened meditator sat in some strange realm beyond desires.
Like a prayer answered by the mysterious essence of life, descended Mencius and a Rabbi. “Ahh… You two, what have you done?” they tromboned in unison.

Aristotle responded, “You mean, what has he done. I… I wanted them to fulfill something, some ideal… something that was human.”

“Pity you, who are pierced at the limbs by the chains of Samsara. Liberation from this existence is what should be sought,” said Buddha.

Mencius and the Rabbi, who had been arguing on their descent through the wormhole, realized at once that their views were much closer than those of the other two. They both believed in emphasizing something higher, something unattainable and glorious which should be bathed in and praised.

The Rabbi called this, this which was the reason for beautiful songs, God; Mencius called it the higher-self in all people. In unison they said: “You are both wrong, what these people need is not to fulfill a single function. They need to exist at the nexus between what is base and what is most-high, ever-seeking to indulge in that which is most-high. Being the Utmost defeats the purpose.”

Hearing these words, Nietzsche activated Transhuman cyborg mode and ascended from the pit, carrying everyone he could on his bionic arms. The rockets bounced him to a stable altitude from which he looked down upon the fallen and selected the spell: Vitrify. And just like that, with a swirl of magical winter dust, the slipping souls were captured in cold nitrogen crystals.

“I see now, there really is something more. More than mere happiness. It is posthuman happiness. It is liberation from mere human cravings, but it is more than what can be achieved through meditation or false utopias. I have pieced all your philosophies together, and now I see. Aristotle, you are right that we are here to do our function. Because we can’t do otherwise. Our function is our destiny, and our destiny is determined by this universe of which we are a part. From the Buddha I learned that I was wrong to mock happiness and yet, was right to emphasize something greater than it. From Mencius and the Rabbi, I reaffirmed that the path should not culminate, but should be a journey of growth.”

Then, with the sheer willpower surging through the transistors in his gadgets, he puffed up his chest and Summoned AGI to carve the world into an orgasmic puzzle for his modded ventral pallidum to chase. “Now the quest shall commence!” And in one bleeping, quickly slithering fractal explosion was the beginning of infinity.

Aesthetic Graphs

Two words: aesthetic graphs. Why hasn’t anyone made a business out of this? Build a platform where data visualization is art by default.

Financial trading brokerages already exploit the fact that their charts just look damn cool. This helps with alluring more customers.

Screen Shot 2018-03-11 at 12.44.43 PM.png

Software like infogram allows you to make airport-whitemom circles. They do not know the way.

The goal is to have a platform where one can easily make this:



In his 1983 book The Visual Display of Quantitative Information, Edward Tufte defines ‘graphical displays’ and principles for effective graphical display in the following passage: “Excellence in statistical graphics consists of complex ideas communicated with clarity, precision and efficiency. Graphical displays should:

  • show the data
  • induce the viewer to think about the substance rather than about methodology, graphic design, the technology of graphic production or something else

The graphs of Evangelion would fail. And rightly so. They are not actually meant to be useful. They are like Picasso’s of women: interpretations of a subject matter that may annoy the subject. Or in the venerable tradition of overanalyzing anime, we can say that they are a critique of scientific overcomplicating.



However, I disagree that the aesthetic is unimportant, or always misleading. One can be both beautiful and intelligent. If we could genetically engineer smart people to look like digital angels, so be it. Life is too short to stare at the offensively dull.

And some of the most acclaimed graphs where artistically sensible.

The true goal of data analysis is to reveal the hidden messages of nature – to surgically wield mathematical tools on numbers. When the message is extracted from her lips, the veil delicately removed, this is beauty.

But such an event deserves to occur in a temple that aspires to be worthy of her soul.