There was a table set out under a tree in front of the house, and the March Hare and the Hatter were having tea at it: a Dormouse was sitting between them, fast asleep, and the other two were using it as a cushion, resting their elbows on it, and talking over its head. “Very uncomfortable for the Dormouse,” thought Alice; “only, as it’s asleep, I suppose it doesn’t mind.”

The table was a large one, but the three were all crowded together at one corner of it: “No room! No room!” they cried out when they saw Alice coming. “There’s plenty of room!” said Alice indignantly, and she sat down in a large armchair at one end of the table.

“Have some wine,” the March Hare said in an encouraging tone.

Alice looked around the table, but there was nothing on it but tea. “I don’t see any wine,” she remarked.

“There isn’t any,” said the March Hare.

“Then it wasn’t very civil of you to offer it,” said Alice angrily.

Making no effort to reply to this, the March Hare and the Hatter started discussing something. Alice tried to listen in, but she couldn’t, which was very shocking because she could listen to everything else, except just this one thing.

After a while, the Hatter asked Alice, “Are you familiar with AI?”.

Alice said, “Why, yes! I am very interested. I did a minor in AI with my undergraduate.”

As she finished, all three of them — the Hatter, the Hare, and the Dormouse — started murmuring to each other, which made Alice slightly annoyed because again, she could not hear it. 

The Dormouse was sleeping and murmuring at the same time. She spoke up first, “That’ll do. So, the Hatter — lead journalist at the most popular AI safety news channel, Cross-Validated News and Self-Fulfilling Views — is interviewing the March Hare, a wonderland-renowned scientist for his works on CDPEBW-Net, a novel architecture that achieved state-of-the-art on the IEWD Dataset. You have to take up my role since I have to sleep now.”

Alice shrugged, “But what is your role, Ms. Dormouse?”. Somehow, Alice just knew what to call everyone.

“Why, it is to keep an eye on the Mad Hatter and ensure that he doesn’t take off his hat.”

“But what happens if he takes off his hat?”, asked Alice.

“Don’t you know about the prize the queen has put up on that impious little white rabbit?”, asked the Dormouse back.

“No. What prize?”, asked Alice.

“The queen had asked the white rabbit to create an AI model to convert the temperature of her castle from Celcius to Fahrenheit. And the rabbit spent one month carefully fine-tuning hyperparameters of a big, deep network. But for some reason, it always predicted the temperature to be lesser than it actually was. The queen got really annoyed at this, and shouted “Off with his head” at the white rabbit. The rabbit ran away into the woods, so the queen put up a grand prize of 700,000 Shannons. And everyone knows white rabbits have a tendency to go and hide inside large black hats.”

“Oh. I feel bad for the poor white rabbit.”, replied Alice.

“Anyway, let the interview begin!”, squeaked the Dormouse, and soon, she was sleeping again. The Hatter started the interview, and Alice made a mental note to stop him if he ever tried to take his hat off.

“Mr. Hare,”, began the Hatter, “to begin with, what are your thoughts on this recent tragedy at the Queen’s castle?”

“Well,”, replied the Hare, “first of all, I do not understand why the white rabbit would create an AI model for this problem. You see, it can be solved by simply multiplying by 1.8 and adding 32. Maybe someone was pressuring him into it, or maybe he wanted to flex his skill. Either way, when a simple portkey works, one shouldn’t try to pull off an elaborate, deadly wizarding tournament.”

Alice had to agree — this problem did not warrant an AI solution.

“But even if he did create a deep model, in theory, it should work, right?”, asked the Hatter.

“Yes.”, said the Hare. “There exists a Universal Function Approximation Theorem that states that neural networks can, under mild assumptions and sufficient width, get arbitrarily good at approximating any continuous function. A proof of the particular case of single-depth networks is very popular.”

“So then, why do you think the network did not do well?”, asked the Hatter.

This was one question that had been bugging Alice for quite some time.

The Hare responded, “There could be multiple reasons behind this. Since the prediction was always a little lesser than the actual, I’m guessing the data distribution changed from training to evaluation. For this particular case, maybe the training data did not account for the temperature rise due to large neural networks. In this case, since the network is huge, the effect of just evaluating the model could be enough to change the temperature.”

“Let’s move on,”, said the Hatter. “Can you tell us more about your work?”

“Gladly. I work on something called Graph Networks.”, replied the Hare.

“What made you interested in this field?”, asked the Hatter. He once tried to remove his hat to scratch his head, but Alice was aware and stopped him right away.

“For one, everything we see around us seems to be a graph. From the molecules that make everything up to the structure of our brain to objects in physics, to images and videos, to maps of the world, to hierarchies, to communities, to the search spaces of games, to language structures, to the computation inside a neural network — all of them can be represented using graphs.”

“And secondly,” continued the Hare, “graph networks are an abstraction that prècise almost all successful learning algorithms on structured data, such as trees, CNNs, and transformers. This, I believe, stems from the fact that they model the data structures themselves very well. Finally, I’m also very interested in learning on even more abstract algebraic structures for this very same reason.”

And just as the Hatter got up to end the interview and shake the Hare’s hand, he toppled over (by mistake) and his hat dislodged from his head. At the same time, Alice jumped out of her chair to catch it and put it back on his head. Unfortunately, she slipped and fell face-first on the table. It did not hurt, but the tea spilled all over the place, and the commotion woke the Dormouse. She yelled “Run”, and everyone started running. Since Alice was never tired in the wonderland, she kept running for hours and hours, or so it felt.

Alice found herself at last in a beautiful garden, among bright flowerbeds and cool fountains.