The Lottery
The morning of October 27th was cloudy and overcast, with the cold of a mid-autumn day; the leaves of the trees showed hints of orange, and the dutifully maintained grass was richly green. The students of Carnegie Mellon began to gather on the Cut around ten o’clock; the whole lottery took only about two hours, so it could begin at nine o’clock in the morning and still be through in time to allow the students to get to Revolution Noodles prior to the crowds that would inevitably gather for lunch.
The first-years assembled first, naturally. Most of the students had already stuffed their backpacks full of textbooks, while a few stragglers had to run back to their dorms, having forgotten them or, perhaps, having been initially unwilling to participate but now feeling left out.
Some of the doctoral candidates began to gather, surveying the freshmen, speaking of research and funding, overfitting and tradeoffs, all alien language to these bright-eyed youths. The faculty came shortly thereafter, wearing whatever chalk-stained trouser or misbuttoned cardigan had struck their fancy in the closet that morning. They greeted one another, exchanging bits of gossip before joining their research groups.
The Pruning, as it was termed, was conducted by Professor Summers, a round-faced, jovial man who ran the machine learning lab. People were sorry for him because he had had no publications at NeurIPS that year and so, when he arrived, carrying a sleek laptop, there was a ripple of conversation throughout the field, not all of it kind. He waved and called, “Little late today, folks. The cluster was slow. You know how it is.” He took a seat on the Fence, his laptop perched precariously on his leg. “Well, now, guess we better get started, get this over with, so we can get back to work.”
A hush fell on the crowd as he cleared his throat and opened the terminal. The students had all heard about this moment as part of a Core module and only half listened to him as he detailed the necessity of “pruning the weights of our neural network, so that we may grow and thrive more efficiently.” And with little aplomb, the script finished execution, midway through a sentence. A name flashed in stark, monospaced font. It was Tessie, a sophomore in Statistics and Machine Learning—though she took any and every opportunity to explain that she was ‘planning to maybe, kind of, transfer to SCS, due to a change of heart since arriving here, not because I was using StatML as an easy route to get admitted to CMU!’ For a moment, she did not move. Then she shouted, “There’s a bug! This is a biased sampler! You didn’t use a proper pseudorandom algorithm! This is selection bias!”
“Be a good sport, Tessie,” someone called out from the crowd.
“I think we ought to start over,” Tessie said, now speaking quietly. “I tell you it wasn’t fair.” As she spoke, a thick heat transfer textbook sailed past her head. Clearly not all CMU students were capable of basic athleticism.
Someone yelled, “Come on, come on, everyone. The Pruning won’t wait.” A balled-up handout hit the side of her head.
“It isn’t fair, it isn’t right,” Tessie screamed, and then they were upon her.