Author: Gregor J. Rothfuss

ML Data validation

anecdotal evidence from some teams suggest a mental shift towards a data-centric view of ML, where the schema is not solely used for data validation but also provides a way to document new features that are used in the pipeline and thus disseminate information across the members of the team.

The Secret Of Our Success

Our mouths are the size of the squirrel monkey’s, a species that weighs less than 1.3 kg. Chimpanzees can open their mouths 2x as wide as we can and hold substantial amounts of food compressed between their lips and large teeth. We also have puny jaw muscles that reach up only to just below our ears. Other primates’ jaw muscles stretch to the top of their heads, where they sometimes even latch onto a central bony ridge. Our stomachs are small, having only a third of the surface area that we’d expect for a primate of our size, and our colons are too short, being only 60% of their expected mass. Compared to other animals, we have such atrophied digestive tracts that we shouldn’t be able to live. What saves us? All of our food processing techniques, especially cooking, but also chopping, rinsing, boiling, and soaking. We’ve done much of the work of digestion before food even enters our mouths.

Continuous Quantum Leaps

By making a kind of high-speed movie of a quantum leap, the work reveals that the process is as gradual as the melting of a snowman in the sun. “If we can measure a quantum jump fast and efficiently enough, it is actually a continuous process.” But there’s more. With their high-speed monitoring system, the researchers could spot when a quantum jump was about to appear, “catch” it halfway through, and reverse it, sending the system back to the state in which it started. What seemed to be unavoidable randomness in the physical world is now shown to be amenable to control. We can take charge of the quantum.

Fermentation Lab

While playing around with MIT’s as-high-tech-as-possible controlled environment mini labs and other prototypes, she also created more stable, robust and practical models with parts anyone can find on Amazon. They’re meant more for restaurant kitchens like at the Standard East Village where she’s already making 2 that fit the needs of their food-and-beverage outlets. “There’s something fun and punk rock about building these things, not out of garbage but out of discarded things”. She also put in a commercial dough proofer that holds koji ferments—i.e., ingredients transformed by a bacteria typically used to make miso, soy sauce and sake—at just the right temperature and humidity level.

Solving Quake

We used a tournament-style evaluation to demonstrate that an agent can achieve human-level performance in a 3D multiplayer first-person video game, Quake III Arena in Capture the Flag mode, using only pixels and game points scored as input. We used a 2-tier optimization process in which a population of independent RL agents are trained concurrently from 1000s of parallel matches on randomly generated environments. Each agent learns its own internal reward signal and rich representation of the world. These results indicate the great potential of multiagent reinforcement learning for artificial intelligence research.

Foursquare Personas

An example on the site illustrates just how much Foursquare can know about the habits of an individual user. For a man aged 45-49, in the Dallas-Fort Worth area, Foursquare can use data about which shops, bars, and other buildings the man visits regularly to assign “personas,” which then can be used by advertisers to target the user more effectively. Because Foursquare can tell the man visited AT&T Stadium between 19:00 and 22:30, and he stopped by Harry’s Sports Bar last week, Foursquare has him pegged as a “Sports Lover.” Similarly, the man’s visit to Pepper’s Grill between 17:00 and 18:30—his second trip to the establishment this week, it notes—confirms the “Casual Diner” persona.