DeepField

LAXCORP RESEARCH · DEEPFIELD PROGRAM

Foundations

The science DeepField stands on is public. That is not a confession of unoriginality. It is how instruments have always been built. The craft of grinding lenses was public for three centuries before two lenses, composed, became a telescope. Every part was known. The instrument was new, and it changed what could be seen.

Each line below was proven alone: in papers, on frozen data, under conditions its authors could control. Composing them into one instrument that runs continuously against the live world is a different order of problem, unsolved, and it is the substance of the lab’s own work. That work is deliberately not published. What belongs to the world is the lineage. This is that reading.

  • Judea Pearlcausal inference

    The Book of Why. The case that correlation is not enough, and that reasoning about cause is both necessary and formally possible.

  • Bernhard Schölkopfcausal representation learning

    Toward Causal Representation Learning. The bridge between modern machine learning and causal structure.

  • Thomas Schreibertransfer entropy

    Measuring Information Transfer. A way to quantify the direction in which information flows between variables.

  • Philip Tetlocksuperforecasting

    Superforecasting, and the Good Judgment Project. Evidence that disciplined, scored forecasting outperforms expert intuition, and the culture of accountability we hold ourselves to.

  • DARPA SCOREmeasuring prediction

    Systematizing Confidence in Open Research and Evidence. How to measure, rather than assert, whether a method actually predicts.

  • Charles Goodhartthe limits of metrics

    Goodhart’s law. Why a method should report calibration bucket by bucket, not a single flattering number.

  • Schoenegger and colleagueslanguage models and forecasting

    Recent evidence on where language models sharpen forecasts, and where they do not.

  • Vladimir Vovkconformal prediction

    Algorithmic Learning in a Random World, with Gammerman and Shafer. How a prediction can carry a guaranteed error bar instead of false precision.

  • Ryan Adams and David MacKaychange-point detection

    Bayesian Online Changepoint Detection. How to separate a genuine change from noise, as it happens.

  • Marten Scheffercritical slowing down

    Early-warning signals for critical transitions. The finding that complex systems often signal a coming shift before it arrives: variance rises, recovery slows.

  • EWSNetlearned early warning

    Royal Society Open Science. That early-warning signatures can be learned by a model, not only derived by hand.

  • The Santa Fe Institutecomplexity economics

    The tradition of treating an economy as a complex adaptive system rather than a machine at rest.

  • Yann LeCun, Yoshua Bengio, and Geoffrey Hintondeep learning

    Deep Learning, Nature 2015. Learned representation: networks that find structure in data too subtle, and too fast-moving, to write down by hand.

  • James Hamiltonregime-switching models

    The econometrics of regime change. Formal machinery for detecting when a system has stopped behaving like its own past.

  • The probabilistic programming traditioncomputational Bayesian inference

    Uncertainty carried end to end: models that hold full distributions over outcomes instead of a single number pretending to be sure.

  • Daniel Kahneman and Amos Tverskybehavioral economics

    Prospect Theory, and Thinking, Fast and Slow. Why human reaction to events is biased in systematic, and therefore modelable, ways.

  • George Sorosreflexivity

    The Alchemy of Finance. That observing and speaking about a market changes it. A system that publishes must account for its own influence.

  • Nassim Nicholas Talebthe unpredictable

    The Black Swan. The discipline of respecting what cannot be forecast, and refusing to pretend otherwise.

The methods are public. The question is whether they compose.