Theory & Signal / Research

From first principlesto new formsof intelligence.

Theory & Signal explores the foundations and frontiers of machine intelligence, from learning and reasoning to perception, generative systems, safety, and scientific discovery.

Our purpose

Intelligence is not a single problem. Neither is our research.

We investigate machine intelligence across multiple levels, from the principles that govern learning to the systems that interpret language, understand visual information, generate new representations, and support scientific discovery. Progress at one level constrains and informs the others — so we study them together.

L1
PrinciplesHow learning, generalization, and reasoning work.
L2
SystemsAgents, language, vision, and generative models.
L3
ApplicationsScientific discovery and reliable deployment.

Research domains

Research across the foundations and frontiers of AI.

Seven connected areas of inquiry, each with its own questions and methods — and many shared ones.

The research directory
  1. 01

    Artificial Intelligence

    Reasoning, planning, autonomous agents, and systems capable of adapting to increasingly complex environments.

  2. 02

    Machine Learning

    Learning algorithms, model generalization, representation learning, optimization, and efficient intelligence.

  3. 03

    Generative AI

    Foundation models and generative systems that create and reason across language, imagery, and other modalities.

  4. 04

    Natural Language Processing

    The structures of language, computational understanding, multilingual intelligence, and grounded reasoning.

  5. 05

    Computer Vision

    Visual representation, perception, spatial understanding, and multimodal interpretation.

  6. 06

    AI Safety

    Interpretability, evaluation, robustness, alignment, uncertainty, and reliable model behavior.

  7. 07

    Scientific Discovery

    Computational approaches that help explore problems in mathematics, physics, biology, chemistry, and related sciences.

At a glance

The shape of the program.

01

Research domains

From learning theory to scientific discovery.

02

Foundations

to applications

Fundamental questions pursued alongside their practical consequences.

03

Theory

and experiment

Ideas held to the standard of measurable, reproducible evidence.

04

Safety

throughout

Reliability treated as part of every domain, not a separate one.

07 — Scientific Discovery

New intelligence. New ways to understand the world.

Advances in machine learning open new possibilities for computational research, from understanding complex systems to exploring scientific questions that traditional methods struggle to address.

  • Scientific machine learning
  • Computational biology
  • Physics-informed learning
  • Chemistry and materials modeling
  • Simulation and surrogate models
  • Scientific reasoning
Fig. 07Schematic
ω₁ω₂
Fig. 07Schematic. Streamlines of a two-dimensional vector field with two sources of rotation — the kind of structure a learned surrogate must reproduce.

Principles

How we approach research.

Four commitments that shape which questions we take on and how we answer them.

  1. 01

    First principles

    Start with clear questions and rigorous foundations.

    We prefer understanding why a method works to knowing that it does. Sharp questions, stated precisely, are the most valuable thing a research program can produce.

  2. 02

    Experimentation

    Test ideas through measurable experiments and reproducible methods.

    Claims should be falsifiable and results repeatable: controlled comparisons, multiple seeds, released configurations, and honest reporting of what did not work.

  3. 03

    Interdisciplinary thinking

    Learn from connections between computational fields and other scientific disciplines.

    Many ideas in machine learning have roots in statistics, physics, neuroscience, and linguistics. We keep those connections open in both directions.

  4. 04

    Responsible progress

    Consider reliability, safety, transparency, and broader consequences as part of technical advancement.

    Understanding how a system can fail is part of understanding the system. We treat evaluation and safety as research questions in every domain, not as a final checklist.

Meaningful progress is rarely made in isolation.

We welcome conversations with researchers, engineers, and organizations exploring related questions in artificial intelligence and scientific computing.