mike andrews

taking care of business (every day, it's all mine)

[mike andrews @algomagic]

Localizing quantum information

An adjusted von Neumann entropy is the minimal measure of localizable quantum information and uniquely characterizes Bell-state redundancy.

https://t.co/Jb27NCDsM2

[mike andrews @algomagic]

Emergence of Fibrations, Compression, and Symmetry Breaking in Artificial Neural Networks

Training induces stable graph-covering symmetries that enable compression to 17% of original size and, when selectively broken, stronger continual learning.

https://t.co/pOaxUPt7uH

[mike andrews @algomagic]

Solomonoff Induction and Singular Integrals

Bayesian evidence for any computable model can be embedded in Solomonoff induction, making singular learning coefficients appear in asymptotic code-length bounds.

https://t.co/HOGty6JehP

[mike andrews @algomagic]

Higher multipoles of the cow

Extends the spherical-cow approximation into a full multipole expansion for modeling bovine geometry, interactions, and symmetry-suppressed effects such as gravitational-wave spindown.

https://t.co/s9aFcU14nz

[mike andrews @algomagic]

Physical Predictions in Closed Quantum Gravity

Conditioning an enlarged Hilbert space on limited observations suppresses microscopic fluctuations, recovering semiclassical predictions, classical probabilities, and an arrow of time.

https://t.co/dJ9ZBaFhXK

[mike andrews @algomagic]

The Price of Remembering: A Calibrated Energy Law for Computation

A calibrated energy law charges operations, storage time, and data movement, yielding quadratic attention costs and joule floors for sorting and scrypt.

https://t.co/KWy8eOqH4u

[mike andrews @algomagic]

From Source Reconstruction to Predictive State Preservation: An Information-Theoretic Framework for AI-Native Communication

Predictive states are minimal targets; log-loss equals lost information, and full-future states support recursion.

https://t.co/CeeaXRz067

[mike andrews @algomagic]

Einstein's equation as a geometric encoding of correlations with quantum reference frames

Posits that spacetime geometry encodes local reference-frame correlations and derives Einstein’s equation under ideal-frame conditions.

https://t.co/WqP1pSgGx4

[mike andrews @algomagic]

The Emergent Symbolic Structure of Artificial Neural Networks

Finds that closed-form symbolic approximations preserve much of neural networks’ behavior and enable targeted interventions in LLMs.

https://t.co/hw5k9TzAgn

[mike andrews @algomagic]

Pragmatic Information, Computation, and the Efficient Market Hypothesis

Links extractable meaning to a receiver’s computational power and reframes market efficiency as a participant-specific computational limitation.

https://t.co/4ZAbO0V3Aj

[mike andrews @algomagic]

A statistical model for quantum spin and photon number states

A binary event-network model recovers quantum spin statistics with vanishing finite-size corrections, extends to photon states, and suggests a tabletop test.

https://t.co/RTHeKK3czV

[mike andrews @algomagic]

Attention as Conditioning: What Classical Learning Theory Predicts About Linear Transformers

State updates match classical conditioning rules, predicting cue competition, capacity scaling, and no spontaneous recovery.

https://t.co/5txnX65hcR

[mike andrews @algomagic]

A Formal Limitation on Learning Human Language From Textual Corpora

Information-theoretic bounds show that no text representation can recover aspects of speaker meaning that only extralinguistic context resolves.

https://t.co/vtlRXKh9cW

[mike andrews @algomagic]

Beyond the Rosetta Stone: Unification Forces in Generalization Dynamics

Synthetic multilingual experiments find that facts transfer across languages only when models form unified rather than language-specific representations.

https://t.co/YUMdjmOWQs

[mike andrews @algomagic]

A not-too-simple solution to Goodman's New Riddle of induction

A proposed solution to Goodman's riddle jointly characterizes model complexity and direct measurement within a scientific view of philosophy.

https://t.co/qsBdo8D6fm

[mike andrews @algomagic]

When the Canonical Completion Is Wrong: Formalizing and Measuring the Jump in Large Language Models

Certified tests find four frontier LLMs always abandon an excluded canonical completion, so this step is not the bottleneck.

https://t.co/eFei59JkmH

[mike andrews @algomagic]

How quantum is quantum geometry?

A classical spinning particle reproduces Berry-curvature and quantum-metric effects, including position spread, orbital moment, induced mass, and Drude weight.

https://t.co/jCrccpf4u1

[mike andrews @algomagic]

A Note on the Measure of Vector and Pythagorean Theorem

Starting from a single invariant vector measure, the note derives squared norm, Euclidean notions, the Pythagorean theorem, and the quantum Born rule.

https://t.co/aXfPAhYKtE

[mike andrews @algomagic]

Theory as data compression

A minimum-description-length framework compares economic models by data compression, capturing flexibility beyond parameter counts and improving model recovery.

https://t.co/ThSiaxdu5C

[mike andrews @algomagic]

Thermodynamic cost of inference and learning in physical neural networks

A Hamiltonian model yields zero quasistatic work for inference, finite-speed transport bounds, and an irreducible thermal cost for writing learned parameters.

https://t.co/6YDTe0PgTI