Flat minima are illusory; generalization is driven by weakness, a reparameterization-invariant measure of compatible completions that predicts performance better than sharpness on MNIST and Fashion-MNIST.
hub Canonical reference
Physical Review 106(4):620–630
Canonical reference. 80% of citing Pith papers cite this work as background.
15
Pith papers citing it
9,268
external citations · Crossref
Background
80% of classified citations
hub tools
citation-role summary
background 4
method 1
citation-polarity summary
representative citing papers
Entropic Autoencoders mitigate posterior collapse by implicitly defining priors via entropy in a free-energy-minimizing encoder ensemble, yielding multimodal latent distributions that preserve data structure on reaction-diffusion, MNIST, and CelebA.
The entropy of the sum of independent ternary random variables is maximized when the first n-1 variables are uniform on {0,2} and the nth follows a specific distribution defined by binomial entropies.
Two calls per example identify the first two moments of latent correctness probability, enabling exact bounds on the vote-accuracy curve for any majority-vote budget under conditional i.i.d. assumptions.
STAR-Teaming uses a Strategy-Response Multiplex Network inside a multi-agent framework to organize attack strategies into semantic communities, delivering higher attack success rates on LLMs at lower computational cost than prior methods.
FQNM realizes the conservative operator as an antisymmetric integer transfer rule, proving exact conservation, monotonicity, TVD, L1 stability and convergence to the entropy solution for scalar laws with monotone flux splitting while collapsing distinct flux formulations to identical dynamics.
Algorithms sample maximum-entropy distributions over citizen assembly panels, yielding better intersectional diversity and higher probability of satisfying unseen representation constraints than standard methods.
Functional renormalization group applied to nearly continuous spectra yields a scale-dependent canonical dimension that undergoes a dimensional phase transition at signal-to-noise ratios below the BBP threshold, correlating with symmetry breaking and eigenvector deviations.
Robots detect underspecified reward features via demonstration variation and query targeted natural language explanations to improve reward recovery from imperfect demos.
A tunable microscopic model of network liquids with a liquid-liquid phase transition, analyzed via RFOT theory, predicts nanonucleation near the glass transition and links thermodynamic and kinetic anomalies when matched to water-like conditions.
Information defined as maximum-caliber deviation derives IIT 3.0 cause-effect repertoires from constrained entropy maximization and equates to prediction error under CLT and LDT.
The quiet-Sun temperature ratio R≈2.4 equals the KL-divergence difference between a κ=2.5 distribution and its EUV and radio Maxwellian projections, satisfying ΔD_KL = (3/2)[R0 − ln R0 − 1] = (3/2) d_IS(T_eff, T_core).
The no-barber principle prohibits selection rules in the inaccessible game that appeal to external adjudicators, favoring the symmetric monoidal category NCFinProb over the cartesian FinProb as its internal language due to the absence of canonical copying maps.
Lecture notes on quantum thermodynamics showing emergence of thermodynamic laws from quantum theory via Markovian master equations for small systems.
citing papers explorer
-
Are Flat Minima an Illusion?
Flat minima are illusory; generalization is driven by weakness, a reparameterization-invariant measure of compatible completions that predicts performance better than sharpness on MNIST and Fashion-MNIST.
-
Entropic Auto-Encoding via Implicit Free-Energy Minimization
Entropic Autoencoders mitigate posterior collapse by implicitly defining priors via entropy in a free-energy-minimizing encoder ensemble, yielding multimodal latent distributions that preserve data structure on reaction-diffusion, MNIST, and CelebA.
-
Maximum Entropy of Sums of Independent Ternary Random Variables
The entropy of the sum of independent ternary random variables is maximized when the first n-1 variables are uniform on {0,2} and the nth follows a specific distribution defined by binomial entropies.
-
Two Calls, Two Moments, and the Vote-Accuracy Curve of Repeated LLM Inference
Two calls per example identify the first two moments of latent correctness probability, enabling exact bounds on the vote-accuracy curve for any majority-vote budget under conditional i.i.d. assumptions.
-
STAR-Teaming: A Strategy-Response Multiplex Network Approach to Automated LLM Red Teaming
STAR-Teaming uses a Strategy-Response Multiplex Network inside a multi-agent framework to organize attack strategies into semantic communities, delivering higher attack success rates on LLMs at lower computational cost than prior methods.
-
Continuum dynamics from quantised interaction rules
FQNM realizes the conservative operator as an antisymmetric integer transfer rule, proving exact conservation, monotonicity, TVD, L1 stability and convergence to the entropy solution for scalar laws with monotone flux splitting while collapsing distinct flux formulations to identical dynamics.
-
Maximally Random Sortition
Algorithms sample maximum-entropy distributions over citizen assembly panels, yielding better intersectional diversity and higher probability of satisfying unseen representation constraints than standard methods.
-
Functional Renormalization for Signal Detection: Dimensional Analysis and Dimensional Phase Transition for Nearly Continuous Spectra Effective Field Theory
Functional renormalization group applied to nearly continuous spectra yields a scale-dependent canonical dimension that undergoes a dimensional phase transition at signal-to-noise ratios below the BBP threshold, correlating with symmetry breaking and eigenvector deviations.
-
Robots That Know What to Ask: Recovering Misaligned Rewards through Targeted Explanations
Robots detect underspecified reward features via demonstration variation and query targeted natural language explanations to improve reward recovery from imperfect demos.
-
Polyamorphism in Glassy Network Materials
A tunable microscopic model of network liquids with a liquid-liquid phase transition, analyzed via RFOT theory, predicts nanonucleation near the glass transition and links thermodynamic and kinetic anomalies when matched to water-like conditions.
-
Information as Maximum-Caliber Deviation: A bridge between Integrated Information Theory and the Free Energy Principle
Information defined as maximum-caliber deviation derives IIT 3.0 cause-effect repertoires from constrained entropy maximization and equates to prediction error under CLT and LDT.
-
Diagnostic Disagreement as an Information-Projection Divergence: An Information-Theoretic Reading of the Quiet-Sun Temperature Ratio
The quiet-Sun temperature ratio R≈2.4 equals the KL-divergence difference between a κ=2.5 distribution and its EUV and radio Maxwellian projections, satisfying ΔD_KL = (3/2)[R0 − ln R0 − 1] = (3/2) d_IS(T_eff, T_core).
-
The No Barber Principle: Towards Formalised Selection in the Inaccessible Game
The no-barber principle prohibits selection rules in the inaccessible game that appeal to external adjudicators, favoring the symmetric monoidal category NCFinProb over the cartesian FinProb as its internal language due to the absence of canonical copying maps.
-
Quantum Thermodynamics
Lecture notes on quantum thermodynamics showing emergence of thermodynamic laws from quantum theory via Markovian master equations for small systems.
- Emergence of Complex Web Structures