Selected work

Representative Papers

Early Signatures of Memorization in Diffusion Models via Basin Geometry and Cyclic Denoising

Oct 08, 2026

This study addresses the challenge of detecting memorization in diffusion models prior to generation by proposing an early auditing method grounded in energy landscape geometry. The research reveals that memorization manifests before output generation, introducing the concept of "latent memorization." By identifying degenerate attractors, it refines the conventional logic that equates memorization solely with basin residence time. Through score divergence, basin volume analysis, and cyclic denoising techniques to probe local basins surrounding training samples, the method enables early identification and isolation of memorized content. Supported by theoretical proofs and multi-dataset experiments, this approach successfully recovers training images on CelebA without explicit replication and achieves an AUC of 0.944 on Stable Diffusion, significantly advancing the detection window for model memorization.

0 citationsRead paper

Dual Computational Horizons: Incompleteness and Unpredictability in Intelligent Systems

Dec 18, 2025

This paper identifies two fundamental computational limitations inherent in intelligent systems: Gödelian incompleteness of formal systems—which constrains the deductive power of consistent reasoning—and finite-precision unpredictability in dynamical systems—which bounds long-term predictive accuracy. Methodologically, it integrates formal logic, computability theory, dynamical systems analysis, and algorithmic information theory to establish a unified framework. The key contribution is the first rigorous characterization of the intrinsic trade-off between inferential completeness and predictive stability, culminating in a proof that no algorithmic agent can *decidably* compute its own maximal prediction horizon. This result establishes an insurmountable theoretical limit on AI interpretability, autonomous reasoning, and long-horizon planning, thereby defining the intrinsic boundary of self-reflective capability for computational agents.

0 citationsRead paper

Recent publications

Latest Papers

Early Signatures of Memorization in Diffusion Models via Basin Geometry and Cyclic Denoising

Oct 08, 2026

This study addresses the challenge of detecting memorization in diffusion models prior to generation by proposing an early auditing method grounded in energy landscape geometry. The research reveals that memorization manifests before output generation, introducing the concept of "latent memorization." By identifying degenerate attractors, it refines the conventional logic that equates memorization solely with basin residence time. Through score divergence, basin volume analysis, and cyclic denoising techniques to probe local basins surrounding training samples, the method enables early identification and isolation of memorized content. Supported by theoretical proofs and multi-dataset experiments, this approach successfully recovers training images on CelebA without explicit replication and achieves an AUC of 0.944 on Stable Diffusion, significantly advancing the detection window for model memorization.

0 citationsRead paper

Dual Computational Horizons: Incompleteness and Unpredictability in Intelligent Systems

Dec 18, 2025

This paper identifies two fundamental computational limitations inherent in intelligent systems: Gödelian incompleteness of formal systems—which constrains the deductive power of consistent reasoning—and finite-precision unpredictability in dynamical systems—which bounds long-term predictive accuracy. Methodologically, it integrates formal logic, computability theory, dynamical systems analysis, and algorithmic information theory to establish a unified framework. The key contribution is the first rigorous characterization of the intrinsic trade-off between inferential completeness and predictive stability, culminating in a proof that no algorithmic agent can *decidably* compute its own maximal prediction horizon. This result establishes an insurmountable theoretical limit on AI interpretability, autonomous reasoning, and long-horizon planning, thereby defining the intrinsic boundary of self-reflective capability for computational agents.

0 citationsRead paper