The Archive

ECG biometrics using Siamese ResNet evaluated across exercise stress and temporal variation for authentication and pretext learning.

Nonparametric tangent vector field regression on Riemannian manifolds via parallel transport and kernel methods.

Geometric Mean Pooling operator combining sign and magnitude information for hierarchical neural network feature composition without learnable parameters.

Theoretical analysis of optimization geometry across equivalent Brownian RKHS coordinate representations.

HATS-en benchmark shows WER poorly correlates with human judgment; BERTScore variants better track semantic fidelity in ASR.

Cross-policy analysis of 15k LIBERO rollouts reveals vision-language-action policies achieve similar end-effector geometry when succeeding.

Theoretical extension of JEPAs beyond Gaussian latent spaces to Riemannian manifolds with conditions for representation identifiability.

Riemannian Neural Hamiltonian Flows combine geodesic geometry with invertible generative modeling for improved interpretability in normalizing flows.

Pretrained neural PDE surrogates reduce CFD data needs 3.25× when geometry changes; quantifies distribution shift effects on pretraining gains.

GeoAAC proposes geometry-based adaptive action chunking for VLA policies that adjusts horizon by prediction reliability.

Deep Noir automatically discovers LLM activation steering parameters via Logit Lens and causal attribution, improving task performance 16–42 percentage points.

Theoretical analysis of TAP approximation and posterior geometry in spherical linear models under Marchenko-Pastur conditions.

NeuSOGA3D combines neural implicit methods with symbolic geometric reasoning for interpretable 3D reconstruction from point clouds.

Matrix generalization of Harvey-Lawson cone proves exact special Lagrangian manifold structure in deep learning geometry.

Study of self-replicating neural cellular automata with mutation-driven diversity metrics to quantify emergent phenotypic and genotypic ecosystem organization.

Benchmark: six frontier models play log(N)-Questions game; Claude Opus 5 underperforms (28/68 wins), revealing communication asymmetry gaps.

Heterogeneous radiologist reporting practices confound chest X-ray report generation model evaluation via reference concordance metrics.

Study quantifies minimum human-to-synthetic data ratio needed to prevent model collapse in LLM training using Fisher-Rao information geometry.

Interpretable multi-instance learning predicts AML mutations from flow cytometry hours faster than standard molecular testing using decision-tree classifiers.

Geometric theory of decision-boundary structures in Markov decision processes for understanding policy complexity and oracle-query efficiency.

TTM-Bench standardizes benchmarking protocol for text-to-music systems with unified metrics across architectures and conditioning modes.

On-the-fly homography calibration for multi-camera tracking eliminates offline 3D site calibration via geometric refinement.

Spatially adaptive noise injection in diffusion samplers applies stochastic correction selectively based on image geometry.

HiLNO applies hierarchical latent representations with multi-scale supervision to improve neural operator learning for PDEs on general geometries.

SWE-bench leaderboard audit: top coding agents converge (0.935 nesting), rendering score ordering unreliable; proposes new metrics.

LLMs form linear belief-state geometry for HMM hidden states during in-context learning; probed across 6 models on 40 HMMs.

PosterVisor framework for scientific poster generation using persistent semantic-geometric contracts instead of transient prompts.

Goal-oriented probabilistic forecasting using Pinball Loss aligns 5G PRB allocation models with asymmetric operational costs.

Their startup, Artificial Intelligence Underwriting Company (AIUC) has raised $40 million in a Series A lound led by Ribbit Capital, with participation from First Harmonic.

AI agent workflow using GPT-6 Astra optimizes 2D CFET thermal design, reducing peak temperature 1.67 K via automated geometry and heat-removal proposals.