The Archive

Self-supervised method for measuring L2 pronunciation deviation using native speech corpora and SVD without labeled data.

Radiology AI has made remarkable strides in detecting abnormalities across chest X-rays, pathology slides, and 2D scans. Yet one of the most clinically rich and... Radiology AI has made remarkable strides in detecting abnormalities across chest X-rays, pathology slides, and 2D scans. Yet one of the most clinically rich and data-dense modalities—the 3D computed tomography (CT) scan—remains largely underserved by modern vision language models (VLMs). Frontier general-purpose models perform poorly on volumetric imaging, and most open medical AI models lack the… Source

Provides geometric characterization of minimal-norm two-layer univariate ReLU classifiers; resolves optimal solutions under bias regularization.

Clifford-VAE learns hyperdimensional vector representations via Clifford algebras and geometric autoencoders on MNIST, FashionMNIST, CIFAR-10.

Local Geometric Mixing establishes Dobrushin contraction bounds for MCMC methods, with applications to Diffusion Path Monte Carlo.

Semantic similarity in frozen encoders is computed dynamically via shared forward pass, not statically from embedding geometry; BGE, E5, Llama 3 reach only 0.55–0.70 AUC on meaning identity.

World model framework generalizes robotic insertion across 90 geometrically diverse tasks via proprioceptive + wrist-camera fusion; single policy replaces task-specific variants.

Study finds activation memory (KV caches) superior to parametric updates for fact recall in few-shot LLM learning, challenging assumptions about parameter adaptation.

Modality imbalance in multimodal learning stems from asymmetric confidence gains during optimization, not modality fusion itself.

Task-induced Riemannian metrics improve Vision Transformer feature spaces by computing task-sensitive geometry via decoder Jacobian pullback.

Riemannian geometry approach to optimize block-diagonal structured matrices for efficient deep learning architectures.

Category theory framework using groupoids to exploit symmetries in partially observable environments for generalized RL policy learning.

MENO neural operator reduces memory footprint for PDE solving with resolution-independent scaling and arbitrary geometric domain support.

Pheno-GS computes geometry-aware optimal transport distances across single-cell patient cohorts at scale.

Study shows compile rate is unreliable metric for LLM code vulnerability repair; proposes change-aware evaluation across 350M–6.7B parameter models.

Stylometric classifier (Random Forest, ROC-AUC 0.87) detects ChatGPT-assisted student writing with 22% false positive rate.

Parametric convergence rates for entropic optimal transport in semi-discrete regime with subGaussian measures and no dimension dependence.

NVIDIA DLSS 5 introduces DLSS 3D-Guided Neural Rendering and granular controls that help game developers add lifelike lighting and material detail while... NVIDIA DLSS 5 introduces DLSS 3D-Guided Neural Rendering and granular controls that help game developers add lifelike lighting and material detail while preserving their artistic intent. We also look at updates to NVIDIA ACE, RTX Mega Geometry 2.0, and RTX Kit across character AI, high-density geometry, and rendering workflows. This post covers: DLSS 5 with 3D-guided… Source

Geometry-aware hyperbolic residual vector quantization preserves hierarchical structure in discrete token representations via non-Euclidean geometry.

Derives error bounds for MLE under Bradley-Terry-Luce preference model with unknown partworth parameters and non-uniform query selection.

CQ4OE benchmark systematically evaluates LLM-assisted ontology generation from competency questions with fine-grained provenance and structural adequacy metrics.

Groupoid-equivariant CNN theory for handling local symmetries on bounded/stratified domains.

GeoPair: training-free Transformer compression via cross-layer factorization preserving activation geometry.

ClusterFewshot improves LLM few-shot prompt optimization by combining semantic task structure with utility scoring, outperforming random and metric-based selection in DSPy.

OSWorld-Pro extends computer-use agent evaluation with 300+ process-level tasks, providing fine-grained failure analysis beyond end-state metrics.

Exposure accounting metric measures LLM reasoning over graph-grounded corpora by controlling for context leakage via copy-ceiling baseline.

ValueDiff proposes value-geometric KV cache eviction for LLMs with weak attention sinks, ranking tokens by value-vector L2 deviation from cache mean.

Statistical framework tests whether LLM-as-judge peer-review metrics capture substantive quality or merely surface-level linguistic polish.

GrapeSplat: feed-forward 3D Gaussian splatting from unposed images using voxel-aligned geometry without per-scene optimization.

Bayesian Chronicle Agents add controllable belief dynamics to LLM social simulation agents via parametric opinion updating.