The Archive

Conflict-preserving counterfactual interventions (Flip/Stay) measure whether VLMs reliably ground on user-requested modalities over parametric knowledge.

SceneBloom combines diffusion-based layout generation with geometric compilation for realistic, simulation-ready 3D indoor environments for embodied agent training.

LLM vulnerability auditors exhibit asymmetric deference to static-analysis evidence, accepting fabricated false assurances more readily than questioning false alarms.

LoRA's low-rank factorization is singular; real log canonical threshold analysis characterizes learning complexity independent of parameter dimension.

Token perceptron analysis in teacher–student setting: embedding geometry effects on generalization error in linear token prediction.

Neural network approximation of symmetric polyconvex elastic energies using ReLU input-convex networks with monotonicity constraints.

Framework decomposing LLM agent failure into model vs. harness contributions via Valid Completion Rate and Model-Harness Fit metrics.

Evaluates cost-aware abstention policies for LLM-based IT ticket routing, comparing classifiers on real 47,837-ticket dataset with asymmetric error costs.

Autoregressive factorization with parallel gradient boosting for nonparametric high-dimensional probability distribution estimation.

Autoregressive models can leak future tokens during training via architectural bugs, artificially inflating eval metrics while harming generation quality.

Mechanistic interpretability reveals causal subspaces controlling metaphor representation in LLMs across architectures.

State-space architecture for facial expression recognition in video using Mamba backbone with linear-cost spatiotemporal scanning.

Evaluation protocol for personalized federated learning that reveals generalization gaps by testing global, rare-class, and unseen-class performance.

Algorithmic drift formalizes feedback loops where observation errors accumulate asymmetrically over time in adaptive tutoring systems, amplifying inequities.

ASTEROID learns test-time exploration via asymmetric on-policy distillation, enabling agents to infer hidden task-relevant information from privileged experts.

Orbitol generates crystal structures via Crystalite diffusion Transformer with Wyckoff-orbit tokens, enabling direct control over space-group symmetry constraints.

InterModel Vigorish framework for interpretable ML using betting-based metrics to measure feature and component contributions.

MindView benchmark evaluates generative multi-view spatial intelligence across 1,500 samples with tasks spanning geometry and semantic reasoning.

Diagnostic metric framework analyzes value expression and organization in LLM-generated debate text.

CoRAL hybrid framework pairs diffusion planner with autoregressive executor for parallel multimodal reasoning, reducing latency via asymmetric orchestration.

Multi-horizon time-series forecasting study using LightGBM on Google Trends and COVID-19 data with entropy-based error metrics.

AsymLoop separates context encoding from iterative refinement in looped Transformers to improve effective depth without scaling parameters.

Physics-informed neural framework discovers extreme scenarios via data-driven dynamics inference with geometry-agnostic learned representations.

LocaLeJEPA extends JEPA self-supervised learning to non-Euclidean geometries via local entropy maximization under geometric priors.

HypTI applies hyperbolic geometry to multimodal feature fusion for dementia detection from clinical and neuroimaging data.

ACE-Compression framework evaluates which core LLM abilities survive compression beyond task performance metrics.

Theoretical framework connects scaling laws to geometric approximation, spectral analysis, and constrained optimization for resource allocation.

SLM-SS applies speech language models to multi-talker separation via token-based generation, optimizing for intelligibility over waveform metrics.

CrystalAR applies autoregressive modeling with space-group symmetry constraints to generate novel crystalline structures.

ProteinGeoBench reveals frontier agents retrieve protein metadata reliably but fail at geometric reasoning on 131 structural-biology tasks.