The Archive

Subject-conditioned causal neural surrogate for pediatric cerebral palsy rehabilitation using OpenSim-derived parameters.

Scaling laws study characterizes optimal learning rate and batch size evolution with model capacity for OpenEuroLLM pretraining.

Memristive-friendly reservoir computing replaces dense matrices with structured orthogonal operators for neuromorphic hardware deployment.

Formalizes KV cache eviction via probabilistic reasoning to improve LLM inference throughput with quality-preserving heuristics.

WikiSkill co-evolves agent skills with persistent knowledge bases, enabling systematic skill reuse across agent interactions.

Production GNN ranking system for social graph friend recommendation using multi-hash embeddings and temporal neighbor sampling at scale.

LeVJEPA: first video encoder using LeJEPA collapse-free objective without architectural asymmetries, achieving efficient self-supervised pretraining.

Speaker anonymization suppresses speaker-identifying attributes from speech while preserving linguistic content and quality. We propose repurposing XTTSv2, a multilingual voice cloning model trained on 27k hours of speech, for speaker anonymization without retraining. Our key…

Workflow for recovering 18th-century essay republication from fragmented text-reuse evidence across ECCO and newspapers.

LLMs can synthesize near-optimal algorithms for operations research problems (inventory control, queueing, assortment) given problem class descriptions and parameter ranges.

Neural operator architecture improvements for enforcing Dirichlet boundary conditions in PDE solution approximation.

Reuters report shows Meta's challenges replacing people with AI agents.

Sparse autoencoders extract interpretable physical concepts from neutrino foundation model; causal analysis reveals direction head underutilizes learned representations.

ICON decomposition isolates true concept contributions in neural networks via multivariate variance partitioning, detecting shortcut learning and spurious correlations.

Graph neural network trained via imitation learning for automated theorem prover proof construction in connection-tableau calculi.

ASR models reveal phonological traces of first language persist in pre-phonemic layers during second-language training.

LivingRAG adds writable experience store to Graph RAG, enabling reuse of verified reasoning across queries.

Neural-operator surrogate replaces expensive 3D electromagnetic forward solver in Bayesian inversion for geophysical surveys.

Control variables mitigate omitted variable bias in deep networks; shows demographic covariates encode into predictions via shortcut learning.

SAMpLE is an open-source SystemC-AMS framework integrating ML models as reusable Timed Dataflow components for virtual prototype simulation.

Loss-based active learning reduces annotation burden for abstractive summarization while achieving competitive model performance with fewer labels.

VINCENT framework for explaining drug synergy predictions by identifying molecular motif pairs jointly driving model outputs.

Weight decay pulses during neural network training grokking reveal stable causal ordering of generalization timing across tasks.

MPNO: physics-informed neural operator with spectral-radius constraints for stable autoregressive transient-dynamics prediction.

MeMark: watermark embedded in spiking neural network membrane state resists output-head replacement and checkpoint reuse attacks.

Parameterized complexity analysis of computing Lipschitz constants for input-convex neural networks via zonotope dual-norm maximization.

LION applies Clifford algebra to multimodal-attributed graph neural networks, improving representation and downstream task performance.

Ensemble CNN system predicts stroke risk from 11 features using seven supervised ML algorithms on medical imaging data.

FactoMap models disentangled representations with factor-space structure (domains, identifications, anisotropy) beyond Euclidean product assumptions.

RACE: Statistical framework for evaluating functional consistency of Transformer neurons at scale via forward-pass perturbation analysis.