Search
15 results for "training"
- ai · arxiv/cs.LG · 4 min
Synthetic Computers Enable Agent Training at Scale
Researchers create realistic digital workspaces to train AI agents on long-horizon productivity tasks, scaling from thousands to potentially billions of simulated user environments.
May 3, 2026 Read → - ai · arxiv/cs.LG · 8 min
Mixed Precision Training Stabilizes Neural ODEs
Researchers demonstrate a framework that reduces memory use by 50% and speeds up neural ODE training 2x by carefully mixing low and high precision arithmetic.
May 3, 2026 Read → - ai · arxiv/cs.LG · 8 min
Model Architecture Controls Whether Errors Stay Hidden
Transformer design determines if internal decision signals remain observable after training, independent of output confidence metrics.
Apr 29, 2026 Read → - ai · arxiv/cs.AI · 8 min
Poisoned Pretraining: Hidden Attacks Embedded in LLM Training Data
Researchers demonstrate how adversaries can plant dormant malicious logic in large language models by seeding poisoned content across obscure websites, evading detection until triggered.
Apr 27, 2026 Read → - ai · arxiv/cs.AI · 5 min
Fast Entropic Approximations cut entropy computation by 37x
Horenko et al. propose non-singular rational approximations of Shannon entropy and KL divergence that preserve mathematical properties while reducing computation cost and improving ML model training.
Apr 27, 2026 Read → - ai · arxiv/cs.AI · 4 min
Cross-Entropy Loss Drives Neural Probe Performance, Not Architecture
Pre-registered study shows cross-entropy training inflates logit norms 15x, accounting for most K-way energy probe gains over softmax baselines.
Apr 24, 2026 Read → - ai · arxiv/cs.AI · 8 min
Supervised Learning Has Built-In Geometric Blindness
Mathematical proof shows empirical risk minimization must preserve sensitivity to label-correlated but test-irrelevant features—a structural constraint, not a training bug.
Apr 24, 2026 Read → - ai · arxiv/cs.AI · 8 min
Junk Data Degrades LLM Reasoning; Twitter Study Shows Lasting Harm
Continual training on low-quality social media text causes measurable cognitive decline in language models, with reasoning and safety capabilities dropping significantly.
Apr 23, 2026 Read → - ai · arxiv/cs.LG · 8 min
Simpler Optimizers Make LLM Unlearning More Robust
Research shows that using lower-order optimization methods during LLM unlearning produces forgetting that resists post-training attacks better than sophisticated gradient-based approaches.
Apr 21, 2026 Read → - ai · arxiv/cs.LG · 4 min
Neural CTMC decouples discrete diffusion into timing and direction
A new parameterization for discrete diffusion models separates when and where tokens jump, aligning training with mathematical structure.
Apr 20, 2026 Read → - ai · arxiv/cs.LG · 4 min
Quantum-LSTM hybrid cuts physics model training data by 100×
Federated learning with quantum-enhanced LSTM achieves classical accuracy on SUSY classification using 20K samples instead of 2M, with under 300 parameters.
Apr 20, 2026 Read → - ai · arxiv/cs.AI · 8 min
Small Models Match Large Ones via Inference Scaffolding
McClendon et al. show that role-based prompt structuring at inference time doubles small-model performance on complex tasks without retraining.
Apr 17, 2026 Read → - ai · arxiv/cs.AI · 5 min
Verifiable model unlearning on edge devices without retraining
ZK-APEX combines sparse masking and zero-knowledge proofs to let providers verify that personalized models forget targeted data while preserving local utility.
Apr 17, 2026 Read → - ai · arxiv/cs.AI · 8 min
Modular Neural Networks Learn Three-Valued Logic Without Symbolic Solvers
THEIA demonstrates that dedicated domain engines enable neural networks to master Kleene three-valued logic and generalize compositionally to sequences 100x longer than training.
Apr 17, 2026 Read → - ai · arxiv/cs.LG · 8 min
INT4 Quantization Fails After FP32 Convergence in Predictable Phases
Post-training quantization assumes converged models are ready to compress, but INT4 quantization collapses in a three-phase pattern tied to weight updates, not learning rate decay.
Apr 17, 2026 Read →