メニュー
このリポジトリの skills
SkillsMP は hiyenwong/ai_collection から 4,114 件の skill を収集しています。skill を開くとソースと詳細を確認できます。
hiyenwong/ai_collection収集済み skill 4,114 件中 40 件を表示しています。
Majorization lattice supermodularity and subadditivity framework — two structural majorization relations (precursors) underlying supermodularity and subadditivity of all sum-concave functions including Tsallis, Rényi, and Shannon entropies on the majorization…
原文の言語: 英語
"Multi-Agent Runtime Grading via Incremental Normalization (MARGIN) — online confidence calibration for multi-agent AI coordination. Use when building multi-agent systems that need to weight agent trustworthiness at runtime: (1) coordinating responses from…
原文の言語: 英語
Medical image domain adaptation and transfer learning methodology. Use when working with medical imaging AI tasks including: (1) adapting pre-trained models to new clinical domains with scarce annotated data, (2) parameter-efficient fine-tuning for medical…
原文の言語: 英語
Category-theory-based brain-DNN alignment methodology using Naturality Violation Score (NVS). Shifts alignment assessment from per-stimulus correspondence to preservation of candidate transformations. Activation: brain-DNN alignment, naturality violation, RSA…
原文の言語: 英語
On-Policy Distillation (OPD) methodology for transforming autoregressive models into diffusion language models efficiently, eliminating train-inference mismatch.
原文の言語: 英語
Persona-Pruner methodology for sculpting lightweight language models for role-playing tasks. Enables efficient pruning of LMs while preserving persona-consistent stylized interactions. Use when: model pruning for role-playing, lightweight character chatbots,…
原文の言語: 英語
Polynomial acyclicity constraints for efficient continuous causal discovery in visual semantic graphs - 33% speedup over exponential baseline with improved F1 scores.
原文の言語: 英語
Readiness-First Pipeline (RRFP) methodology — treating pipeline schedules as non-binding hint orders rather than pre-committed execution sequences. Reduces bubbles and stage misalignment in distributed training under runtime variability. Up to 1.77x speedup…
原文の言語: 英語
Reconfigurable Nonlinear Photonic Decision Network (RNPDN) methodology for adaptive photonic neuromorphic computing. Local physical learning rules with tunable stability-plasticity tradeoff, controlled memory formation via bistable photonic states, and…
原文の言語: 英語
Train hundred-billion-parameter sparse MoE models on single nodes using reversible recurrence stacks and state-preserving growth principles with TQP optimizer strategy.
原文の言語: 英語
Information-geometry unified memory architecture combining Riemannian retrieval (Fisher-Rao metric) with Fisher-guided discrete token distillation for resource-efficient long-term memory in dialogue agents.
原文の言語: 英語
Contrastive on-policy self-distillation methodology for reasoning models that mitigates privilege-induced style drift
原文の言語: 英語
Rollout-Adaptive Supervised Fine-Tuning (RASFT) for reasoning tasks - policy-aware SFT that calibrates expert supervision based on problem-level solvability from verified rollouts.
原文の言語: 英語
SAE 最优性结构理论 - 解释 Sparse Autoencoders 如何从最优性条件提取可解释特征。涵盖层次分裂与吸收、残差结构、密集对立特征等现象的理论基础。
原文の言語: 中国語
Theory explaining how optimality conditions structure SAE (Sparse Autoencoder) dictionaries - hierarchical splitting, absorption, residuals, and dense antipodal features
原文の言語: 英語
Self-evolution skill that uses dual-agent challenge design and execution to expand capabilities over time.
原文の言語: 英語
自我验证技能，基于 ReVeal 论文实现多轮生成-验证迭代，支持代码和推理任务的可靠自我验证。触发词：自我验证、self-verification、verify、验证代码、验证推理。
原文の言語: 中国語
GPU-accelerated semidefinite programming for causal game analysis — using SDP hierarchies to compute bounds in causal inference games, with GPU acceleration for scalability. From arXiv:2606.20519.
原文の言語: 英語
Efficient semi-structured LLM sparsification via annealing of Hessian-mask guided pruning. Achieves high sparsity with minimal accuracy loss using second-order importance estimation.
原文の言語: 英語
Analysis of Neural Tangent Kernel (NTK) collapse near dynamical bifurcations in state-space models. Studies how the NTK spectrum degrades as recurrent networks approach critical transitions. Activation: NTK collapse, bifurcation analysis, state-space NTK,…
原文の言語: 英語
SuCo - Sufficiency-guided Continuous Adaptive Reasoning for LRM efficiency. Minimal Sufficient CoT (MSC) defines shortest prefix adequate for correct answer. Two-stage training: MSC-Aligned Fine-Tuning + Sufficiency-Aware Policy Optimization. Use when: (1)…
原文の言語: 英語
Super Factory multi-agent pipeline system — E2E testing, real execution, and contract debugging.
原文の言語: 英語
Methodology from Anthropic research for improving alignment training to reduce agentic misalignment through principle-based training, "difficult advice" datasets, and counterfactual data augmentation.
原文の言語: 英語
Universal complementarity identity for quantum interferometry — exact trade-off relation between path distinguishability and interference visibility for polarized double-slit experiments, with extensions to quantum information protocols. Activation:…
原文の言語: 英語
Vibe Calibration methodology for autonomous quantum processor bring-up using LLM skill orchestration. Distills expert tacit knowledge into reusable calibration skills for superconducting quantum processors. Use when designing autonomous calibration systems…
原文の言語: 英語
Quantum-inspired GAN methodology for high-resolution medical image generation with prototype-guided skip connections and dual-stream generator. Addresses data scarcity, class imbalance, and privacy constraints in medical imaging through variational quantum…
原文の言語: 英語
End-to-End Encrypted Control Pipeline for Multi-Agent Coordination via CKKS Homomorphic Encryption. Enables privacy-preserving cloud-based coordination by redesigning control loops for FHE constraints. Activation: encrypted control, homomorphic encryption,…
原文の言語: 英語
Metabolic Multi-Agent Optimizer (MMAO) - bio-inspired optimization with endogenous resource allocation. Each agent carries internal energy with private-public metabolic loop. Fitness improvements converted to metabolic gains regulating sensing, search…
原文の言語: 英語
Methodology from Anthropic research (Jun 2026) on making biological data infrastructure agent-friendly. Case study shows that adding deterministic retrieval layers (like gget virus) to scientific research agents improves accuracy from inconsistent results to…
原文の言語: 英語
Runtime governance framework for heterogeneous agent systems using action certificates. Model-agnostic governance centered on action proofs rather than vendor-native session records. From Anthropic research (arXiv:2606.04104).
原文の言語: 英語
Compiler-driven sub-microsecond feedback control stack for trapped-ion quantum experiments. Use when designing quantum control software stacks, compiler pipelines for hardware control, deterministic low-latency feedback systems, DSL transpilation, or…
原文の言語: 英語
Execution-time AI alignment architecture using an unfireable safety kernel that operates outside the agent's address space. Ensures safety controls cannot be bypassed by the AI agent itself, addressing the fundamental vulnerability of in-process guardrails.
原文の言語: 英語
Boltzmann Attention methodology — energy-based generalization of attention using interacting Ising models with learnable pairwise couplings. Opens path to quantum annealing-based training.
原文の言語: 英語
Brain-Prompt Injection security audit framework for BCI-LLM agents with route-safety audit contract, C3 decomposition, and split-conformal calibration
原文の言語: 英語
Comprehensive open-source toolkit unifying continuous attractor neural network (CANN) research workflow. Three co-designed components (Python/Rust/GUI) for 1D/2D attractor modeling, spike-frequency adaptation, grid cells, path integration, and persistent…
原文の言語: 英語
CANNs toolkit for continuous attractor neural network research - unified Python/Rust/PySide6 framework for modeling spatial navigation, grid cells, head-direction cells, and attractor dynamics analysis. Activation: CANNs toolkit, continuous attractor neural…
原文の言語: 英語
CARVE (Content-Aware Recurrent with Value Efficiency) methodology — content-aware gating for recurrent models that resolves memory-blind gating, value-axis waste, and enables WY-form chunk-parallel training.
原文の言語: 英語
Computational modeling methodology for chronic stress effects on prefrontal working memory networks via excitatory-inhibitory (E/I) balance perturbation. Use when modeling stress-induced cognitive dysfunction, E/I ratio alterations, or prefrontal cortex…
原文の言語: 英語
Computational modeling methodology for chronic stress as E/I perturbation in recurrent working-memory networks. Identifies enhanced inhibitory-to-excitatory synaptic strength as best-fit mechanism and reveals resilience-generalization trade-off.
原文の言語: 英語
Framework for functional dissociation between cortical and subcortical systems during learning under memory constraints - cortex supports general structure learning while subcortex specializes in reward-based learning
原文の言語: 英語