🎯 AI: techniques

Attack and defence techniques and machine learning methods that a second item could name too, with an ATT&CK, ATLAS or CWE id when one is stated. A procedure only one article describes is item text. Each one opens the items that name it, and the list is ranked by how many do.

This workspace's kept items name 295 of them. This is page 2 of 3, the most named first.

• Batch processing2 items
• Burn-in and RMA of GPU nodes2 items
• Cache engineering2 items
• Causal Encoder-Decoder architecture2 items
• Chain-of-thought and action monitoring2 items
• Chain-of-Thought Prompting2 items
• choice questions2 items
• Chunked prefill2 items
• CISPO2 items
• Clean-room rebuild2 items
• Codemode2 items
• compaction2 items
• Compute-based prefill scheduling2 items
• Compute-in-Flash2 items
• Constitutional classifiers2 items
• Constrained decoding2 items
• Continuous monitoring2 items
• Contrastive learning2 items
• Controlling where reasoning can go2 items
• Conversation analysis2 items
• CoT controllability evaluation2 items
• Cross-model review2 items
• Cyber data removal from training2 items
• DAPO2 items
• decision adapter2 items
• Decode routing by request count and KV length2 items
• decomposition2 items
• Dedicated memory agents2 items
• Defense in depth2 items
• Delta weight synchronization2 items
• Deployment simulation2 items
• Dictionary-based KV cache compression2 items
• Disaggregated prefill/decode2 items
• Dr. GRPO2 items
• Dreaming2 items
• drop2 items
• DSpark adaptive speculative verification2 items
• DSpark speculative decoding2 items
• Dual-track incremental scanning2 items
• Duration transformation of GPU capacity2 items
• Dynamic Client Registration2 items
• Dynamic GGUF quantisation2 items
• Dynamic multi-turn adversarial user simulation2 items
• ECI2 items
• End-to-end integer-only quantization2 items
• Explicit cache breakpoints2 items
• Explicit state handles2 items
• Exponential Time Hypothesis2 items
• Extended thinking / chain-of-thought reasoning2 items
• Fast mode2 items
• Few-step denoising2 items
• Flow Matching2 items
• Forced forgetting2 items
• Formally verified policy changes2 items
• Full-KL co-distillation2 items
• Generative UI2 items
• Global capacity pooling2 items
• Grouped GEMM2 items
• GSPO2 items
• Gumbel-max watermarking2 items
• Hardware-in-the-loop architecture search2 items
• Harness distillation2 items
• Harness reversal2 items
• Harness self-editing2 items
• Head-wise gated attention2 items
• HealthBench length adjustment2 items
• Hierarchical token pooling2 items
• History compaction2 items
• History-aware routing2 items
• History-dependent cache reuse2 items
• Honeypot tests2 items
• Human-in-the-Loop Validation2 items
• Hybrid Reasoning Agents2 items
• Hybrid search2 items
• HybridRL2 items
• HybridSFT2 items
• In-context learning2 items
• In-place tokenizer expansion2 items
• Inference2 items
• Inference-time scaling2 items
• Interpretability2 items
• Item Response Theory2 items
• JEPA2 items
• Kernel-level enforcement2 items
• Kind as posture, type as format2 items
• KV caching2 items
• LLM-as-judge2 items
• LLM-as-judge evaluation2 items
• Load Balancing2 items
• Log-barrier objective2 items
• Lookahead depth supervision2 items
• Looped architectures2 items
• Lossy self-improvement2 items
• Low-rank adaptation2 items
• LSTM2 items
• Making repeated attempts difficult2 items
• MCP server for tool access2 items
• Memorization detection2 items
• Memory token compression2 items
• Merlin-Arthur protocol2 items