Algorithms & Systems for AI¶
What this subject is for: Making models fast, cheap, and deployable — distributed training, parallelism, kv-cache, FlashAttention, quantization, inference optimization.
Track status: 33 substantive concept pages · 4 stubs awaiting next cycle. See the live generation status and the latest retrospective.
Concepts¶
- Attention Mechanisms
- Automatic Differentiation
- Communication Collectives
- Compiler Optimizations for ML
- Scaling Agent Coordination Through Systems Architecture
- Constrained Learning
- Convex Optimization
- Curriculum Learning
- Curriculum resampling
- Data Parallelism
- Differentiable Optimization
- Distributed Training Arc
- Distributed Training
- Flash Attention
- Gradient bucketing
- Inference optimization
- KV Cache Management
- KV cache
- LLM Architecture Optimizations
- LLM Inference
- Mixed-precision training
- Model Deployment
- Model Parallelism
- Pipeline Parallelism
- Policy Gradient Theory
- Post-Training Quantization
- Precision scaling
- Quantization-Aware Training
- Quantization Basics
- Quantization
- Reinforcement Learning Schedulers
- Tensor Cores
- Tensor parallelism
Auto-seeded stubs awaiting next cycle: collective-communication, long-context-models, reinforcement-learning, rlhf-infrastructure-overview
Arcs through this subject¶
No arcs yet — the retrospective proposes these once concept coverage hits ≥4 pages per track.
Key thinkers¶
Author pages pending.
Builds tied to this subject¶
MVB recipes pending — currently they live inside concept pages' Build it sections.
Auto-rebuilt from filesystem state by scripts/rebuild_track_indexes.py — see system architecture.