multi-objective nas

**Multi-Objective NAS** is a **neural architecture search approach that simultaneously optimizes multiple competing objectives** — such as accuracy, latency, model size, energy consumption, and memory, producing a Pareto frontier of architectures representing different trade-offs. **How Does Multi-Objective NAS Work?** - **Objectives**: Accuracy ↑, Latency ↓, Parameters ↓, FLOPs ↓, Energy ↓. - **Pareto Frontier**: The set of architectures where no objective can be improved without degrading another. - **Methods**: Evolutionary algorithms (NSGA-II), scalarization (weighted sum), or Bayesian optimization. - **Selection**: User picks from the Pareto frontier based on deployment constraints. **Why It Matters** - **Real-World Trade-offs**: No single architecture is best — deployment requires balancing multiple constraints. - **Design Space Exploration**: Reveals the fundamental trade-off curves between competing metrics. - **Flexibility**: The Pareto set provides multiple deployment options from a single search. **Multi-Objective NAS** is **architectural diplomacy** — finding the set of optimal compromises between accuracy, speed, size, and power consumption.