Skip to content

[Phase 3] Implement Neural Architecture Search Algorithms #403

@ooples

Description

@ooples

Problem

Has AutoML framework but missing specific NAS algorithms.

Existing

  • src/AutoML/NeuralArchitectureSearch.cs (framework)
  • src/AutoML/NeuralArchitectureSearchStrategy.cs
  • src/AutoML/IAutoMLModel.cs

Missing Implementations

Differentiable NAS (CRITICAL):

  • DARTS (Differentiable Architecture Search)
  • GDAS (Gradient-based NAS)
  • PC-DARTS (Partial Channel DARTS)

Efficient NAS (HIGH):

  • ENAS (Efficient Neural Architecture Search)
  • ProxylessNAS
  • FBNet (hardware-aware NAS)

One-Shot NAS (HIGH):

  • Once-for-All Networks
  • BigNAS
  • AttentiveNAS

Search Spaces (MEDIUM):

  • MobileNet search space
  • ResNet search space
  • Transformer search space

Architecture

  • Expand src/AutoML/NAS/
  • Search space definitions
  • Hardware cost modeling

Success Criteria

  • ImageNet architecture search
  • Transfer to downstream tasks
  • Hardware latency constraints
  • Parity with NAS-Bench-201

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions