Research map / Architectures and efficiency / Neural architecture search
Teacher Guided Architecture Search
Pouya Bashivan, Mark Tensen, James J. DiCarlo
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add teacher RDM similarity (TG) to premature performance score (ablation) | P-guided search · CIFAR-100 architecture search (RL, 10-layer space, 2 epochs/sample) | best C100 error (%): 41.0 -> 38.3 | Improved |
| add teacher RDM similarity (TG) to premature performance score (ablation) | P-guided search · CIFAR-100 architecture search (RL, 10-layer space, 20 epochs/sample) | best C100 error (%): 40.5 -> 39.2 | Improved |
| add teacher RDM similarity (TG) to premature performance score (ablation) | P-guided search · CIFAR-100 architecture search (RL, 20-layer space, 2 epochs/sample) | best C100 error (%): 37.5 -> 33.2 | Improved |
| add teacher RDM similarity (TG) to premature performance score (ablation) | P-guided search · CIFAR-100 architecture search (RL, 20-layer space, 20 epochs/sample) | best C100 error (%): 32.7 -> 32.2 | Improved |
| add teacher RDM similarity (TG) to premature performance score (ablation) | P-guided search · CIFAR-100 architecture search (TPE, 10-layer space, 2 epochs/sample) | best C100 error (%): 42.5 -> 37.6 | Improved |
| add teacher RDM similarity (TG) to premature performance score (ablation) | P-guided search · CIFAR-100 architecture search (TPE, 20-layer space, 2 epochs/sample) | best C100 error (%): 37.0 -> 33.0 | Improved |
| use P+TG at 2 epochs/sample instead of P at 20 epochs/sample (ablation) | P-guided search · CIFAR-100 architecture search (RL, 10-layer space) | best C100 error (%): 40.5 -> 38.3 | Improved |
| increase TG weight alpha (ablation) | P+TG score · predicting mature performance | correlation with mature performance | Mixed |
| use NASNet instead of ResNet as the teacher network (ablation) | TG-SAGE · predicting mature performance | correlation with mature performance | Improved |
| use macaque neural RDMs as teacher with P+TG instead of P alone (ablation) | P-guided search · ImageNet top-1 error | top-1 error (%): 34.4 -> 32.54 | Improved |
| replace 7x1/1x7 convolutions with 7x7 separable convolutions (ablation) | SAGENet · CIFAR-100 test error | error (%): 17.42 -> 17.51 | No clear effect |
| replace 7x1/1x7 convolutions with 7x7 separable convolutions (ablation) | SAGENet · CIFAR-10 test error | error (%): 3.66 -> 3.88 | Hurt |
Closest papers by meaning
- TransNAS-Bench-101: Improving Transferability and Generalizability of Cross-Task Neural Architecture Search (CVPR 2021 · similarity 0.89)
- Vision-Oriented Lightweight Neural Architecture Search with Budget-Adaptive Evaluation (CVPR 2026 · similarity 0.89)
- Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective (ICLR 2021 · similarity 0.88)
- A Semi-Supervised Assessor of Neural Architectures (CVPR 2020 · similarity 0.88)
- Teacher Guided Neural Architecture Search for Face Recognition (AAAI 2021 · similarity 0.88)
- einspace: Searching for Neural Architectures from Fundamental Operations (NeurIPS 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
