Research map / Architectures and efficiency / RNNs and normalization
COLA: Orchestrating Error Coding and Learning for Robust Neural Network Inference Against Hardware Defects
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 COLA components (separation, AAWO, TC regularization) to ECOC | AlexNet with ECOC · CIFAR10 state error | clean accuracy (γ=0): 68.04 -> 79.04 | Helped |
| Add COLA components to ECOC | VGG-16 CIFAR100 under bit-flip errors · bit-flip α=0.05 | accuracy (%): 01.47 -> 41.51 | Helped |
| Add separation architecture plus AAWO and TC regularization to ECOC | LeNet-5 MNIST · state error γ=0.4 | accuracy (%): 81.09 -> 93.12 | Helped |
| Conventional ECOC replacing one-hot softmax | AlexNet CIFAR10 · clean accuracy (γ=0) | accuracy (%): 72.33 -> 68.04 | Hurt |
| Add separation architecture to ECOC (ECOC+Sep, prior work) | AlexNet CIFAR10 · state error γ=0 | accuracy (%): 68.04 -> 71.67 | Helped |
| Add TC output regularizer to ECOC (ablation) | AlexNet CIFAR10 · state error γ=0 | accuracy (%): 68.04 -> 78.35 (bit-flip α=0) / 77.08 is COLA-only | Helped |
| Add AAWO orthogonalization to separation architecture (ECOC+Sep+orth) (ablation) | AlexNet CIFAR10 · state error γ=0.5 | accuracy (%): 54.48 -> 56.34 (vs ECOC+Sep) | Helped |
| Relax weight orthogonalization to AAWO (diagonal Dm constraint) instead of strict WO (ablation) | VGG-16 CIFAR100 · state error γ=0.3 | accuracy (%): Strict WO 25.11-ish range; AAWO 66.12 vs WO 60.51 at 0.2 | Helped |
| Use stronger code (Hadamard-15) without COLA (ablation) | AlexNet CIFAR10 ECOC · state error across variation levels | accuracy (%): ECOC-15 performs worse than ECOC-10 at all variation levels | Hurt |
| Apply COLA to a stronger code (Hadamard-15 vs one-hot code length 10) (ablation) | AlexNet CIFAR10 ECOC · state error | accuracy (%): ECOC+TC+Sep+orth-15 outperforms ECOC+TC+Sep+orth-10 | Helped |
| Apply COLA to original softmax one-hot DNN (Original-COLA) (ablation) | AlexNet CIFAR10 · state error γ=0 and γ=0.5 | accuracy (%): 72.33 -> 76.68 (clean); 36.11 -> 66.74 (γ=0.5) | Helped |
| Apply COLA to original DNN and use ECOC codes (ECOC-COLA) rather than one-hot (ablation) | AlexNet CIFAR10 · state error γ=0.5 | accuracy (%): 66.74 -> 68.55 (Original-COLA vs ECOC-COLA) | Helped |
Closest papers by meaning
- Error Correction Output Codes for Robust Neural Networks against Weight-errors: A Neural Tangent Kernel Point of View (NeurIPS 2024 · similarity 0.90)
- Tightening Robustness Verification of MaxPool-based Neural Networks via Minimizing the Over-Approximation Zone (CVPR 2025 · similarity 0.84)
- SURE: SUrvey REcipes for building reliable and robust deep networks (CVPR 2024 · similarity 0.84)
- VNN: Verification-Friendly Neural Networks with Hard Robustness Guarantees (ICML 2024 · similarity 0.84)
- Towards General Robustness Verification of MaxPool-based Convolutional Neural Networks via Tightening Linear Approximation (CVPR 2024 · similarity 0.84)
- Verification of Neural Networks Against Convolutional Perturbations via Parameterised Kernels (AAAI 2025 · similarity 0.83)
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.