Research map / Label-efficient and robust learning / Recognition and long-tail
Reducing Network Agnostophobia
Akshay Raj Dhamija, Manuel Günther, Terrance Boult
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 |
|---|---|---|---|
| Train with Objectosphere loss instead of a background class | LeNet++ · Devanagari | CCR at FPR 10^-2: 0.7527 -> 0.9658 | Improved |
| Train with Objectosphere loss instead of a background class | LeNet++ · NotMNIST | CCR at FPR 10^-2: 0.9068 -> 0.9563 | Improved |
| Train with Entropic Open-Set loss instead of a background class | LeNet++ · CIFAR-10 unknowns | CCR at FPR 10^-2: 0.9726 -> 0.9804 | Improved |
| Use Scaled Objectosphere instead of a background class | ResNet-18 · SVHN | CCR at FPR 10^-2: 0.4803 -> 0.5454 | Improved |
| Use Scaled Objectosphere instead of a background class | ResNet-18 · CIFAR-100 subset | CCR at FPR 10^-2: 0.3429 -> 0.4334 | Improved |
| Add a feature-magnitude margin for known samples and penalize unknown feature magnitude (ablation) | Entropic Open-Set on LeNet++ · CIFAR-10 unknowns | CCR at FPR 10^-2: 0.9804 -> 0.9785 | Hurt |
| Add the Objectosphere feature-magnitude constraint (ablation) | Entropic Open-Set on LeNet++ · NotMNIST | CCR at FPR 10^-2: 0.9515 -> 0.9563 | Improved |
| Scale the softmax score by deep feature magnitude (ablation) | Objectosphere on ResNet-18 · SVHN | CCR at FPR 10^-2: 0.5074 -> 0.5454 | Improved |
| Scale the softmax score by deep feature magnitude (ablation) | Objectosphere on ResNet-18 · CIFAR-100 subset | CCR at FPR 10^-2: 0.3595 -> 0.4334 | Improved |
| Scale the softmax score by deep feature magnitude (ablation) | Objectosphere on LeNet++ · CIFAR-10 unknowns | CCR at FPR 10^-2 | No clear effect |
| Train with CIFAR samples as negative unknowns instead of NIST Letters (ablation) | LeNet++ · NIST Letters unknowns | robustness to unknowns | Hurt |
Closest papers by meaning
- Outliers with Opposing Signals Have an Outsized Effect on Neural Network Optimization (ICLR 2024 · similarity 0.81)
- Initial Guessing Bias: How Untrained Networks Favor Some Classes (ICML 2024 · similarity 0.81)
- Simplicity Bias of Two-Layer Networks beyond Linearly Separable Data (ICML 2024 · similarity 0.81)
- The Rich and the Simple: On the Implicit Bias of Adam and SGD (NeurIPS 2025 · similarity 0.80)
- DeNetDM: Debiasing by Network Depth Modulation (NeurIPS 2024 · similarity 0.80)
- Graph Agent Network: Empowering Nodes with Inference Capabilities for Adversarial Resilience (AAAI 2025 · similarity 0.80)
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.
