Research map / Label-efficient and robust learning / Recognition and long-tail
Hide&Seek: Learning to Explain in an End-to-End Differentiable Network
Tal Ellinson, Hadi Mohasel Afshar, Sally Cripps
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 |
|---|---|---|---|
| Replace joint INVASE selector-predictor with Hide&Seek | Selector-predictor IWFS · Synthetic Syn4 | TPR (%): 90 -> 99 | Improved |
| Use Hide&Seek instead of REAL-x | Selector-predictor IWFS · Synthetic Syn5 | TPR (%): 88 -> 97 | Improved |
| Use Hide&Seek instead of INVASE for training time | Selector-predictor IWFS · Synthetic data run time | run time (hh:mm:ss): 01:18:52 -> 00:00:05 | Improved |
| Replace zero/mean ablation with random marginal draws and continuous masks (vs INVASE) (ablation) | Hide&Seek · Switch accuracy Syn4-6 | switch accuracy: 51 -> 100 | Improved |
| Replace zero/mean ablation with random marginal draws (vs L2X) (ablation) | Hide&Seek · Switch accuracy Syn4-6 | switch accuracy: 58 -> 100 | Improved |
| Correct INVASE's failures to identify the switch feature (ablation) | INVASE · Synthetic Syn4 | TPR (%): 90 -> 99.9 | Improved |
| Add high feature correlation rho up to 0.9 | Hide&Seek · Correlated Syn4-6 | switch accuracy: 0.998 -> 0.994 | No clear effect |
| Use Hide&Seek instead of INVASE under high feature correlation | Selector-predictor IWFS · Credit default semi-synthetic Syn4 | F1 (%): 57 -> 74 | Improved |
| Use Hide&Seek instead of INVASE under high feature correlation | Selector-predictor IWFS · Credit default semi-synthetic Syn2 | F1 (%): 100 -> 67 | Hurt |
| Use Hide&Seek instead of INVASE on real credit default labels | Selector-predictor · Credit default CDY (real labels) | AUROC: 0.771 -> 0.770 | No clear effect |
| Add parsimony-weight annealing (quadratic growth of lambda) (ablation) | Hide&Seek · Synthetic data | feature importance metrics | Improved |
| Use Hide&Seek instead of INVASE for AUROC on Syn1 credit labels | Selector-predictor · Credit default semi-synthetic Syn1 labels | AUROC: 0.601 -> 0.664 | Improved |
Closest papers by meaning
- Estimating Conditional Mutual Information for Dynamic Feature Selection (ICLR 2024 · similarity 0.84)
- SAND: One-Shot Feature Selection with Additive Noise Distortion (ICML 2025 · similarity 0.84)
- Contextual Feature Selection with Conditional Stochastic Gates (ICML 2024 · similarity 0.84)
- Explainable Neural Networks with Guarantee: A Sparse Estimation Approach (AAAI 2025 · similarity 0.84)
- Complexity Matters: Feature Learning in the Presence of Spurious Correlations (ICML 2024 · similarity 0.83)
- Local Feature Selection without Label or Feature Leakage for Interpretable Machine Learning Predictions (ICML 2024 · 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.
