Research map / Learning theory and algorithms / Sample complexity
Error Analysis for Deep ReLU Feedforward Density-Ratio Estimation with Bregman Divergence
Siming Zheng, Guohao Shen, Yuanyuan Lin, Jian Huang
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 |
|---|---|---|---|
| use mixing chain (mTRE) instead of convolution chain (cTRE) | telescoping density ratio estimation · Beta (d=5), n=5000, K=5 | MSE: 1.4670 -> 0.9850 | Improved |
| use mixing chain instead of convolution chain | telescoping density ratio estimation · Normal setting, n=5000, K=10 | MSE: 2.7076 -> 2.8330 | Hurt |
| use mixing chain instead of convolution chain | telescoping density ratio estimation · Normal setting, n=5000, K=5 | MSE: 2.7987 -> 2.7426 | No clear effect |
| use mixing chain instead of convolution chain, chain length 5 | telescoping density ratio estimation · MNIST | ANLL: 1.41 -> 1.40 | Improved |
| use mixing chain with chain length 10 instead of convolution chain | telescoping density ratio estimation · MNIST | ANLL: NAN failure -> 1.39 | Improved |
| increase training sample size n from 100 to 100000 | BDD · Model (TN) | average MSE: 3.48502 -> 0.00376 | Improved |
| increase training sample size n from 100 to 100000 | BDD · Model (NI) | average MSE: 6.32138 -> 0.17945 | Improved |
| make X heavy tailed (Student-t(2)) violating Assumption 12 instead of normal | BDD · Assumption 12 check, n=100000 | average MSE: 0.19501 -> 1.95932 | Hurt |
| let alpha K^-1 approach 1 (0.9 to 0.9999999) | TRE · d=1 | average MSE: 1.473×101 -> 1.542×10−7 | Improved |
| let alpha K^-1 approach 1 (0.9 to 0.9999999) | TRE · d=5 | average MSE: 5.296×1013 -> 6.689×10−2 | Improved |
| increase test sample size m from 100 to 400 in KL divergence CI | KL divergence estimate · d=5, n=5000, rho=0.4 | coverage probability: 0.935 -> 0.805 | Hurt |
Closest papers by meaning
- Binary Losses for Density Ratio Estimation (ICLR 2025 · similarity 0.86)
- Bounds on $L_p$ Errors in Density Ratio Estimation via $f$-Divergence Loss Functions (ICLR 2025 · similarity 0.86)
- Overcoming Saturation in Density Ratio Estimation by Iterated Regularization (ICML 2024 · similarity 0.85)
- Nonparametric Estimation of a Factorizable Density using Diffusion Models (NeurIPS 2026 · similarity 0.85)
- Revive Re-weighting in Imbalanced Learning by Density Ratio Estimation (NeurIPS 2024 · similarity 0.84)
- Projection Pursuit Density Ratio Estimation (ICML 2025 · similarity 0.84)
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.
