Research map / Learning theory and algorithms / Prediction and decision losses
ResNEsts and DenseNEsts: Block-based DNN Models with Improved Representation Guarantees
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 |
|---|---|---|---|
| Use a sufficiently wide expansion with bottleneck residual blocks | ResNEst · Training empirical risk across block counts | empirical risk | Helped |
| Add prediction weights for each feature to decouple prediction from feature learning | ResNEst, yielding A-ResNEst · Training empirical risk with fixed features | empirical risk | Helped |
| Use dense connectivity that concatenates each block's features | DenseNEst · Training empirical risk across dense block counts | empirical risk | Helped |
| Use bottleneck blocks rather than basic residual blocks | ResNEst · Feature reuse and training performance | empirical risk | Helped |
| Increase input-space expansion enough to satisfy the bottleneck condition | ResNEst · Training empirical risk with multiple blocks | empirical risk | Helped |
| Add a feature that is linearly unpredictable from previous features | A-ResNEst · Residual representation improvement between blocks | empirical risk | Helped |
| Use dense concatenation instead of summing previous features | DenseNEst · Feature construction | empirical risk | Helped |
| Represent a DenseNEst with an equivalent wide ResNEst using orthogonal feature subspaces | ResNEst · Training empirical risk | empirical risk | Helped |
| Add nonlinear blocks under the stated conditions | ResNEst · Training empirical risk versus the best linear predictor | empirical risk | Helped |
Closest papers by meaning
- Residual Alignment: Uncovering the Mechanisms of Residual Networks (NeurIPS 2023 · similarity 0.90)
- Are deep ResNets provably better than linear predictors? (NeurIPS 2019 · similarity 0.89)
- Compute Better Spent: Replacing Dense Layers with Structured Matrices (ICML 2024 · similarity 0.87)
- Neural Collapse is Globally Optimal in Deep Regularized ResNets and Transformers (NeurIPS 2025 · similarity 0.86)
- Revisiting Residual Connections: Orthogonal Updates for Stable and Efficient Deep Networks (NeurIPS 2025 · similarity 0.86)
- Depthwise Hyperparameter Transfer in Residual Networks: Dynamics and Scaling Limit (ICLR 2024 · similarity 0.86)
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.