Research map / Learning theory and algorithms / Sample complexity
Optimal Feedback Law Recovery by Gradient-Augmented Sparse Polynomial Regression
Behzad Azmi, Dante Kalise, Karl Kunisch
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 |
|---|---|---|---|
| add gradient information to the regression of the value function (ablation) | P-l2 polynomial regression · Test 1 validation (2D) | ErrL2: 1.46 × 10−1 -> 9.38 × 10−3 | Improved |
| add gradient information to the regression of the value function (ablation) | P-l2 polynomial regression · Test 1 validation (2D) | ErrH1: 1.17 -> 3.25 × 10−2 | Improved |
| add l1 sparsity penalty to the gradient-augmented regression (ablation) | AP-l2 regression · Test 1 nonzero components | nonzero components: 52/52 -> 19/52 | Improved |
| add l1 sparsity penalty to the gradient-augmented regression (ablation) | AP-l2 regression · Test 1 validation (2D) | ErrL2: 9.38 × 10−3 -> 1.20 × 10−2 | No clear effect |
| use sparse gradient-augmented regression instead of dense l2 for the feedback law | Test 1 feedback controller · Test 1 operation count | floating-point operations: 63% reduction in the number of operations | Improved |
| add gradient information so fewer training samples are needed (ablation) | sparse polynomial regression · Test 1 (2D) | training samples needed: Nd=120 -> Nd=40 | Improved |
| add gradient information in the 80-dimensional problem (ablation) | sparse polynomial regression · Test 3 (80D) | training samples needed: two orders of magnitude less training samples | Improved |
| add gradient information to sparse regression (ablation) | Test 3 sparse regression · Test 3 validation (80D) | ErrL2: 7.46 × 10−2 -> 3.56 × 10−2 | Improved |
| add gradient information to sparse regression (ablation) | Test 3 sparse regression · Test 3 validation (80D) | ErrH1: 9.38 × 10−2 -> 4.11 × 10−2 | Improved |
| add gradient information to sparse regression (ablation) | Test 3 control law · Test 3 stabilization (80D) | stabilization of dynamics: over 4000 training samples -> 50 and 70 samples | Improved |
| increase l1 penalty lambda (ablation) | AP-l1 sparse regression · Test 2 (Figure 9) | nonzero components and error | Mixed |
| increase polynomial order s from 4 to 8 (ablation) | gradient-augmented sparse regression · Test 2 (Figures 9-10) | validation error: s=4 -> s=8 | Improved |
Closest papers by meaning
- Learning Optimal Feedback Operators and their Sparse Polynomial Approximations (JMLR 2023 · similarity 0.93)
- Constrained Optimization From a Control Perspective via Feedback Linearization (NeurIPS 2025 · similarity 0.88)
- Iterative Linearized Control: Stable Algorithms and Complexity Guarantees (ICML 2019 · similarity 0.86)
- Sparsity in Partially Controllable Linear Systems (ICML 2022 · similarity 0.85)
- Optimal Control Operator Perspective and a Neural Adaptive Spectral Method (AAAI 2025 · similarity 0.84)
- Differentiable Nonlinear Model Predictive Control (NeurIPS 2026 · 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.
