Research map / Learning theory and algorithms / Matrix and tensor methods
Geometric Analysis of Matrix Sensing over Graphs
Haixiang Zhang, Ying Chen, Javad Lavaei
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 prior incoherence regularizer with a twice-differentiable convolution-smoothed regularizer | regularized MSoG objective (12) · strict-saddle sampling-rate requirement | required sampling rate np (bound): np >= Theta[mu^4 r^6 (kappa)^6 log n] -> np >= Theta[mu^2 r^3 (kappa)^2 log n] | Improved |
| generalize RIP to partial observations (Omega-RIP) for losses on graph-observed entries | matrix sensing over graphs · strict saddle property | admissible RIP constant delta: delta < 1/16 | Improved |
| raise Omega-RIP constant from 1/16 to 1/2 using loss f_{3/2} | problem (12) · counterexample, any p | spurious second-order critical point probability: probability at least (3 - sqrt5)/2 ~ 0.38 | Hurt |
| reduce Omega-RIP constant delta in the simulation | perturbed accelerated gradient descent on f_c · n=10, r=5 synthetic, 100 instances | success rate (optimization complexity) | Mixed |
| increase sampling rate p from 0.7 toward 1.0 | perturbed accelerated gradient descent on f_c · n=10, r=5 synthetic, 100 instances | success rate (optimization complexity) | Mixed |
| use partial sampling (p<1) instead of full observation (p=1) when delta > 0.5 | perturbed accelerated gradient descent on f_c · n=10, r=5 synthetic | success rate | Improved |
| vary delta below 1/2 versus vary p | perturbed accelerated gradient descent on f_c · n=10, r=5 synthetic | success rate | No clear effect |
| gaussian linear measurements with m >= c n r / delta^2 on graph-restricted entries | linear matrix sensing over graphs · Omega-RIP 2r,2r | RIP holds w.h.p.: m >= cnr/delta^2 | Improved |
Closest papers by meaning
- Decentralized Matrix Sensing: Statistical Guarantees and Fast Convergence (NeurIPS 2023 · similarity 0.86)
- No Spurious Local Minima in Nonconvex Low Rank Problems: A Unified Geometric Analysis (ICML 2017 · similarity 0.86)
- Over-parametrization via Lifting for Low-rank Matrix Sensing: Conversion of Spurious Solutions to Strict Saddle Points (ICML 2023 · similarity 0.86)
- Sharp Restricted Isometry Property Bounds for Low-Rank Matrix Recovery Problems with Corrupted Measurements (AAAI 2022 · similarity 0.86)
- Semi-Random Matrix Completion via Flow-Based Adaptive Reweighting (NeurIPS 2024 · similarity 0.84)
- Symmetric Matrix Completion with ReLU Sampling (ICML 2024 · 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.
