Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Stabilizing LTI Systems under Partial Observability: Sample Complexity and Fundamental Limits
Authors: Ziyi Zhang, Yorie Nakahira, Guannan Qu
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To validate the claim in Theorem 5.3, we use the LTS-P to stabilize a randomly generated partially observable system and compare the result with two state-of-the-art benchmarks [43, 51] in Appendix A. ... Figure 1: The above figure shows the length of rollouts needed to identify and stabilize an unstable system with the unstable dimension k = 5. The solid line shows the average length of rollouts the learner takes to stabilize the system. The shaded area shows the standard deviation of the length of rollouts. ... Figure 2: The above figure shows the number of rollouts needed to identify and stabilize an unstable system with the unstable dimension k = 5. ... Figure 3: The above figure shows the probability of stabilization for a randomly generated matrix with n = 15 and k = 5. |
| Researcher Affiliation | Academia | Ziyi Zhang, Yorie Nakahira, Guannan Qu Department of Electrical and Computer Engineering Carnegie Mellon University Pittsburgh, PA 15213 ziyizhan,yorie,EMAIL |
| Pseudocode | Yes | Algorithm 1 LTS-P: learning the Hankel Matrix |
| Open Source Code | Yes | The code is attached to the submission, which exactly reproduces the desired result. In the simulation section, we also extensively discuss how to reproduce the result. ... The code is attached in the supplemental material. |
| Open Datasets | No | For each dimension n, we randomly generate a matrix with k unstable eigenvalues λi ∼ unif(1.1, 2) and n − k stable eigenvalues λj ∼ unif(0, 0.5). |
| Dataset Splits | No | The paper describes generating M trajectories each of length T for data collection, but does not specify any training/test/validation splits for a fixed dataset as the data is generated dynamically for simulations. |
| Hardware Specification | No | The experiment is simple and is not device-dependent. |
| Software Dependencies | No | No specific software dependencies with version numbers are mentioned in the paper. |
| Experiment Setup | Yes | In both parts of the simulations, we fix m = 4 and p = q = T/4 − 2 for the proposed algorithm, LTS-P. ... The system is simulated in three different settings, each with noise wt, vt ∼ N(0, σ), with σ = 0.4, 0.6, 0.8. |