Streaming Linear System Identification with Reverse Experience Replay
Authors: Suhas Kowshik, Dheeraj Nagaraj, Prateek Jain, Praneeth Netrapalli
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Furthermore, SGD RER can be applied to more general settings like sparse LTI identification with known sparsity pattern, and non-linear dynamical systems. Our work demonstrates that the knowledge of data dependency structure can aid us in designing statistically and computationally efficient algorithms which can decorrelate streaming samples. Finally, we also provide empirical validation of our method on simulated data, and demonstrate that the proposed method is indeed able to provide error rate similar to the OLS method while methods like SGD and standard experience replay, lead to biased estimates. |
| Researcher Affiliation | Collaboration | Prateek Jain Google AI Research Lab, Bengaluru, India 560016 prajain@google.com Suhas S Kowshik Department of EECS MIT, Cambridge, MA 02139 suhask@mit.edu Dheeraj Nagaraj Department of EECS MIT, Cambridge, MA 02139 dheeraj@mit.edu Praneeth Netrapalli Google AI Research Lab, Bengaluru, India 560016 pnetrapalli@google.com |
| Pseudocode | Yes | Algorithm 1: SGD RER Input :Streaming data {Xτ}, horizon T, buffer size B, buffer gap u, bound R, tail average start: a Output :Estimate ˆAa,t, for all a < t N 1; N = T/(B + u) |
| Open Source Code | No | The paper does not provide a specific link or explicit statement about releasing the source code for the methodology described. |
| Open Datasets | No | Synthetic data: We sample data from VAR(A , µ) with X0 = 0, µ N(0, σ2I) and A Rd d is generated from the 'Rand Bi Mod' distribution. That is, A = UΛU with random orthogonal U, and Λ is diagonal with d/2 entries on diagonal being ρ and the remaining diagonal entries are set to ρ/3. We set d = 5, ρ = 0.9 and σ2 = 1. We fix a horizon T = 107 and set the buffer size as B = 100 and u = 10. To estimate R from the data, we use the first 2 log T = 32 samples and set R as the sum of the norms of these samples. This describes synthetic data generation rather than the use of a publicly available dataset with concrete access information. |
| Dataset Splits | No | The paper describes generating synthetic data for experiments and mentions using the first 2 log T samples to estimate R, but it does not specify explicit training, validation, or test dataset splits (e.g., percentages, sample counts, or predefined splits) for reproducibility. |
| Hardware Specification | No | The paper does not mention any specific hardware (e.g., GPU models, CPU types, or cloud resources with specifications) used for running the experiments. |
| Software Dependencies | No | The paper does not list specific software dependencies with version numbers (e.g., Python 3.x, TensorFlow 2.x, PyTorch 1.x). |
| Experiment Setup | Yes | We set d = 5, ρ = 0.9 and σ2 = 1. We fix a horizon T = 107 and set the buffer size as B = 100 and u = 10. To estimate R from the data, we use the first 2 log T = 32 samples and set R as the sum of the norms of these samples. We let the stepsize to be γ = 1 2R which is aggressive compared to our theorems. We start the SGD RER and other SGD-like algorithms from the second buffer onward. |