Guarantees for Nonlinear Representation Learning: Non-identical Covariates, Dependent Data, Fewer Samples
Authors: Thomas Tck Zhang, Bruce D Lee, Ingvar Ziemann, George J. Pappas, Nikolai Matni
ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We support our analysis with numerical experiments performing imitation learning over non-linear dynamical systems. and 3. Numerical Validation To validate our theoretical observations, we consider a nontrivial regression task over dynamical systems: balancing a pole atop a cart from visual observations, as pictured in Figure 1(a). |
| Researcher Affiliation | Academia | 1Department of Electrical and Systems Engineering, University of Pennsylvania, Philadelphia, USA. Correspondence to: Thomas Zhang <ttz2@seas.upenn.edu>. |
| Pseudocode | No | The paper describes algorithmic schemes and mathematical derivations but does not include any clearly labeled 'Pseudocode' or 'Algorithm' block. |
| Open Source Code | No | The paper does not contain any statement or link indicating the release of open-source code for the described methodology. |
| Open Datasets | No | Figure 1(a) is one such observation generated by the Py Bullet simulator when the system is at the origin. and A single keypoint extractor is used by all experts (across the parameter variations of the system), and is trained from labeled data across a variety of parameter settings. The data is generated from a simulator for the experiments, not a publicly available dataset with access info. |
| Dataset Splits | No | Expert demonstrations are obtained from 10 independent realizations of the actuation noise sequence for each system. The length of the rollout trajectory is 500 steps (recalling the discretization timestep of 0.02.) and The second stage consists of 10 target tasks... Here, we consider a variable number of trajectories, Ntarget. The paper describes data generation and usage but does not provide specific train/validation/test dataset split information. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware (e.g., GPU, CPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper mentions 'Py Bullet simulator' but does not provide its version number or any other specific software dependencies with versions. |
| Experiment Setup | Yes | To obtain an approximate minimizer to the above problem, we employ the adam optimizer using a batch size of 32, weight decay of 1e 3, and learning rate of 1e 3 with a decay factor of 0.5 every 10 epochs for a total of 100 epochs. |