Desired Dynamics-Based Generalized Inverse Solver for Estimation Problems

Liu, Shaojie; Zhang, Yulong; Gao, Zhiqiang; Chen, YangQuan; Li, Donghai; Zhu, Min

doi:10.3390/pr10112193

Cited by 5 publications

(5 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Figure 10. Samples for different number of steps nt of the DPMSolver++[32,33]. As expected, more steps correspond to more realistic samples.…”

supporting

confidence: 57%

“…we obtain a factorized p(z|x) = Stable Diffusion (SD) [42] using 20 steps with the DPM-Solver++ [32,33] as realization of p T 2I (x|t) and sample x of resolution 512 × 512. For unconditional synthesis p(x), we use the prompt "An image of a car.".…”

Section: Coverage Analysis Of Conditional Versus Unconditional Synthesismentioning

confidence: 99%

“…Figure 9 illustrates samples for different versions of Stable Diffusion [42]. Figure 10 illustrates samples for different number of steps n t of the DPMSolver++ [32,33].…”

Section: B Samples Zero-shot Benchmarkmentioning

confidence: 99%

See 2 more Smart Citations

Identification of Systematic Errors of Image Classifiers on Rare Subgroups

Metzen¹,

Hutmacher²,

Hua³

et al. 2023

Preprint

View full text Add to dashboard Cite

Despite excellent average-case performance of many image classifiers, their performance can substantially deteriorate on semantically coherent subgroups of the data that were under-represented in the training data. These systematic errors can impact both fairness for demographic minority groups as well as robustness and safety under domain shift. A major challenge is to identify such subgroups with subpar performance when the subgroups are not annotated and their occurrence is very rare. We leverage recent advances in text-to-image models and search in the space of textual descriptions of subgroups ("prompts") for subgroups where the target model has low performance on the prompt-conditioned synthesized data. To tackle the exponentially growing number of subgroups, we employ combinatorial testing. We denote this procedure as PROMPTAT-TACK as it can be interpreted as an adversarial attack in a prompt space. We study subgroup coverage and identifiability with PROMPTATTACK in a controlled setting and find that it identifies systematic errors with high accuracy. Thereupon, we apply PROMPTATTACK to ImageNet classifiers and identify novel systematic errors on rare subgroups.

show abstract

“…Figure 10. Samples for different number of steps nt of the DPMSolver++[32,33]. As expected, more steps correspond to more realistic samples.…”

supporting

confidence: 57%

Section: Coverage Analysis Of Conditional Versus Unconditional Synthesismentioning

confidence: 99%

See 1 more Smart Citation

Identification of Systematic Errors of Image Classifiers on Rare Subgroups

Metzen¹,

Hutmacher²,

Hua³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…The authors have proposed a short and general tuning procedure for ADRC 21,22 for inner loop tuning of ADRC. It initializes the parameters and iterates through a simple set of rules to find the optimal solution.…”

Section: Tuning Procedures For Bfc-adrcmentioning

confidence: 99%

“…The development of linear ADRC 13 substantially aided the industrial use of ADRC. [14][15][16][17] ADRC has been used for a long time to study the control of large inertial systems, 18 such as chemical production process, 19 fully distributed automatic generation control, 20 distributed parameter system, [21][22][23][24] main steam temperature control. 25 In their work, 26 the authors achieved almost feedback-linearization by employing a low-power extended observer.…”

mentioning

confidence: 99%

Robust compensation for active disturbance rejection control of a class of high‐order systems with large time delay: Simulation and experimental studies

Liu,

Wu,

Dai

et al. 2023

Intl J Robust & Nonlinear

Self Cite

View full text Add to dashboard Cite

Rejection of unknown disturbances and uncertainties in high‐order nonlinear systems with large delays is an important control problem. Active disturbance rejection control (ADRC) solves disturbances and uncertainties in a simple and effective way. In engineering, low‐order ADRC is often used to control high‐order large‐delay systems because of its ease of implementation and simplicity of operation. However, this leads to the problem that the low‐order ADRC has a limited ability to configure the high‐order nonlinear systems. The desired dynamics and the closed‐loop dynamics are almost always very biased. For a class of high‐order nonlinear systems with delay, a bias feedback compensation (BFC) approach to ADRC is suggested. This strategy works by adding a feedback loop based on the desired dynamics over the ADRC loop. The BFC strategy can improve the poor performance of low‐order ADRC for higher‐order nonlinear systems. In addition, a synthesis parameter tuning procedure of BFC‐ADRC is proposed in this article. Then, the stability problem is analyzed in this article. Finally, high‐fidelity simulation studies on a 1000 MW power plant for superheater steam temperature control and experiments on three interacting water tanks for level control are performed to verify the tracking, disturbance rejection, and robustness of the control strategy. Both simulation and experiments show that the method is highly robust and significantly improves both disturbance rejection and tracking performance.

show abstract