Search

This paper presents a kernelized offset-free data-driven predictive control scheme for nonlinear systems. Traditional model-based and data-driven predictive controllers often struggle with inaccurate predictors or persistent disturbances, especially in the case of nonlinear dynamics, leading to tracking offsets and stability issues. To overcome these limitations, we employ kernel methods to parameterize the nonlinear terms of a velocity model, preserving its structure and efficiently learning unknown parameters through a least squares approach. This results in a offset-free data-driven predictive control scheme formulated as a nonlinear program, but solvable via sequential quadratic programming. We provide a framework for analyzing recursive feasibility and stability of the developed method and we demonstrate its effectiveness through simulations on a nonlinear benchmark example.

Through exposing items to users, implicit feedback recommender systems influence the logged interactions, and, ultimately, their own recommendations. This effect is called exposure bias and it can lead to issues such as filter bubbles and echo chambers. Previous research employed the multinomial logit model (MNL) with exposure information to reduce exposure bias on synthetic data. This extended abstract summarizes our previous study in which we investigated whether (i) these findings hold for human-generated choices, (ii) other discrete choice models mitigate bias better, and (iii) an item's estimated relevance can depend on the relevances of the other items that were presented with it. We collected a data set of biased and unbiased choices in a controlled online user study and measured the effects of overexposure and competition. We found that (i) the discrete choice models effectively mitigated exposure bias on human-generated choice data, (ii) there were no significant differences in robustness among the different discrete choice models, and (iii) only multivariate discrete choice models were robust to competition between items. We conclude that discrete choice models mitigate exposure bias effectively because they consider item-co-exposure. Moreover, exposing items alongside more or less popular items can bias future recommendations significantly and item exposure must be tracked for overcoming exposure bias. We consider our work vital for understanding what exposure bias it, how it forms, and how it can be mitigated., Comment: Accepted at the CONSEQUENCES '24 workshop, co-located with ACM RecSys '24

Recent claims of achieving exponential quantum advantage have attracted attention to Gaussian boson sampling (GBS), a potential application of which is dense subgraph finding. We investigate the effects of sources of error including loss and spectral impurity on GBS applied to dense subgraph finding algorithms. We find that the effectiveness of these algorithms is remarkably robust to errors, to such an extent that there exist efficient classical algorithms that can simulate the underlying GBS. These results imply that the speedup of GBS-based algorithms for the dense subgraph problem over classical approaches is at most polynomial, though this could be achieved on a quantum device with dramatically less stringent requirements on loss and photon purity than general GBS., Comment: 10 pages, 7 figures

This project brings music to sight. Music can be a visual masterpiece. Some people naturally experience a visualization of audio - a condition called synesthesia. The type of synesthesia explored is when sounds create colors in the 'mind's eye.' Project included interviews with people who experience synesthesia, examination of prior art, and topic research to inform project design. Audio input, digital signal processing (including Fast Fourier Transforms (FFTs)) and data manipulation produce arguments required for our visualization. Arguments are then applied to a physics particle simulator which is re-purposed to model a synesthetic experience. The result of the project is a simulator in MAX 8, which generates a visual performance using particles by varying each particle's position, velocity, and color based on parameters extracted via digital processing of input audio., Comment: 7 pages, 2 figures

The challenging requirements of large scale quantum information processing using parametric heralded single photon sources involves maximising the interference visibility whilst maintaining an acceptable photon generation rate. By developing a general theoretical framework that allows us to include large numbers of spatial and spectral modes together with linear and non-linear optical elements, we investigate the combined effects of spectral and photon number impurity on the measured Hong--Ou--Mandel interference visibility of parametric photon sources, considering both threshold and number resolving detectors, together with the effects of spectral filtering. We find that for any degree of spectral impurity, increasing the photon generation rate necessarily decreases the interference visibility, even when using number resolving detection. While tight spectral filtering can be used to enforce spectral purity and increased interference visibility at low powers, we find that the induced photon number impurity results in a decreasing interference visibility and heralding efficiency with pump power, while the maximum generation rate is also reduced., Comment: 24 pages, 10 figures

We propose to learn invariant representations, in the data domain, to achieve interpretability in algorithmic fairness. Invariance implies a selectivity for high level, relevant correlations w.r.t. class label annotations, and a robustness to irrelevant correlations with protected characteristics such as race or gender. We introduce a non-trivial setup in which the training set exhibits a strong bias such that class label annotations are irrelevant and spurious correlations cannot be distinguished. To address this problem, we introduce an adversarially trained model with a null-sampling procedure to produce invariant representations in the data domain. To enable disentanglement, a partially-labelled representative set is used. By placing the representations into the data domain, the changes made by the model are easily examinable by human auditors. We show the effectiveness of our method on both image and tabular datasets: Coloured MNIST, the CelebA and the Adult dataset., Comment: Published as a conference paper at the 16th European Conference on Computer Vision (ECCV), Glasgow, UK, 2020

We prove a duality result for the analytic cohomology of Lie groups over non-archimedean fields acting on locally convex vector spaces.

We prove standard results of group cohomology -- namely, existence of a long exact sequence, classification of torsors via the first cohomology group, Shapiro's lemma, the Hochschild-Serre spectral sequence, a decomposition of the cochain complex in the direct product case, and Jannsen's result on the recovery problem -- for cohomology theories such as continuous, analytic, bounded, and pro-analytic cohomology. We also prove these results for certain monoids. The cohomology groups considered here all have very concrete interpretations by means of a cochain complex. Therefore, we do not use methods of homological algebra, but explicit calculations on the level of cochains, using techniques dating back to Hochschild and Serre.

We perform $N$-body simulations of the early phases of open cluster evolution including a large population of planetesimals, initially arranged in Kuiper-belt like discs around each star. Using a new, 4th-order and time-reversible $N$-body code on Graphics Processing Units (GPUs), we evolve the whole system under the stellar gravity, i.e. treating planetesimals as test particles, and consider two types of initial cluster models, similar to IC348 and the Hyades, respectively. In both cases, planetesimals can be dynamically excited, transferred between stars or liberated to become free-floating (such as A/2017 U1 or 'Oumuamua) during the early cluster evolution. We find that planetesimals captured from another star are not necessarily dynamically distinct from those native to a star. After an encounter both native and captured planetesimals can exhibit aligned periastrons, qualitatively similar to that seen in the Solar system and commonly thought to be the signature of Planet 9. We discuss the implications of our results for both our Solar system and exoplanetary systems., Comment: 15 pages, 7 figures, 4 tables, accepted for publication in MNRAS

Interpretability and fairness are critical in computer vision and machine learning applications, in particular when dealing with human outcomes, e.g. inviting or not inviting for a job interview based on application materials that may include photographs. One promising direction to achieve fairness is by learning data representations that remove the semantics of protected characteristics, and are therefore able to mitigate unfair outcomes. All available models however learn latent embeddings which comes at the cost of being uninterpretable. We propose to cast this problem as data-to-data translation, i.e. learning a mapping from an input domain to a fair target domain, where a fairness definition is being enforced. Here the data domain can be images, or any tabular data representation. This task would be straightforward if we had fair target data available, but this is not the case. To overcome this, we learn a highly unconstrained mapping by exploiting statistics of residuals - the difference between input data and its translated version - and the protected characteristics. When applied to the CelebA dataset of face images with gender attribute as the protected characteristic, our model enforces equality of opportunity by adjusting the eyes and lips regions. Intriguingly, on the same dataset we arrive at similar conclusions when using semantic attribute representations of images for translation. On face images of the recent DiF dataset, with the same gender attribute, our method adjusts nose regions. In the Adult income dataset, also with protected gender attribute, our model achieves equality of opportunity by, among others, obfuscating the wife and husband relationship. Analyzing those systematic changes will allow us to scrutinize the interplay of fairness criterion, chosen protected characteristics, and prediction performance., Comment: IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2019