Herding for Structured Prediction

Chen, Y.; Gelfand, Andrew E.; Welling, Max

doi:10.7551/mitpress/9969.003.0010

Cited by 2 publications

(2 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The parameter s can be tuned by users, allowing a trade-off between information content and computational complexity. scLKME offer three possible sketching methods, including, geometric sketching [22], kernel herding [23,24], and simple random sampling without replacement.…”

Section: Cell Sketchingmentioning

confidence: 99%

scLKME: A Landmark-based Approach for Generating Multi-cellular Sample Embeddings from Single-cell Data

Yi,

Stanley

2023

Preprint

View full text Add to dashboard Cite

Single-cell technologies enable high-dimensional profiling of individual cells, therefore offering profound insights into subtle variation between specialized cell-types. However, translating the multitude of nuanced cellular profiles into meaningful per-sample representations is challenging due to heterogeneous cellular composition across individual profiled samples. To compute informative per-sample representations, we developed scLKME, a novel approach that uses a landmark-based kernel mean embedding method to convert multi-sample single-cell data into compact per-sample embeddings. Treating each sample as a distribution over cells, scLKME identifies landmarks across samples and maps these distributions into a reproducing kernel Hilbert space. Overall, scLKME outperforms state-of-the-art techniques in robustness, efficiency, accuracy, and practical usefulness of sample embeddings. Its application on a CyTOF dataset profiling immune responses in preterm birth highlighted its capacity to accurately identify patient-specific variations correlating with gestational age, suggesting broad applicability to multi-sample single-cell datasets with complex experimental designs. scLKME is available as an open-sourced python package athttps://github.com/CompCy-lab/scLKME.

show abstract

Section: Cell Sketchingmentioning

confidence: 99%

scLKME: A Landmark-based Approach for Generating Multi-cellular Sample Embeddings from Single-cell Data

Yi,

Stanley

2023

Preprint

View full text Add to dashboard Cite

show abstract

“…The exploration is therefore performed within this discrete database made of the 90 000 viable geometries. The DOE is built as a selection of 150 guide fins within this database using the Kernel Herding Algorithm as proposed by [23]. The refinement sets are selected within this database.…”

Section: Optimization Processmentioning

confidence: 99%

Beneficial Effects of Guide Fins on Corner Separation

Mondin,

Riéra,

Duquesne

et al. 2023

Volume 13A: Turbomachinery — Axial Flow Fan and Compressor Aerodynamics

View full text Add to dashboard Cite

Corner separation is a major phenomenon limiting the operability of aeronautical compressors. Passive control devices such as guide fins are envisioned to reduce its deleterious effects. In this paper, guide fins are optimized to reduce the total pressure losses downstream of a subsonic compressor cascade (M = 0.11, Re = 3.2 105, modern stator blades) at design and near-stall incidences. Guide fins shapes are defined using 15 parameters. An optimization strategy relying on the iterative refinement of surrogate model with Reynolds-Averaged Navier-Stokes (RANS) computations is carried out. A database of 1033 guide fins is obtained, a large number of which strongly reduces the near-stall losses. Three guide fins are thoroughly investigated and referred to as Short Fence (short and straight guide fin), Long Fence (long and straight guide fin) and 3D (long guide fin with a strong pyramidal aspect). Their effect on the downstream measurement plane is validated experimentally, as well as their integrated performances. Large reductions of total pressure losses at near-stall incidence are obtained (−2.4 pt to −2.9 pt, or a relative reduction of −50% to −60%) without degrading neither the nominal losses nor the stator deflection. These guide fins are used to highlight three beneficial mechanisms that alter the corner separation development within the blade passage a) Tip Vortex (energizes the passage flow) b) Fence (induces 2 passage vortices) c) Guide (induces a local favourable pressure gradient). The role of these mechanisms in the generation of total pressure losses are analyzed in depth with streamlines visualisations and with a topological analysis. Kriging surrogate models reveal that three parameters drive the guide fins performances : the axial position, height, and thickness value at hub. The performances of the investigated guide fins are robust to significant variations of these parameters, as well as small variation of the inlet boundary layer thickness.

show abstract

Herding for Structured Prediction

Cited by 2 publications

References 12 publications

scLKME: A Landmark-based Approach for Generating Multi-cellular Sample Embeddings from Single-cell Data

scLKME: A Landmark-based Approach for Generating Multi-cellular Sample Embeddings from Single-cell Data

Beneficial Effects of Guide Fins on Corner Separation

Contact Info

Product

Resources

About