paper-with-me

홈 › Papers

How Good is a Single Basin?

2024-02-05 · Kai Lion, Lorenzo Noci, Thomas Hofmann, Gregor Bachmann

The multi-modal nature of neural loss landscapes is often considered to be the main driver behind the empirical success of deep ensembles. In this work, we probe this belief by constructing various "connected" ensembles which are restricted to lie in the same basin. Through our experiments, we demonstrate that increased connectivity indeed negatively impacts performance. However, when incorporating the knowledge from other basins implicitly through distillation, we show that the gap in performance can be mitigated by re-discovering (multi-basin) deep ensembles within a single basin. Thus, we conjecture that while the extra-basin knowledge is at least partially present in any given basin, it cannot be easily harnessed without learning it from other basins.

📄 PDF Abstract BibTeX arXiv:2402.03187

Code (0)

등록된 구현이 없습니다.

Methods 이 논문이 사용한 방법론

Deep Ensembles 설명 없음

Similar Papers 제목 키워드 기반

A Performance Analysis of Basin Hopping Compared to Established Metaheuristics for Global Optimization

2024-03-09 · Marco Baioletti, Valentino Santucci, Marco Tomassini

During the last decades many metaheuristics for global numerical optimization have been proposed. Among them, Basin Hopping is very simple and straightforward to implement, although rarely used outside its original Physi…

global-optimization

Image Segmentation by Size-Dependent Single Linkage Clustering of a Watershed Basin Graph

2015-05-01 · Aleksandar Zlateski, H. Sebastian Seung

We present a method for hierarchical image segmentation that defines a disaffinity graph on the image, over-segments it into watershed basins, defines a new graph on the basins, and then merges basins with a modified, si…

ClusteringImage SegmentationSemantic Segmentation

WASH: Train your Ensemble with Communication-Efficient Weight Shuffling, then Average

2024-05-27 · Louis Fournier, Adel Nabli, Masih Aminbeidokhti, Marco Pedersoli 외

The performance of deep neural networks is enhanced by ensemble methods, which average the output of several models. However, this comes at an increased cost at inference. Weight averaging methods aim at balancing the ge…

image-classificationImage Classification

To Stay or Not to Stay in the Pre-train Basin: Insights on Ensembling in Transfer Learning

2023-03-06 · NeurIPS 2023 11 · Ildus Sadrtdinov, Dmitrii Pozdeev, Dmitry Vetrov, Ekaterina Lobacheva

Transfer learning and ensembling are two popular techniques for improving the performance and robustness of neural networks. Due to the high cost of pre-training, ensembles of models fine-tuned from a single pre-trained …

DiversityTransfer Learning

Explicitly Multi-Modal Benchmarks for Multi-Objective Optimization

2021-10-07 · Ryosuke Ota, Reiya Hagiwara, Naoki Hamada, Likun Liu 외

In multi-objective optimization, designing good benchmark problems is an important issue for improving solvers. Controlling the global location of Pareto optima in existing benchmark problems has been problematic, and it…

Benchmarking