tabular foundation model
PulseAugur coverage of tabular foundation model — every cluster mentioning tabular foundation model across labs, papers, and developer communities, ranked by signal.
2 天有情绪数据
-
新方法在无需微调的情况下实现最先进的多模态分类
研究人员开发了CoMET，一种利用冻结的预训练骨干网络和表格基础模型（TFMs）的多模态分类新方法。该方法使用主成分分析（PCA）压缩模态嵌入，然后将其输入TFM，从而无需微调。为了提高表示质量，尤其是在CLS标记未对齐的情况下，他们提出了PALPooling，一种自适应标记池化器。CoMET在各种多模态基准测试中取得了最先进的成果，并且可以在没有进行任何训练的情况下处理包含超过50万个样本和2000个类别的大规模数据集。
-
New benchmark GHGbench tackles fragmented carbon emission prediction
Researchers have introduced GHGbench, a new benchmark and dataset designed to unify and improve the prediction of carbon emissions at both company and building levels. The benchmark addresses fragmentation in existing d…
-
Tabular foundation models improve landslide prediction with scarce data
Two new research papers propose novel approaches to landslide susceptibility prediction using tabular foundation models, addressing the common issue of data scarcity and imbalance. The first paper introduces a generativ…