LeituraPesquisaRadarFramework de investimento
Entrar / Cadastrar
← Voltar para a leitura
TópicoEditorial DisconfirmAIAtualizado 2026-11-13 19:45

Mitra-v2

2 etapas1 peças3 Fontes

亚马逊科学团队发布表格基础模型Mitra-v2,其核心突破在于通过扩大合成预训练数据的规模与多样性来提升性能,而非加深网络架构。该模型在TabArena基准测试的51个真实数据集上领先单模型赛道,并在ScoringBench回归基准中排名第一。【背景】Mitra-v2沿用了上一代的12层Transformer架构,但将预训练表的行数扩大10倍、列数约3倍,并引入新型混合因果模型先验,使单一合成表能模拟真实数据中复杂的列间关系。尽管被归类为基础模型,Mitra-v2并非零样本(zero-shot)系统:默认情况下,它会在每张新表上进行短暂的自动微调,并对8个副本的预测结果进行平均(bagging)。这一机制虽增加了计算开销,但显著提升了准确率。目前,Amazon Connect Customer已采用该模型处理结构化业务与客户体验数据。

Como chegamos aqui

  1. 2026-09-03

    亚马逊发布《Mitra-v2技术报告》(arXiv:2609.04540),详细阐述模型架构、合成预训练策略及在TabArena和TALENT基准上的评估结果。

    arxiv.org ↗
  2. 2026-11-10

    亚马逊科学博客由Chris (Luyang) Kong发文介绍Mitra-v2,宣布其在TabArena单模型赛道领先,并开源Apache-2.0许可的代码与权重。

    amazon.science ↗

O que ainda está indefinido

  • Mitra-v2在大规模生产环境中的实际推理延迟与成本效益相比传统梯度提升树的具体表现如何?
  • 除了Amazon Connect Customer,其他行业客户对Mitra-v2的独立采纳案例及其带来的具体业务指标提升是否有公开数据支持?

Fontes

  • amazon.scienceAmazon Science Blog↗
Entrar / Cadastrar
LeituraPesquisaRadarFramework de investimento
arxiv.orgMitra-v2 Technical Report (arXiv)↗
  • huggingface.coHugging Face Model Card (Regressor)↗
  • Cobertura neste tópico

    • Amazon Mitra-v2: Expanding Synthetic Pretraining Diversity to Replace Network Depth, Self-Reported Leading TabArena2026-11-11Notícias

      The Amazon Science team has released the tabular foundation model Mitra-v2. Its architecture remains identical to its predecessor; performance gains stem from scaling up synthetic pretraining data (10x rows, ~3x columns) and a novel hybrid causal model prior. This prior combines neural networks, decision trees, Gaussian processes, and autoregressive mechanisms within a single causal graph, allowing one synthetic table to carry mixed inter-column relationships found in real tables. Self-reported results show it leads the single-model track on 51 datasets in TabArena and ranks first on the ScoringBench regression benchmark.