当前位置:首页 > 报告详情

金融系统的下一个前沿:基于Transformer的基础模型(由NVIDIA赞助).pdf

上传人: 明**** 编号:1013036 2025-12-21 28页 647.39KB

1、 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.A I M 3 3 1 7-SThe Next Frontier in Financial Systems:Transformer-based Foundation ModelsPahal PatangiaSudhir KalidindiHe/HimHead of Global Industry Business Develop

2、ment,PaymentsNVIDIAHe/HimPrincipal,Solutions ArchitectAWS 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.AgendaWhy transformers for payments Foundation models for Transactional DataPersonalization&pattern miningCloud AI Factory on AWS+NVIDIA accelerationKey takeawaysCall to actio

3、n 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Why payments need next-generation AI Exploding Volume and Complexity Sophisticated Attack surfacesLow latency without false negatives 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.o

4、r its affiliates.All rights reserved.Why transformers for payments?2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.How they work?Look at all data points at onceUse self-attentionLearn contextWhy they are powerful?Process all at onceCapture full contextLearn their own representatio

5、nsKey Building BlocksEmbeddingsAttention-LayersPretrainIn PaymentsModel transaction sequences like sentencesDetect anomalies-fraud ringsFraud detection,personalizationTransformers are deep learning models that understand relationships within sequencesWhat are Transformers?The engine behind Modern AI

6、 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Tabular Machine Learning in Financial ServicesMany workflows are dominated by decision tree-based modelsNameDateAmountMerchant.FraudAdam2/18/2025$50Online Retail.NoBecky2/20/2025$35Pet Shop.NoCharlie2/21/2025$60TelecomNo.Adam2/28/20

word格式文档无特别注明外均可编辑修改,预览文件经过压缩,下载原文更清晰!
三个皮匠报告文库所有资源均是客户上传分享,仅供网友学习交流,未经上传用户书面授权,请勿作商用。
根据《The Next Frontier in Financial Systems: Transformer-based Foundation Models》的内容,以下是全文关键点的概括: 1. **支付领域AI需求**:支付领域正面临数据量爆炸、复杂性增加和攻击面复杂化等问题,需要下一代AI技术。 2. **Transformer模型优势**:Transformer模型能够一次性处理所有数据点,使用自注意力机制学习上下文,捕捉完整信息,学习自身表示。 3. **Transformer在支付领域的应用**:模型可以处理交易序列,检测异常(如欺诈团伙),进行欺诈检测、个性化推荐。 4. **Transformer架构**:其架构可大规模并行化,适合GPU处理。 5. **Tabular数据模型**:从Transformer到Tabular模型,通过预训练策略处理交易数据。 6. **云AI工厂**:AWS和NVIDIA合作提供可扩展的云平台,加速模型训练和推理。 7. **成功案例**:大型银行和金融科技公司通过使用Transformer模型提高了欺诈检测准确率。 8. **关键成果**:Transformer模型将交易转换为嵌入,学习复杂模式,为欺诈检测、身份验证和个性化提供更深入的理解。
Transformer如何助力?" Transformer模型揭秘" Transformer在金融领域的应用"
客服
商务合作
小程序
服务号
折叠