发帖
雷达卡
Deep Learning in Banking Integrating Artificial Intelligence for Next-Generation.epub
(9.3 MB, 需要: RMB 13 元)
内容新，2025年新上线的资料包
内容丰富，390多页的大型资料包
内容精彩，epub格式，方便阅读翻译，全部矢量文字，适合翻译学习！Acknowledgments
Chapter 1: Introduction
1.1 WHY WE WROTE THIS resource AND WHO IT IS FOR
1.2 DO YOU NEED DEEP LEARNING?
1.3 resource STRUCTURE
1.4 SOME RELEVANT REGULATORY FRAMEWORKS
1.5 TECHNICAL BACKGROUND
NOTES
Chapter 2: Image Processing and Convolutional Neural Networks
2.1 IMAGE DATA IN BANKING
2.2 REGULATORY CONSTRAINTS
2.3 DEEP LEARNING FOR IMAGE PROCESSING: CONVOLUTIONAL NEURAL NETWORKS
2.4 TRAINING CNNs
2.5 CNN APPLICATIONS IN BANKING AND FINANCE
2.6 CASE STUDY: LIDAR-BASED MORTGAGE DEFAULT RATE PREDICTION
NOTE
Chapter 3: Time Series and Recurrent Models
3.1 TIME SERIES AND PANEL DATA IN BANKING
3.2 REGULATORY ASPECTS OF TIME SERIES AND TABULAR DATA
3.3 DEEP LEARNING TIME SERIES MODELS
3.4 TIME SERIES APPLICATIONS IN BANKING AND FINANCE
3.5 CASE STUDY: BEHAVIORAL SCORING USING DEEP LEARNING
Chapter 4: Text Data and Transformers
4.1 TEXT DATA IN BANKING
4.2 REGULATORY AND ETHICAL CONCERNS
4.3 RECURRENCY AND ATTENTION FOR TEXT ANALYTICS
4.4 MULTI-HEAD ATTENTION
4.5 TRANSFORMERS
4.6 TRANSFORMER-BASED ARCHITECTURES FOR TEXT: FROM BERT AND BEYOND
4.7 FINE-TUNING OF PRE-TRAINED MODELS
4.8 TRAINING A TEXT MODEL FROM SCRATCH
4.9 TEXT-BASED APPLICATIONS IN BANKING AND FINANCE
4.10 CASE STUDY: RECESSION CHANCES ARISING FROM FED SPEECHES USING BERT
Chapter 5: Financial Contagion and Network Models
5.1 NETWORK DATA IN BANKING
5.2 REGULATORY CONCERNS OF USING NETWORK DATA
5.3 NETWORK FEATURES: CENTRALITY AND NEIGHBORHOOD
5.4 GRAPH NEURAL NETWORKS
5.5 NETWORK APPLICATIONS IN BANKING AND FINANCE
5.6 CASE STUDY: DEFAULT CORRELATION IN MORTGAGE LENDING
NOTES
Chapter 6: Generative AI and Large Language Models
6.1 LARGE LANGUAGE MODELS IN BANKING
6.2 REGULATORY FOCUS ON LARGE LANGUAGE MODELS
6.3 LLM THEORY
6.4 ALIGNMENT AND REINFORCEMENT LEARNING FROM HUMAN FEEDBACK (RLHF)
6.5 LLMs AS AGENTS, NOT JUST PREDICTORS
6.6 AGENTIC AI
6.7 LICENSING ISSUES AND DATA USE
6.8 LLM APPLICATIONS IN BANKING AND FINANCE
6.9 CASE STUDY: COMPARISON OF PROMPT ENGINEERING AND FINE-TUNING FOR A CREDIT SCORING EXPLAINER
NOTES
Chapter 7: Multimodal Data and Information Fusion
7.1 AI MULTIMODAL MODELS
7.2 CREATING A MULTIMODAL MODEL
7.3 SOME SPECIFIC CHALLENGES IN THE FINANCIAL SECTOR
7.4 OPERATIONAL IMPLICATIONS
7.5 CASE STUDY: A MULTIMODAL MORTGAGE MODEL
Chapter 8: Fairness, Accountability, Explainability, and Causality
8.1 FAIRNESS
8.2 ACCOUNTABILITY
8.3 EXPLAINABILITY
8.4 EXPLAINABILITY VERSUS CAUSALITY
8.5 CAUSAL DEEP LEARNING
8.6 CASE STUDY: FAIRNESS AND EXPLAINABILITY
NOTE
Chapter 9: Perspectives on the Future of AI in Banking
9.1 ROI ANALYSIS AND PROJECT PRIORITIZATION
9.2 DEPLOYING DEEP LEARNING MODELS
9.3 FUTURE GLOBAL PERSPECTIVES
9.4 CLOSING WORDS
Bibliography
For most of our careers, we have been fortunate to have built data science and, since the last decade, deep learning models across multiple fields. Although Cristián works mainly with financial institutions and regulators, María and Sebastián have been working on deploying varied AI solutions, in topics such as sleep research, customer service management, telecommunications, gaming, and a long list of applications. Banking was, however, the intersection of our work, which has led to many joint projects and publications for most of our careers. One thing that was common throughout these collaborations was that we needed a unified resource to direct our business partners and students towards. What we wanted was a one-stop shop, a single resource that discussed the challenges of working with unstructured data in banking and financial institutions, both in practical and regulatory terms, while also providing the theoretical background about the models and showed how to train these models with practical exercises.
During January 2024, coincidentally, the three of us met at the Faculty of Business and Economics at the University of Chile. Sebastián was working with María and Cristián and had been invited by Jaime Miranda, the director of the School of Information Systems and Auditing, to teach a short course on, you guessed it, deep learning in banking. It was there that this idea was born. Cristián had been mulling over the idea that there was no textresource for this and wanted to write it during his sabbatical. The problem was clear: A user interested in deep learning in banking would need to choose one of the many resources on deep learning, search for relevant papers on AI in banking, read the Basel regulation and its local implementations, and look for specific regulatory documents discussing the use of unstructured data by the likes of the Fed, the European Central Bank, and other institutions. There was no single resource available, so we may as well write one. The visit to Chile served as the perfect place to bring everyone on board. After planning the resource for a few months, consulting with colleagues on its contents across multiple countries, and pitching it to Wiley, the stage was set. This resource would become a reality.
Our wish for this work is for it to be a reference resource within financial institutions and schools, particularly in those areas and programs focused on data-driven risk management. As banking is arguably the most regulated use of data science worldwide, it makes sense to have one resource that can point the reader to the challenges that are faced in our industry. The nuances of deploying deep learning models in banking are varied. Some uses strictly regulate the use of alternative data, such as images (Chapter 2) or social network data (Chapter 5) in consumer credit scoring. Some encourage it, such as text data (Chapter 4) in anti-money laundering. Knowing when these data sources appear and what are the challenges to use them in banking will, we believe, accelerate the adoption of new technologies and the responsible development and deployment of AI.
With that in mind, throughout the resource, we used AI ourselves in a few ways. First, we used it to help us with spell and grammar checking. We used Grammarly, Writefull, and GPT-4o for these purposes. Second, we used Gemini’s and OpenAI’s Deep Research to help us search the web and summarize literature during our literature review, expanding the initial one we conducted ourselves and find blind spots. We also used Gemini AI to turn the resource code we created, which you can find in the resource’s website,1to the language-agnostic algorithms that accompany the case studies, and to turn some equations into LaTeX code. Finally, Figure 9.1was drawn from AI-generated Mermaid code from cruder diagrams we originally created. All AI-generated content was carefully reviewed, edited, and approved. At no point in this work did we use AI to generate any insights or analyses of any kind. All analyses, conclusions, and interpretations represent our views and expertise.
Finally, we are optimistic in believing that the use of responsible AI in banking can bring about innovation and growth. With this in mind, this resource is intended to provide the tools to navigate the sea of new AI deployments, without falling into the pitfalls of its development. It also provides warnings where the current state of the art is simply not there yet to ensure the safe deployment of AI, and highlights best practices for ethical implementation (Chapter 8). There is no putting the genie back in the bottle; AI will be a key part of banking practice. Our aim in writing this resource was to create a resource to support you, the reader, in creating solutions that are safe, profitable, and productive. We hope you think the same after reading these pages.
京公网安备 11010802022788号
