Explainable AI for Financial Services

Centelon develops AI based decision systems for financial services. One of the key regulatory requirements is to explain the model to the authorities. We have deployed multiple methods such as Local Interpretable Model Agnostic Explanations (LIME) and Game theory based Shapley Values.

In the talk, I would take the audience through the business case, mathematics behind various explanatory models, design considerations and code demonstration

 
 

Outline/Structure of the Workshop

1. Need for explainable AI in Financial Services

2. Two Approaches: Node explanations versus Feature Importance

3. Model specific explanation methods in NLP

4. Model Agnostic Methods a) Drop-one-feature method and its limitation

5. Model Agnostic Methods b) Theoretical background of LIME and its limitations

6. Model Agnostic Methods c) Theoretical background of Shapley Values

7. Equivalence conditions for Shapley Values and LIME

8. Code Demo: ( Financial Services use case, None Financial Services use case)

Learning Outcome

The audience will begin to appreciate the need for building explainable systems. They would learn the basic tool set to deploy in various use cases.

Target Audience

Developers, Executives

Prerequisites for Attendees

Basic Machine Learning, Deep Learning

schedule Submitted 7 months ago

Public Feedback

comment Suggest improvements to the Speaker
  • Dr. Vikas Agrawal
    By Dr. Vikas Agrawal  ~  6 months ago
    reply Reply

    Dear Prabhash: I was trying to locate your LinkedIn profile and your bio on confengine.com. Could you please add those?

    We have two other proposals submitted on a very similar topic and so we are considering those as well.

    Warm Regards

    Vikas

    • Prabhash Thakur
      By Prabhash Thakur  ~  6 months ago
      reply Reply
      Hi Vikas - Thank you for your email. We recently connected on LinkedIn. My linkedin profile isĀ www.linkedin.com/in/prabhash-thakur. I will update it on confengine as well.

      Best Regards,
      Prabhash Thakur

      --000000000000a5fff6058a142f62--