Artificial Intelligence and Machine Learning Model-Agnostic Interpretability and Explain-ability - QC-175

Preferred Disciplines: Computer Science, Risk Management & Financial Mathematics, Mathematics & Statistics (PhD, Master or Post-Doc may also be considered)
Project length: TBD
Approx. start date: Winter 2019
Location: Montreal, QC
No. of Positions: 5
Preferences: None
Company: Societe Generale, Montreal Solution Center (MSC)

About Company:

Societe Generale Corporate & Investment Banking (SG CIB) ranks among the world leaders in the areas of investment banking, global finance and global markets. In Canada SG CIB is present in Montreal, Toronto and Calgary and has been serving Canadian businesses for more than 40 years. Our goal is to be the international reference bank for our Canadian clients in our areas of expertise: financial institutions, multinationals and governments, infrastructure, energy and utilities, oil and gas, mining and agricultural commodities.

SG Montreal Solution Center (SG MSC) is part of Groupe SG and its mandate is to provide local services in support of Societe Generale Group's banking and investment banking activities (SG CIB). SG MSC provides support, development and infrastructure services across multiple services.

Summary of Project:

Société Générale is currently expanding its R&D activities by supporting the development of an AI hub in Montreal, Canada. This hub will become part of Société Générale’s Montreal Solution Center (MSC) whose role is to provide development and infrastructure services across multiple services in Canada and the world.

One of the first mandate of this hub will be to support the Model Risk Management Americas group within their use of artificial intelligence approaches. The company would like to better understand AI models currently available and/or in use in risk management and their implication considering existing regulatory obligations within the financial sector.

As part of this project, Société Générale would also like to set the basis for long-term collaborations with the academic sector. As such, Société Générale would like, as a start, to co-fund, along with Mitacs, the academic research of 5 PhD students (open to Master and post-doc as well) on this topic with the potential of adding more researchers to this project over the years.

Research Objectives/Sub-Objectives:

The project’s main goal is to establish and implement a model-agnostic protocol that will enable independent review and validation of AI and ML Models used in the Bank. The protocol will also include the design and implementation of Ongoing Monitoring Framework for these types of models. 

Different tasks of the project include: study and research the latest literature in the topic of Explaining and Interpreting Predictions made using Machine Learning. Application of the protocol and the techniques identified and explored during the project into different actual cases in the Bank (different models developed in the entire Americas Region). The application will also include the implementation of the Ongoing Monitoring Framework and Performance Metrics.

The project also encompasses a strong aspect of research as exhaustive studies must be conducted regarding the development of Machine Learning, traditional statistical models for prediction and benchmarking.

Methodology:

    • Investigate different literature in the topic of Interpretability in AI and ML Models
    • Investigate industry-wide research in the topic and create a compendium/library of case studies.
    • Design, develop and implement a framework and protocol for the interpretation and explanation of any AI and ML model
    • Create Proof-of-Concept of the protocol for one specific ML model developed internally by the Bank in the Region
    • Design a Business Requirement Document for the Implementation of the Framework and Protocol in the Models in the entire Region. 
    • Identify model types and architectures requiring evaluation & analysis as scoped by discussions with domain experts (risk management officers) and researchers
    • Research performance indicators and explain-ability metrics that can be applied to the model at hand
    • Build POC (proof-of-concept) model evaluation framework
    • Propose solution to domain experts and determine the effectiveness of the solution from their perspective
    • Go/No-go decision given by domain experts
    • Enhance solution based on feedback from domain experts
    • Deploy solution in global model-agnostic tool

    Expertise and Skills Needed:

    Required Skills:

    • Programming skills: R, Python (and the libraries)
    • Machine Learning and Artificial Intelligence techniques: classification, regression, clustering, neural networks, support vector machines, decisions trees, dimensionality reduction, performance estimation
    • Experience with handling of large datasets (SQL, CSV)
    • Mathematical and statistical modeling: regression, general linear models, stochastic processes, time series analysis

    Context (Technologies used in the company):

    • Python, R
    • Scikit-Learn
    • Tensorflow
    • Pandas & Numpy
    • H2O AutoML
    • Alteryx
    • Tableau

    For more info or to apply to this applied research position, please

    1. Check your eligibility and find more information about open projects.
    2. Interested students need to get the approval from their supervisor and send their CV along with a link to their supervisor’s university webpage by applying through the webform.
    Program: