WP3: Towards explainable and fair AI-generated decisions

Research • Lead: 14-BFH • M4–M48 • 300 PM

WP 3 is led by BFH and supported by all partners. The work is divided into the following tasks.

Objectives

The WP will work towards a unifying framework of explainability for AI models applied to financial use cases.
O 3.1. To answer the main research questions on solving explainability deployment hurdles for financial applications.
O 3.2. To demonstrate the proposed framework for audience-dependent explanations, through use cases (SWE, UNA, BFH).
O 3.3. To disseminate the knowledge, validated by an international research centre (FRA and ARC) or ECB.

Tasks (6)

Task	Name	Description
T3.1	Technical coordination	Monitoring the related IRPs, store the output generated in a location accessible to the entire network
T3.2	Research training	Support the research training for all assigned DCs and contribute to advanced training content
T3.3	Post hoc explainability	Provide global and local post hoc explainability techniques that address the explainability needs of different stakeholders
T3.4	Explainability functions	Propose explainability functions, tailored for financial time series, preserving the non-stationary dependence structure
T3.5	Portfolio optimization	Develop new portfolio optimization models that address challenges of incorporating fairness considerations into investments
T3.6	Dissemination	Disseminate, communicate and exploit the results (Conferences, OS Day, policy paper, two prototypes, use case, media coverage)

Doctoral Candidates (4)

DC1: Strengthening European financial service providers through applicable reinforcement learning

Host: UTW • PhD: UTW

DC9: Audience-dependent explanations

Host: UNA • PhD: UNA

DC16: Investigating the utility of classical XAI methods in financial time series

Host: BFH • PhD: UTW

DC17: Fair Algorithmic Design and Portfolio Optimization under Sustainability Concerns

Host: BFH • PhD: UTW

Deliverables (3)

Code	Name	Type	Diss.	Due	Description
D3.3	Summary report on time-series explainability	R	SEN	M24	Summary report on all results and impacts related to explainability for time-series
D3.1	Documentation of explainable AI methods	R	SEN	M48	Documentation of test setups for applying explainable AI methods
D3.2	Technical report on trustworthy AI methods	R	SEN	M48	Technical report showing the achievements on trustworthy and fair AI models

Effort Breakdown

Participant	Person-Months
1-UTW	6
2-WWU	6
4-UNA	54
5-KUT	6
6-ASE	6
7-BBU	6
8-CAR	6
9-RAI	6
10-SWE	6
13-ROY	6
14-BFH	144
15-ARC	6
16-EIT	6
17-FRA	6
18-ECB	6
19-POZ	6
20-DBA	6
21-UKL	6
22-BIS	6
Total	300