Mathematical Finance & Financial Data Science Seminar
AI and machine learning are entering every aspect of our life. Marketing, autonomous driving, personalization, computer vision, finance, wearables, travel are all benefiting from the advances in AI in the last decade. As more and more AI applications are being deployed in enterprises, concerns are growing about potential "AI accidents" and the misuse of AI. With increased complexity, some are questioning whether the models actually work! As the debate about fairness, bias, and privacy grow, there is increased attention to understanding how the models work and whether the models are thoroughly tested and designed to address potential issues.
The area "Responsible AI" is fast emerging and becoming an important aspect of the adoption of machine learning and AI products in the enterprise. Companies are now incorporating formal ethics reviews, model validation exercises, and independent algorithmic auditing to ensure that the adoption of AI is transparent and has gone through formal validation phases.
In this talk, Sri will introduce Algorithmic auditing and discuss why Algorithmic auditing will be a formal process industries using AI will need. Sri will also discuss the emerging risks in the adoption of AI and discuss how QuSandbox, his company is building, will address the emerging needs of formal Algorithmic auditing practices in enterprises.
50 ĐỀ LUYỆN THI IOE LỚP 9 - NĂM HỌC 2022-2023 (CÓ LINK HÌNH, FILE AUDIO VÀ ĐÁ...
Algorithmic auditing 1.0
1. Pragmatic
Algorithmic Auditing 1.0
2021 Copyright QuantUniversity LLC.
Presented By:
Sri Krishnamurthy, CFA, CAP
sri@quantuniversity.com
www.quantuniversity.com
NYU Mathematical Finance
and Data Science Seminar
April 20, 2021
2. 2
Speaker bio
• Advisory and Consultancy for Financial
Analytics
• Prior Experience at MathWorks, Citigroup
and Endeca and 25+ financial services and
energy customers.
• Columnist for the Wilmott Magazine
• Author of forthcoming book
“Pragmatic AI and ML in Finance”
• Teaches AI/ML and Fintech Related topics in
the MS and MBA programs at Northeastern
University, Boston
• Reviewer: Journal of Asset Management
Sri Krishnamurthy
Founder and CEO
QuantUniversity
3. 3
QuantUniversity
• Boston-based Data Science, Quant
Finance and Machine Learning
training and consulting advisory
• Trained more than 1000 students in
Quantitative methods, Data Science
and Big Data Technologies using
MATLAB, Python and R
• Building a platform for AI
and Machine Learning
Experimentation
4. 1. Algorithmic Auditing – Introduction
2. Algo auditing Frameworks
3. 5 things to note when auditing an algorithm
1. Use case
2. Data
3. Model
4. Environment
5. Process
4. Case study
Agenda
18. 18
• Algorithmic auditing is a structured process conducted internally or
by a qualified independent third party that involves:
▫ Verifying and/or validating the working of the algorithm along with the
data, model, environment, process, contextual to the use-cases in
which the algorithm is intended to be used.
▫ Identifying issues that are clearly articulated and scoped for the
algorithm. Criteria could include:
– Bias, fairness, discrimination, explainability, interpretability etc.
▫ Documenting the understanding of the algorithm’s behavior, uses as
observed and evaluated by a qualified individual.
▫ Recommending mitigation, control and elimination of noted risks.
Algorithmic Auditing
21. 21
How would you structure an algorithmic audit?
• Typically after the model is done
• Independence
• Subject matter expertise
• Disclosure
• Legal/Industry standard
External
• During or before model deployment
• Due diligence
• Policy – Model risk/Governance
• Best practices
• Workflow
Internal
22. 22
Why an algorithmic audit?
• Fraud detection
• Credit decision
• Facial recognition
Potential Systemic issues
• Blackbox models
• Vendor models
• Proprietary models
Transparency
• Methods used for the decision making process
• Security and Privacy
• Use of variables and data
Accountability
23. 23
• SMACTR (Google, Partnership of AI)
• SAI : Supreme Audit Institutions (Finland, Germany, the Netherlands,
Norway and the UK)
• ICO : UK
• TUV Austria: Trusted Artificial intelligence white paper
AI Auditing frameworks
24. 24
• Scoping
• Mapping
• Artifact Collection
• Testing
• Reflection
SMACTR
Ref: https://arxiv.org/abs/2001.00973 Closing the AI Accountability Gap: Defining an End-
to-End Framework for Internal Algorithmic Auditing
25. 25
Toward Trustworthy AI Development: Mechanisms for
Supporting Verifiable Claims
Ref: https://arxiv.org/pdf/2004.07213.pdf
26. 26
Toward Trustworthy AI Development: Mechanisms for
Supporting Verifiable Claims
Ref: https://arxiv.org/pdf/2004.07213.pdf
27. 27
Toward Trustworthy AI Development: Mechanisms for
Supporting Verifiable Claims
Ref: https://arxiv.org/pdf/2004.07213.pdf
29. 29
Main general problem areas and risks:
• Developers of ML algorithms will often focus on optimising specific numeric
performance metrics. As a result, there is a high risk that requirements of
compliance, transparency and fairness are neglected.
• Product owners within the auditee organisation might not communicate their
requirements well to ML developers, leading to ML algorithms that could, in a worst
case scenario, increase costs and make routine tasks more time-consuming.
• Auditee organisations often lack the resources and competence to develop ML
applications internally and thus rely on consultants or procure ready-made solutions
from commercial businesses. This increases the risk of using ML without the
understanding necessary both for ML-based production/maintenance and
compliance requirements.
• There is significant uncertainty among public-sector entities in the MoU member
states about the use of personal data in ML models. While the data protection
agencies have begun to issue guidelines, organisational regulatory structures are not
necessarily in place and accountability tends to be unclarified.
Auditing machine learning algorithms: A white paper for
public auditors
https://www.auditingalgorithms.net/executive-summary.html
30. 30
• Auditors need a good understanding of the high-level principles of ML
algorithms and up-to-date knowledge of the rapid technical
developments in this field - this is sufficient to perform a baseline audit
by reviewing the respective documentation of an ML-system.
• For a thorough audit that includes substantial tests, auditors need to
understand common coding languages and model implementations, and
be able to use appropriate software tools.
• ML-related IT infrastructure often includes cloud-based solutions due to
the high demand on computing power. Therefore, auditors need a basic
understanding of cloud services for this kind of audit work.
Auditing machine learning algorithms: A white paper for
public auditors
https://www.auditingalgorithms.net/executive-summary.html
31. 31
• This paper reaches the following conclusions and recommendations for SAIs:
• SAIs should be able to audit ML-based AI applications in order to fulfil their statutory
mission and to assess whether use of ML contributes to efficient and effective public
services, in compliance with relevant rules and regulations.
• ML audits require special auditor knowledge and skills, and SAIs should build up the
competence of their auditors.
• The ML audit catalogue and helper tool proposed in this paper have been tested in
our case studies and may be used as templates. They are living documents and thus
should be refined by application to more cases and to more diverse cases, and
consistently updated with new AI research results.
• SAIs should build up their capacities to perform more ML audit work.
• The authors hope that the guidance and good practices provided within this paper,
alongside the audit helper tool, will enable the international audit community to
begin auditing ML.
Auditing machine learning algorithms: A white paper for
public auditors
https://www.auditingalgorithms.net/executive-summary.html
33. 33
The framework:
• gives us a clear methodology to audit AI applications and ensure
they process personal data fairly, lawfully and transparently;
• ensures that the necessary measures are in place to assess and
manage risks to rights and freedoms that arise from AI;
• and supports the work of our investigation and assurance teams
when assessing the compliance of organisations using AI.
ICO - UK
https://ico.org.uk/media/for-organisations/guide-to-data-
protection/key-data-protection-themes/guidance-on-ai-and-
data-protection-0-0.pdf
34. 34
The framework output:
• Auditing tools and procedures which our investigation and
assurance teams will use when assessing the compliance of
organisations using AI. The specific auditing and investigation
activities they undertake vary, but can include off-site checks, on-
site tests and interviews, and in some cases the recovery and
analysis of evidence, including AI systems themselves.
• This detailed guidance on AI and data protection for organisations,
which outlines our thinking.
• A toolkit designed to provide further practical support to
organisations auditing the compliance of their own AI systems
ICO - UK
https://ico.org.uk/media/for-organisations/guide-to-data-protection/key-data-protection-
themes/guidance-on-ai-and-data-protection-0-0.pdf
37. 37
Questions to ask:
• Do we really need this algorithm?
• How will this algorithm be used?
• Who/What will it affect?
1. Use cases are important
38. 38
Things to think about:
• How much data do we
have?
• How will this affect the
model?
• Do we have enough data?
• Are their privacy concerns?
2. Don’t forget the data
39. 39
All scenarios haven’t
played out
• Stress scenarios
• What-if scenarios
Challenges with real datasets
Figure ref: http://www.actuaries.org/CTTEES_SOLV/Documents/StressTestingPaper.pdf
41. 41
Questions to ask
• Blackbox/Whitebox
• Does the model work?
• How do we handle imbalanced
classes?
• Is it fair/biased?
• Can you explain the model?
3. Model Audit
59. Thank you!
Sri Krishnamurthy, CFA, CAP
Founder and CEO
QuantUniversity LLC.
srikrishnamurthy
www.QuantUniversity.com
Contact
Information, data and drawings embodied in this presentation are strictly a property of QuantUniversity LLC. and shall not be
distributed or used in any other publication without the prior written consent of QuantUniversity LLC.
59