Foundational Research

Core methodological work that underpins our understanding of how and why AI systems make decisions.

Foundational

Feature Attribution Methods

Understanding which input features drive model predictions is fundamental to explainability. We study methods like SHAP and LIME, investigating their faithfulness, stability, and alignment with human intuition across model architectures and application domains.

  • Local vs. global explanation trade-offs
  • Attribution faithfulness and consistency metrics
  • Interaction effects in high-dimensional feature spaces
  • User studies on explanation comprehension
Feature Importance Waterfall Income +0.34 History +0.23 Tenure +0.12 Age -0.18 Region -0.09 − Decreases prediction Increases prediction +
Foundational

Counterfactual Explanations

Counterfactual explanations answer the question: “What would need to change for the outcome to be different?” We research methods for generating actionable, sparse, and plausible counterfactuals that respect causal structure and domain constraints.

  • Actionable recourse in automated decisions
  • Causal counterfactual generation
  • Plausibility constraints and feasibility
  • Comparative evaluation of counterfactual methods
Decision boundary Approved Denied Original Counterfactual minimal change
Foundational

Concept-based Explanations

Rather than explaining predictions in terms of raw features, concept-based methods use human-understandable concepts as the unit of explanation. We study concept bottleneck models, concept activation vectors, and methods for discovering latent concepts aligned with human reasoning.

  • Concept bottleneck architectures
  • Testing with Concept Activation Vectors (TCAV)
  • Automated concept discovery
  • Concept completeness and fidelity
Raw Input Striped pattern Has wings Yellow color Prediction: “Bee” Interpretable concept layer
Foundational

Human-in-the-Loop Evaluation

Explanations are only as good as their impact on human decision-making. We design and conduct rigorous user studies to evaluate how explanations affect comprehension, trust calibration, and task performance in real-world settings.

  • Trust calibration through explanations
  • Explanation modality and cognitive load
  • Task-grounded evaluation frameworks
  • Longitudinal studies of explanation utility
AI Model Explanation SHAP / text / visual Evaluator generates presents to feedback Trust • Comprehension • Task accuracy

Model Interpretability Audits

Post-hoc and ante-hoc analysis of model decision-making across architectures

Clinical Risk Model Review

Conducted post-hoc interpretability analysis of a clinical risk prediction model serving 2M+ patients. Identified feature interactions that contradicted domain knowledge, leading to model retraining and improved clinician trust.

SHAP Analysis Model Audit Healthcare

Credit Scoring Transparency Assessment

Evaluated a consumer lending model for explanation consistency across demographic groups. Delivered a technical report with actionable recommendations for regulators and internal stakeholders.

Fairness LIME Finance

Supply Chain Anomaly Detector Audit

Analyzed a deep learning anomaly detection system used in logistics. Produced global and local explanation reports that enabled operations teams to validate flagged events with confidence.

Anomaly Detection Global Explanations Logistics

Algorithmic Fairness Reviews

Bias detection, disparate impact analysis, and equity-centered remediation

Hiring Algorithm Bias Evaluation

Assessed an automated resume screening tool for disparate impact across protected characteristics. Recommended calibration strategies that preserved predictive performance while reducing bias metrics by 40%.

Bias Audit HR Tech Disparate Impact

Lending Decision Equity Analysis

Performed a comprehensive fairness audit of an automated underwriting system. Mapped decision boundaries across subpopulations and provided remediation strategies aligned with regulatory guidance.

Equity Analysis Underwriting Compliance

Content Moderation Fairness Review

Evaluated a large-scale content moderation classifier for inconsistent enforcement across languages and cultural contexts. Identified systematic gaps and proposed retraining protocols.

NLP Multilingual Platform Safety

XAI Dashboard Design

Human-centered explanation interfaces for operational AI systems

Fraud Detection Explanation Interface

Designed and prototyped a real-time explanation dashboard for a fraud detection system. Enabled analysts to inspect feature contributions per flagged transaction, reducing false positive review time by 30%.

Dashboard Real-time Fraud Detection

Clinical Decision Support Interface

Co-designed an explanation layer for a clinical decision support tool with practicing physicians. Iterative user testing ensured explanations aligned with clinical reasoning patterns.

HCI Clinical DSS User Research

Insurance Underwriting Explanation Portal

Built an interactive explanation portal that allowed underwriters to explore model reasoning through counterfactual queries and feature sensitivity analysis.

Counterfactuals Interactive Insurance

AI Policy & Governance

Frameworks, compliance roadmaps, and responsible AI strategy for organizations

Enterprise AI Governance Framework

Developed a comprehensive governance framework for a Fortune 500 company deploying AI across business units. Included model risk tiers, explanation requirements, and escalation protocols.

Governance Enterprise Risk Management

Financial AI Regulatory Compliance Roadmap

Created a compliance roadmap mapping existing AI systems to emerging regulatory requirements. Prioritized remediation efforts based on risk exposure and explanation gaps.

Regulation Finance Compliance

Government Responsible AI Policy

Advised a government agency on responsible AI deployment policy, including transparency standards for public-facing automated decision systems and citizen recourse mechanisms.

Public Sector Transparency Policy

Interested in our research?

We welcome collaborations with academic groups, industry labs, and policy organizations.

Get in Touch