Internship Opportunities at ADIA Lab

Project Description:

Collaborative data science methods help harness collective intelligence from fragmented and siloed organizational assets or device ecosystems. A major challenge in enabling collaborative intelligence in the LLM era is maintaining high resource efficiency. Praneeth’s research group has developed state-of-the-art methods for extremely efficient LLM fine-tuning, such as LoRA-Silver Bullet. This project will apply these methods in a Spoke-and-Hub fashion to explore various approaches, as outlined below:

Goals:

• Achieve alignment to non-toxic, ethical, and safe answers after fine-tuning LLMs. Starting with an aligned LLama LLM, the project will apply the LoRA-Silver Bullet method for extremely efficient fine-tuning on a client's local task.
• Measure alignment properties post fine-tuning to establish a baseline.
• Research ways to improve alignment using LoRA-SB while maintaining strong performance on the client’s task.
• Apply socially fair learning with LoRA-SB to achieve fairness goals and analyze the trade-offs.

Students will receive structured guidance from the research group. Previous students working on similar projects have gone on to PhD programs at top universities and secured strong industry positions.

For more on the group's advances in extremely efficient LLM fine-tuning, see:

1.) Initialization using Update Approximation is a Silver Bullet for Extremely Efficient Low-Rank Fine-Tuning

2.) FedEx-LoRA: Exact Aggregation for Federated and Efficient Fine-Tuning of Foundation Models

Project Description:

This project focuses on analyzing AI safety policies and regulations in healthcare, with a particular emphasis on the UAE. The intern will conduct a comprehensive literature review, examine existing AI regulations, and carry out a comparative analysis to identify best practices and gaps in AI governance, with a focus on ensuring AI safety and fostering human-AI collaboration in healthcare.

The project will also explore the technological, systemic, and societal factors influencing AI safety policies, including their impact on healthcare providers, insurers, patient trust, and accessibility. A key objective is to investigate how AI safety regulations shape human-AI collaboration in healthcare, ensuring AI tools are safe, effective, and seamlessly integrated into clinical and administrative workflows.

The intern will contribute to the development of a conceptual framework for the safe and effective adoption of AI in healthcare. This framework will synthesize key insights from various regions and highlight essential principles and strategies for balancing innovation, regulatory compliance, ethical considerations, and human-AI synergy in AI-driven healthcare systems.

Project Description:

There are ongoing efforts to incorporate health data into an individual’s Verified Credential data structure, aiming to create a Health Data Wallet. This data wallet would enable an individual’s smartphone to store copies of relevant medical records. These records, along with related patient behavioral datasets, could then be shared with the appropriate medical professional (e.g., doctor) without the need to interconnect medical databases across institutions—an otherwise complex process.

Project Description:

Population, life expectancy, and well-being are key indicators of sustainability. In this internship, you will model these factors from a causal perspective, examining their relationships with variables such as education, healthcare costs, and GDP. The goal is to generate insights that can guide effective policy decisions for developing nations.

This project focuses on identifying the key drivers of population dynamics across different countries. To ensure a data-driven approach, we prioritize factors with sufficient historical data. We then apply data-based algorithms (such as PC and GES) and LLM-based methods (like Automated Causal Discovery and Efficient Causal Discovery) to construct a Directed Acyclic Graph (DAG) representing causal relationships between these factors. Using the DAG and historical data, we quantify causal effects (causal betas) for the edges through do-calculus.

Finally, we translate our findings into policy recommendations based on these causal insights.

Project Description:

Abu Dhabi’s commitment to sustainability, carbon reduction, and financial innovation makes this project essential for developing dynamic models to support real-time decision-making. This project focuses on detecting structural breaks in time series data related to carbon emissions, energy prices, and macroeconomic indicators using advanced machine learning techniques, including Reinforcement Learning (RL) and Bayesian Change Point (BCP) detection. Interns will explore how abrupt shifts in environmental and financial systems, such as carbon pricing or energy transitions, influence Abu Dhabi’s policy and investment strategies.

The project has practical applications for policymakers and investors, particularly within entities like ADIA and Masdar, to better understand when structural changes occur and adapt investments or policies accordingly.

Project Description:

Federated learning is a machine learning technique that enables multiple devices or servers to collaboratively train a model without sharing raw data. Each participant (e.g., smartphones, hospitals) processes data locally and only shares model updates, enhancing privacy and security—especially in sensitive domains such as healthcare and finance.

Open-source platforms play a crucial role in federated learning by providing frameworks and tools to manage these distributed learning processes. The FLEX library, along with TensorFlow Federated (TFF) and PySyft, enables researchers to implement federated learning efficiently and securely.

This project focuses on client selection strategies and their impact on model quality and scalability. Client selection plays a critical role in federated learning, affecting efficiency, performance, and communication overhead. The project will involve setting up a federated learning environment, training models, and analyzing how different client selection approaches influence the overall system.

Reference:
F. Herrera et al., "Flex: Flexible Federated Learning Framework," Information Fusion, Vol. 117, May 2025, 102792. Link

Project Description:

Explainable Artificial Intelligence (XAI) focuses on making AI systems transparent and understandable, fostering trust, accountability, and fairness. By offering clear explanations, XAI helps bridge the gap between complex AI models and human comprehension.

A key challenge in XAI is tailoring explanations to different audiences. AI developers, domain experts, regulatory bodies, and end-users all require different levels of detail and clarity. Designing effective explanations involves understanding these diverse needs and ensuring accessibility without compromising accuracy.

For example, in a healthcare AI system:

• For doctors: AI-generated explanations may include detailed visualizations of medical images, statistical insights, and highlighted areas of concern.
• For patients: The system should generate simplified explanations with clear, non-technical language and visual aids to enhance understanding.

This project will explore how to design and implement effective XAI explanations for different user groups, focusing on a healthcare-related AI application.

Project Description:

This project explores the relationship between climate variables and economic indicators, aiming to provide meaningful estimations and predictions using statistical and machine learning methods. The findings will help inform environmental policies and economic decision-making. Over the course of eight weeks, the intern will work under the supervision of Emilio Porcu and collaborate with a group of PhD students in a co-mentorship setup.