Singapore

Learning from human feedback enables AI systems and robots
to learn policies that align with human intent. While
existing work has primarily examined learning from
demonstrations, corrections, and preferences in
single-agent settings, these ideas have yet to be fully
extended to multi-agent domains—where cooperation,
decentralization, and non-stationary dynamics demand new
methods. In this thesis summary, I highlight my current
work and outline future directions for multi-robot learning
from human feedback, offering deployment strategies that
align supervisor intent with robot teams in the real world.

AAAI 2026 Main Conference

Multi-Robot Learning from Human Feedback

technical paper

We are pleased to announce the Fortieth AAAI Conference on Artificial Intelligence (AAAI-26), which will be held in Singapore EXPO from January 20 to January 27, 2026.

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines. AAAI-26 will feature technical paper presentations, special tracks, invited speakers, workshops, tutorials, poster sessions, senior member presentations, competitions, and exhibit programs, and a range of other activities to be announced.<br><br>

To access this event page, you need to log in with the **email address you registered with**. <br>Access credentials will be sent to your email from Underline -  subject line "Welcome to AAAI 2026". Please be sure to check your spam email folder if you do not see an email confirmation right away.

Please log in

To access this event page, you are required to register.
Please complete your registration to continue.

We recommend reading [**the registration information**](https://aaai.org/conference/aaai/aaai-26/registration/) first.

**Online Registration Form**: https://aaai.getregistered.net/conference-2026 

Registration Required

We are pleased to announce the Fortieth AAAI Conference on Artificial Intelligence (AAAI-26), which will be held in Singapore EXPO from January 20 to January 27, 2026.

Model development in AI is shaped by developer decisions.
While there is significant research on the opportunities
and risks of multiplicity, little attention has been paid
to how developer decisions impact multiplicity. My thesis
focuses on (a) introducing broader frameworks to better
situate and analyze developer decisions in AI, (b)
identifying theoretical connections to characterize the
influence of these decisions on multiplicity, and (c)
operationalizing these insights across various
applications, thus building responsible AI models with
multiplicity.

From Decisions to Multiplicity: Frameworks, Theories, and Applications

Explainable AI (XAI) seeks to answer the question: which
features of the data led a model to make its decision?
Existing approaches are either model-agnostic (e.g., LIME,
SHAP)—flexible but unstable—or logic-based (e.g.,
sufficient reasons,
knowledge compilation)—principled but often overly complex.
This work introduces a probabilistic relaxation of
sufficient reasons, termed probabilistic sufficient
reasons, which balances flexibility with theoretical
guarantees. We analyze
its computational properties, propose tractable subclasses,
and outline future directions for scalable algorithms and
applications.

On the Computational Tractability of Probabilistic Global and Local Sufficient Explanation

Achieving globally desirable outcomes in networked
multi-agent systems—such as high social welfare, stable
allocations, and widespread cooperation—is a fundamental
challenge in AI. This paper outlines a research agenda that
explores two complementary pathways to this goal. The first
is a top-down approach, where a central mechanism designer
proposes rules to guide strategic agents towards
theoretically optimal equilibria. The second is a bottom-up
approach, where desirable farsighted policies, like
cooperation in social dilemmas, emerge from the
decentralized interactions of agents via multi-agent
reinforcement learning. We argue that the integration of
these paths constitutes a promising frontier for creating
robust and adaptive multi-agent systems.

Designing Incentives for Networked Multi-agent Systems

Transformers have reshaped modern artificial intelligence, yet their theoretical foundations remain incomplete. This thesis investigates the approximation power and memory limitations of transformers. I combine tools from approximation theory and statistical learning theory to provide provable guarantees on expressivity, memorization capacity, and inherent architectural constraints. My contributions include the first rigorous proof of memory bottlenecks in prompt tuning and new results on the expressivity of transformers. The long-term goal of my doctoral research is to develop a principled theoretical framework that grounds the empirical behavior of large-scale transformer models in formal approximation-theoretic results.

Memorization and Expressivity in Transformers: A Learning-Theoretic Perspective

Explainability has emerged as a pillar of Trustworthy AI
for achieving safety in critical domains. However,
introducing explainability to boost transparency of
black-box AI systems can create unforeseen vulnerabilities.
Previous research has drawn attention to privacy leakage,
malicious or otherwise, that explainable interfaces can
cause thus leading to inadvertent identification of
individuals and/or exposure of sensitive personal
information. Privacy preservation methods used in response
to this leakage, are found to adversely affect utility of
the system such as the model accuracy and explanation
quality. The proposed thesis will examine the advancement
of Privacy Enhancing Technologies (PETs) in XAI keeping
users at the core of the design process. The main
objectives of the research are determining defenses for
privacy attacks, building interpretable algorithms for
private models and examining user requirements for privacy
preserving XAI. The research is expected to yield
characteristics of privacy preserving XAI, guidelines, and
recommendations for effectively building privacy compliant
XAI while considering the diverse needs of users. The
research outcomes will enable developers and researchers in
designing XAI that is safe for deployment and balances the
triad of privacy, explainability and utility.

Conceptualisation and Implementation of Human-centric Privacy Preserving Framework for Explainable AI

The rapid advancement of generative mod- els (Roy et al.
2023; Pal et al. 2024; Roy et al. 2024) has opened new
avenues for addressing critical challenges in computer
vision (Dhar et al. 2021; Fazlyab et al. 2023), such as
data scarcity, image quality enhancement, and person-
alization. Recent progress has concentrated on improving
the adaptability, efficiency, and quality of these models
to meet the growing demand for parameter-efficient
fine-tuning and adaptation of large vision-language and
generative mod- els (Roy et al. 2025b; Pramanick, Roy, and
Patel 2022). In this work, we begin by tackling the
challenges of resource- constrained learning (Roy et al.
2022). We then leverage powerful vision-language models to
address these issues in a parameter-efficient manner.
Additionally, we aim to en- hance state-of-the-art
generative models—specifically dif- fusion models—by
incorporating natural image priors (Roy et al. 2023). We
also explore joint concept merging through the lens of
low-rank adapter merging, applying it to content- style
personalization. Finally, we address the challenge of
zero-shot personalization of any object without requiring
additional training. We conclude by devising a frequency-
guided method for training-free multi-LoRA composition,
which is more appropriate for deployment on edge devices.

Learning More from Less: Resource-Constrained Generative AI for Classification, Generation, and Personalization

Large Language Models (LLMs) have demonstrated remarkable
capabilities in reasoning, yet their efficacy is
constrained by a fundamental memory limitation: a static
context window that resets with each interaction. This
prevents them from accumulating experience and adapting to
dynamic, long-term tasks. To address the limitations of
long-term memory in agentic LLMs, this work introduces a
neuro-inspired framework with two key contributions. First,
we propose \textbf{ARTEM} (Agentic Retrieval with
Temporal-Episodic Memory), a system that organizes
experiences into structured events and manages
utility-based memory consolidation. Second, we extend this
framework with a distinct governance component,
\textbf{Value-driven ARTEM}, that validates candidate
outputs against core principles before finalization.
Together, these components equip LLM agents with continual
learning, adaptive reasoning, and robust value-aligned
decision-making. Looking forward, we outline future
directions including dynamic memory adaptation, memory
decay mechanisms, and applications in interactive
multi-agent environments.

Value-Driven Memory-Augmented Generation for Agentic LLMs: Towards Structured and Adaptive Knowledge Utilization

Autonomous systems are increasingly deployed in complex,
uncertain environments, where they must make their own
decisions and adapt to unexpected conditions without human
intervention. System decisions have critical implications
for safety, reliability, and task success, yet current
approaches often address only one isolated aspect of this
challenge. For instance, there have been numerous, yet
separate, advances in planning under uncertainty with
optimal control methods, anticipating failures with
conformal prediction thresholding, and integrating large
language models with AI-based planners. This gap raises the
question: how can these capabilities be unified in a
framework that enables autonomy to operate reliably across
uncertain domains without human oversight? My dissertation
will address this challenge by developing methods that link
these threads, to contribute towards advancing trustworthy
autonomy that can operate robustly and transparently in
real environments.

Trustworthy Autonomy Without Human Intervention in Uncertain Domains

The submission contains the following documents:
1. Cover Sheet
2. Thesis Summary
3. Curriculum Vitae
4. Personal Statement

Factuality Evaluation using Reasoning and World Modeling

My research investigates how to evaluate and enhance large
language models’ (LLMs) alignment with human values in
collective decision-making scenarios. I focus on three
inter-
related aspects of this challenge: (i) normative alignment,
(ii)
procedural competence, and (iii) personalization.

Content not yet available

Downloads

Next from AAAI 2026 Main Conference

From Decisions to Multiplicity: Frameworks, Theories, and Applications

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

Content not yet available

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from AAAI 2026 Main Conference

From Decisions to Multiplicity: Frameworks, Theories, and Applications

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

Downloads