Research

Machine learning security, model behavior, and mathematically grounded AI.

A research overview page for technical direction, paper-level work, methodology, and rigorous investigation.

Primary Lens

ML Security

Focused on attack surfaces, evaluation, and trustworthy deployment questions.

Method

Theory + Experiment

Research framing combines system reasoning with technical testing and analysis.

Scope

Models to Infrastructure

The work connects model behavior to the systems and environments that contain it.

Research Frame

Security questions become more interesting when architecture is part of the problem.

The research direction on this site is not only about isolated attacks. It is about how model structure, representation geometry, deployment assumptions, and system boundaries interact.

Methods

A blend of technical analysis, experimental evaluation, and infrastructure-aware reasoning.

The goal is to keep research grounded enough to matter in practice, while still being precise enough to contribute at the paper level.

Research Overview

The research direction centers on machine learning security, model behavior, and mathematically grounded approaches to AI systems that must operate under real-world constraints.

Core themes

privacy and security risk in modern machine learning systems
recommendation-model behavior and attack surfaces
mathematically informed reasoning about model structure and learnability
the gap between laboratory capability and deployable AI systems

Orientation

The goal is to pursue work that is both technically rigorous and operationally relevant. That means choosing problems where theory, system design, and deployment consequences can inform one another instead of living in separate silos.

Output direction

This section is designed to grow into a bilingual research overview spanning papers, technical investigations, working notes, and system-level questions that connect security, infrastructure, and product reality.

Flagship Threads

The current research narrative is anchored by these project surfaces.

ALOA research story

Follow the two-tower membership inference work as a scientific progression from architecture to findings.

Elementization story

See how infrastructure thinking and trust boundaries connect back to deployable AI research questions.