SANUP S. ARABALLI // RESEARCH OS v2.0
>LOADING EVOLUTIONARY SUBSTRATE...
>INITIALIZING POPULATION ···· 28 AGENTS READY
>GEN 001 → GEN 002 → GEN 003 ···
>FITNESS LANDSCAPE MAPPED // READY

PhD Candidate · Open to Research Scientist & Applied Scientist roles

Sanup S.
Araballi

I make hard problems tractable.

Bilevel OptimizationSurrogate EAExplainable AIGraph NNReinforcement Learning

Explore Research Projects CV

What I work on

Research areas

Bilevel Optimization

Nested optimization for large-scale, NP-hard combinatorial problems

Surrogate-Assisted EA

Evolutionary search guided by learned fitness models

Explainable AI

Distilling black-box models into interpretable rules

Graph Learning

Embeddings and information cascades on networked data

Reinforcement Learning

Distributional and interpretable control

Affiliations & service

Air Force Research LabSummer Research Intern

Syracuse UniversityPhD Candidate · CS & Engineering

Peer ReviewerNeurIPS · IJCNN · Tapia · IEEE WCCI

Primary InstructorCPS196: Intro to Python

Published at AAAI· ECTA / IJCCI · BEA · ARF · Collaboration with ComScore

Publications

Selected work

All publications →

2026

AAAI Spring Symposium

Distilling Deep Reinforcement Learning into Interpretable Fuzzy Rules

Compressed 26,000+ agent experiences into 74 human-readable IF-THEN rules, making black-box RL certifiable for safety-critical control systems.

Key equation

IF x₁ is Ã₁ AND x₂ is Ã₂ → y = p₁x₁ + p₂x₂ + r

2025

ECTA / IJCCI

A Surrogate-Assisted Co-Evolutionary Framework for Bilevel Optimization

>96% reduction in computation cost on NP-hard bilevel problems. Validated on Max-Cut, extended to Gene Regulatory Networks.

Key equation

min F(x, y*) s.t. y* ∈ argmin f(x, y)

2021

BEA ConferenceOpen Paper Winner

Developing an Ad Viewing Retention Model for TV Comedy Through Machine Learning

80%+ accuracy predicting TV ad retention from ComScore data. Recognised in a ComScore industry press release.

Projects

Research in action

All projects →

Deep RL · Finance

Reinforced Trading Bot

C51 Distributional DQN, consistent profitability with zero hand-crafted rules

Evolutionary Algorithms

Aerodynamic Shape Optimizer

Genetic algorithm + Blender physics, evolved minimum-drag geometries from scratch

Graph Neural Networks

Information Cascade Predictor

DFS-biased node2vec outperforms state-of-the-art CasFlow on social network data

Let's connect

Get in touch

Open to discussing research collaborations, industry opportunities, or questions about my work. Actively exploring Research Scientist and Applied Scientist roles.

ssarabal@syr.edu

linkedin.com/in/sanuparaballi

github.com/sanuparaballi

Syracuse, NY

Email me directly

The fastest way to reach me. I read every message and reply personally.

ssarabal@syr.edu