Prospective PhD Applicant • Reinforcement Learning

Negar Arianfar

Negar Arianfar

Applied Reinforcement Learning • Decision Systems • From Theory to Implementation

I work on applied reinforcement learning and sequential decision-making bridging theory to reproducible implementations (PyTorch, custom environments, evaluation).

GitHub View Research Projects Email

Research Interests

Reinforcement Learning • Decision-Making Systems • Reward Design • Safe & Robust Learning • Multi-Agent / Distributed Optimization

RL DQN MDP Modeling Reward Shaping Evaluation

Currently Working On

Rewriting my earlier work in energy-aware optimization into an RL formulation and building an implementable research prototype.

Working Paper RL for Optimization From Theory → Implementation

About

I am an applied ML researcher focused on building reinforcement learning systems that can be tested, measured, and extended. I enjoy designing environments, implementing agents, and running structured evaluations to answer research questions with reproducible code.

M.Sc. in Business Management International Marketing & Sales Hochschule der Wirtschaft für Management (HDWM), Germany (2023–2025)
M.Sc. in Software Engineering University of Najaf Abad Azad, Iran (2014–2018)
B.Sc. in Hardware Engineering University of Najaf Abad Azad, Iran (2008–2014)

Research Vision

My research focuses on designing reinforcement learning systems that move beyond theoretical formulation toward robust, reproducible implementations. I am particularly interested in how environment modeling, reward design, and stability mechanisms influence long-term decision performance.

I aim to develop learning-based optimization frameworks for distributed and strategic decision-making problems, with an emphasis on reliability, evaluation methodology, and practical scalability.

Research Questions

  • How does reward design influence stability and long-term performance in Deep Q-Network based agents?
  • What role does environment modeling play in shaping learning efficiency and robustness in constrained decision systems?
  • How can reinforcement learning be formulated for distributed optimization problems (e.g., energy-aware resource allocation) with reproducible evaluation protocols?
  • Which stability mechanisms (replay buffers, target networks, masking strategies) most effectively reduce training variance in small-scale strategic environments?

Publications & Preprints

Energy-Efficient Clustering Protocol for Wireless Sensor Networks
Conference Paper (in Persian), 2018

This paper proposes an energy-aware clustering protocol to extend network lifetime in wireless sensor networks through optimized cluster-head selection mechanisms.

PDF
Reinforcement Learning for Strategic Decision Systems

This research transforms the Business Chessboard Strategy (BCS) into an AI-executable framework using Hierarchical Reinforcement Learning. By modeling strategic decisions as a Hierarchical Markov Decision Process (H-MDP), the project enables RL agents to simulate, evaluate, and optimize long-term strategies under uncertainty. A custom simulation environment (BCS-Simulator) is developed to train and analyze intelligent agents, combining deep RL methods with explainable AI tools. The goal is to bridge human strategic theory and computational decision optimization.

Research Proposal
PDF

Research Projects

MiniChess-RL: Deep Q-Network Agent in a Custom 4×4 Chess Environment

Code

Built a custom 4×4 mini-chess environment and trained a DQN agent end-to-end using Python + PyTorch. The project demonstrates environment modeling, reward shaping, illegal move masking, training loops, and evaluation.

Reinforcement Learning Deep Q-Network Environment Design Illegal Action Masking Evaluation PyTorch
View Details

Working Paper: RL Formulation of Energy-Aware Optimization (in progress)

— details coming soon

Translating an earlier energy optimization / clustering idea into an RL problem definition (state/action/reward), and planning reproducible experiments and baselines.

MDP Formulation Baselines Experiment Design

Selected practical projects focused on automation, data, and applied AI.

Python Data Automation Tool

CLI-based tool for cleaning raw CSV sales data and generating a structured summary report. Built for fast, reusable data cleanup workflows.

  • Removes duplicates and trims messy text
  • Standardizes dates and numeric fields
  • Handles missing values safely
  • Generates summary.txt + cleaned CSV outputs
Python Pandas Argparse
View on GitHub
python src/run.py --input data/raw_sales.csv --outdir output

Books

Learning Reinforcement Learning Through Projects: From Zero to your First Deep Q-Network

Learning Reinforcement Learning Through Projects: From Zero to your First Deep Q-Network

A practical and intuitive introduction to reinforcement learning, guiding readers from bandit problems to Deep Q-Networks (DQN). The book emphasizes environment design, reward modeling, instability analysis, and hands-on implementation.

View on Amazon
Chessboard of Business

Chessboard of Business

A strategic guide to building successful businesses through structured decision-making, foresight, and chess-inspired thinking. Focused on practical frameworks for sustainable growth.

View on Amazon

CV

Download Academic CV

Contact

Email: negarinarianfar@gmail.com
GitHub: github.com/negarinarianfar
Linkedin: linkedin.com/in/negar-arianfar