Spatio-Temporal ML • Network Science • Computational Social Systems

I’m a graduate researcher working on machine learning for complex social and urban systems.
My work focuses on how spatial structure, temporal dynamics, and network interactions shape real-world phenomena such as urban growth, elections, mobility patterns, and scientific collaboration.

I build models that operate under distribution shift, limited supervision, and heterogeneous real-world constraints, with an emphasis on interpretability, robustness, and domain grounding.

I’m currently completing my M.S. in Data Science at UMass Dartmouth (2024–2026) and preparing for PhD applications for Fall 2027.

📧 rohanbaliwork@gmail.com
📍 Boston, MA

• Spatio-Temporal Machine Learning
• Graph Neural Networks & Network Science
• Computational Social Science
• Learning Under Distribution Shift
• Machine Learning for Cities, Elections, and Public Policy

Long-Horizon Urban Growth Forecasting

A ConvLSTM-based model for forecasting multi-decadal urban expansion using GHSL built-up rasters and OpenStreetMap road networks.

• Trained on 2,313 CONUS tiles (1975–2000)
• Achieved MSE: 0.000218, RMSE: 0.00641
• 67% improvement over U-Net baseline
• Stable generalization across geographies

Tech: TensorFlow, Keras, NumPy, GDAL
Repo: https://github.com/rohanbalixz/NeuralTimeCapsule
Paper: https://github.com/rohanbalixz/NeuralTimeCapsule/blob/main/paper/Bali2025_NeuralTimeCapsule_UrbanGrowthPrediction.pdf

Small-World Analysis of Scientific Communities

• Analyzed 180K+ co-authorship edges across 12 fields
• Found 72% of high-impact researchers occur within a 3-hop neighborhood
• Built pipelines for graph construction, clustering, and author-impact inference

Tech: NetworkX, Gephi, Python
Repo: https://github.com/rohanbalixz/Understanding-Academic-Collaboration-Networks-Through-Small-World-Theory

NVIDIA DLI – Remote Sensing ML

U-Net trained on Sentinel-1 radar imagery for flood detection.

• Dice: 0.82, IoU: 0.78
• SAR-specific preprocessing pipeline
• Integrated into Google Earth Engine

Tech: TensorFlow, GDAL
Repo: https://github.com/rohanbalixz/Disaster-Risk-Monitoring-Using-Satellite-Imagery

A streaming pipeline combining:

• Bayesian hierarchical models
• Transformer-based sentiment analysis
• News and poll ingestion
• Sub-5-minute update latency

Tech: PyTorch, XGBoost, Kafka, NGBoost, RoBERTa
Repo: https://github.com/rohanbalixz/Bihar-Election-Analytics

Retrieval-augmented travel assistant with:

• Geographic embeddings
• OpenRoute integration
• 35% hallucination reduction over GPT-only baselines

Repo: https://github.com/rohanbalixz/Tripsy-AI-Travel-Assistant

Standalone MapReduce system using LevelDB + Python multiprocessing.

• Near-linear scaling (R² = 0.94)
• 1.8× faster than naive baselines

Repo: https://github.com/rohanbalixz/LocalMapReduce

Physics-guided intrusion detection for SunSpec/Modbus networks.

Repo: https://github.com/rohanbalixz/clad-pv

Emotion-aware transformer-based conversational agent.

Repo: https://github.com/rohanbalixz/Aether

• NVIDIA DLI: Disaster Risk Monitoring with Satellite Imagery
• Docker Certified: Foundations Professional
• Azure Fundamentals & ML Engineer (Microsoft)

• “Understanding Academic Collaboration Networks through Small World Theory.”
  https://swn465.home.blog/2024/12/12/understanding-academic-collaboration-networks-through-small-world/

• “Neural Time Capsule: Urban Growth Prediction.”
  https://github.com/rohanbalixz/NeuralTimeCapsule/blob/main/paper/Bali2025_NeuralTimeCapsule_UrbanGrowthPrediction.pdf

I’m open to collaborations in:

• Spatio-temporal ML
• Network science
• Computational social systems
• Urban data and policy modeling

📧 rohanbaliwork@gmail.com
🌐 Portfolio: https://rohanbalixz.github.io/rohan-bali-portfolio/
📚 Scholar: https://scholar.google.com/citations?user=CBgU9IIAAAAJ&hl=en