Curriculum Vitae

Dartmouth, MA

Thesis: Learning from Minimal Data under Distribution Shift (BrainLab OS)

Jaipur, India

Master's thesis developing frameworks for robust machine learning under distribution shift and limited supervision. Building neuroanatomically-inspired cognitive architecture that learns efficiently from minimal labeled data.

• •Developing theoretical foundations for generalization under distribution shift
• •Implementing neuroscience-inspired learning mechanisms for data-efficient training
• •Investigating robustness and adaptation in low-data regimes across multiple domains

Developed a ConvLSTM-based forecasting model for long-horizon urban expansion using GHSL builtup rasters and OpenStreetMap road networks.

• •Trained on 2,313 CONUS tiles (1975-2000); achieved MSE = 0.000218 and RMSE = 0.00641
• •Demonstrated 67% improvement over U-Net baseline with stable generalization across regions

Analyzed 180K+ co-authorship edges across 12 academic domains using small-world statistics.

• •Found that 72% of high-impact researchers cluster within a 3-hop neighborhood
• •Developed reproducible pipelines for citation graph construction and author-impact inference

Trained U-Net on Sentinel-1 radar imagery for flood segmentation (Dice = 0.82, IoU = 0.78).

• •Designed preprocessing and augmentation pipeline optimized for noisy radar data
• •Integrated predictions into Google Earth Engine for real-time risk assessment

Built retrieval-augmented assistant with geographic embeddings and OpenRoute integration.

• •Reduced hallucinations by 35% and improved route relevance by 22% over GPT-only models

Implemented local MapReduce architecture with LevelDB storage and multiprocessing.

• •Achieved near-linear scaling on 8 cores (R² = 0.94) and 1.8× improvement over naive baselines

Neural Time Capsule: Urban Growth Prediction

Rohan Bali

In preparation for AAAI 2026 or IJCAI 2026 (2025)

Understanding Academic Collaboration Networks through Small World Theory

Rohan Bali

Medium Blog Post (2024)