CI-Guided Data Curation: Using prediction instability to detect label noise. Validated on SST-2 with 10% noise injection. Recovered 40% of accuracy gap to perfect labels. Part of the Collapse Index framework.

NLP Course | Home Works | Fall 2021 | Dr. Behrooz Minaei

BERT-based sentiment analysis for IMDB and SST-2 datasets using Hugging Face Transformers. Includes training, evaluation, and batch prediction scripts under MIT License.

MOCA-Net: Novel neural architecture with sparse MoE, external memory, and budget-aware computation. Real Stanford SST-2 integration, O(L) complexity, 96.40% accuracy. Built for efficient sequence modeling.