Yugandhar Reddy Gogireddy

Proposes a hybrid pipeline combining Multi-Scale Retinex defogging with Super-Resolution CNNs to improve clarity, contrast, and color restoration for degraded underwater images; evaluates improvements using SSIM and PSNR on real underwater datasets.

Reproduction and evaluation study of dialogue summarization models (e.g., Hierarchical Memory Networks and Pointer-Generator variants) on the AMI dataset, including human assessments of summary quality and analysis of reproduction discrepancies.

Describes an embedding-based retrieval pipeline for large-scale PDF document search using distributed FAISS and sentence-transformer embeddings to enable efficient document retrieval.

Examines parameter-efficient approaches for NLP models to make them more suitable for resource-constrained settings, discussing techniques to reduce model footprint while maintaining performance.

Investigates combining graph-based sentence selection with transformer fusion methods to improve abstractive summarization performance.

A paper on CAPTCHA image recognition employing CNNs and LSTM models.