Shylaja S S

Research profile

Research summary

Shylaja S S is a researcher at PES University, India, whose OpenAlex-resolved record runs from 2007 to 2026 with an h-index of 9 and an i10-index of 9. The profile's foundation is computer vision: OpenAlex topic counts are led by image enhancement techniques (15), image and signal denoising (14), face recognition and analysis (14), face and expression recognition (13), and advanced neural network applications (12). The current Evolve catalog, however, reflects a pivot toward two newer lines — machine learning (federated learning / graph neural networks) and educational technology (multimodal AI systems) — each holding exactly 50% of the field distribution. Activity is the standout signal, scored 100/100: 30 works in the last three years, an annual peak of 15 works in 2025, and 7 more already in 2026. Both cataloged papers are 2026 arXiv preprints with the researcher in last-author, non-corresponding positions: a diagnosis of structural negative transfer in federated graph neural networks, and an NCERT-textbook-to-video pipeline built on RAG, Stable Diffusion, DynamiCrafter, and gTTS. Reproducibility averages 38/100 across the two reviews, with Evolve praising honesty in one while flagging anecdotal evaluation in the other. Collaboration is compact — 5 distinct recent coauthors and a 36/100 score — while focus scores 0/100, marking genuine breadth rather than a narrow specialty.

Research keywords

Highlights

• Federated GNN negative transfer: honest diagnosis, open questions

  Last-authored 2026 arXiv preprint identifying structure-driven harm in federated graph neural networks through a degree-divergence signal, with causal investigation and an explicit account of where divergence-aware mitigation falls short. Evolve credits the paper's honesty while flagging unresolved causality and small samples; reproducibility readiness is rated medium.

• NCERT textbooks turned into personalized educational videos

  Last-authored 2026 arXiv preprint presenting a system that converts NCERT PDF textbooks into video using retrieval-augmented generation, Stable Diffusion, DynamiCrafter, and gTTS. Evolve calls the engineering practical but the evaluation anecdotal, with citations that appear misattributed; reproducibility readiness is rated low.

• A two-decade computer-vision base beneath a new agenda

  OpenAlex topic counts root the profile in image analysis: image enhancement techniques (15), image and signal denoising (14), face recognition and analysis (14), face and expression recognition (13), and advanced neural network applications (12) — the substrate on which the newer federated-learning and generative-education lines rest.

Timeline

1. 2007First recorded publication year in the OpenAlex counts-by-year profile (1 work), opening a record that runs continuously to 2026.
2. 2010Early peak with 4 works — the highest annual count of the profile's first decade.
3. 2015Only 1 work recorded, and no works at all in 2016 — the quietest stretch of the record.
4. 2017Output recovers with 5 works, resuming the steady cadence that continues through 2021 at 6–8 works per year.
5. 2025Most productive year on record: 15 works, within a 30-work total across the last three years.
6. 20267 works already recorded, including both Evolve-cataloged arXiv preprints on federated GNN negative transfer and NCERT-to-video generation.

Papers in the Evolve catalog

• 2026-09-15

  Structural Negative Transfer in Federated Graph Neural Networks: Diagnosis, Causal Investigation, and the Limits of Divergence-Aware Mitigation

• 2026-09-13

  Dynamic Learning Solutions: A System for Personalized Educational Video Generation

Frequent co-authors