Md Kishor Morol

Research profile

Research summary

Md Kishor Morol is a Bangladesh-based AI researcher at American International University-Bangladesh whose profile, name-resolved from OpenAlex bibliographic data, shows a fast-moving and deliberately broad applied agenda. Three Evolve-reviewed 2026 arXiv preprints split evenly across the very fields the record reports — zero-shot evaluation of vision-language models (IchthyoNoma), UAV remote sensing (AAMBERS-UAV), and LLM agent security (CAPTURE) — with Morol in last-author position on two, a middle position on one, and no corresponding authorship. The underlying OpenAlex record lists 85 works in the last three years, including 50 dated 2026 — up from 2 per year in 2020-2021 — which drives a maximum activity score of 100. Topic counts concentrate on advanced neural network applications, sentiment analysis, and AI in healthcare and education (16 each), with topic modeling (12) and natural language processing techniques (10) behind. Collaboration is comparatively narrow: 5 distinct coauthors in the recent window yield a collaboration score of 36, anchored by five recurring partners. Across the 3 papers Evolve has reviewed, the reproducibility mean is 70, with the fish-recognition audit rated highest for readiness. The h-index stands at 12 with an i10-index of 12; composite output and impact scores are not yet computed.

Research keywords

Highlights

• Acquisition exposure flips multimodal rankings in UAV weedy rice segmentation

  Evolve's review of AAMBERS-UAV finds that acquisition exposure reverses SegFormer's RGB-vs-fusion ranking while leaving U-Net's unchanged; the same-test design is judged sound, though only 2 backbones and 1 U-Net seed constrain the scope.

• Zero-shot audit of CLIP, BioCLIP and BioCLIP2 on Bangladeshi freshwater fish

  IchthyoNoma tests zero-shot biological vision-language models on 7 Bangladeshi freshwater fish classes, carefully controlling nomenclature, Bengali naming and context effects; Evolve rates the audit solid but narrow in scope, with high reproducibility readiness.

• Disentangling preference drift from memory poisoning in personalized LLM agents

  CAPTURE introduces a memory gateway that distinguishes legitimate preference drift from poisoning, reporting an 11.5% attack success rate and an 83.5% compliance rate; Evolve calls the evaluation rigorous and honest about limitations, though the core evidence is synthetic.

Timeline

1. 20202 works recorded in the OpenAlex year counts — the earliest year in the provided record.
2. 202214 works recorded, an early ramp up from the 2020-2021 baseline of 2 per year.
3. 202414 works recorded, matching the 2022 level just before the recent acceleration.
4. 202521 works recorded — the last full year before the 2026 peak.
5. 202650 works dated 2026 in the record; three arXiv preprints — AAMBERS-UAV, IchthyoNoma and CAPTURE — enter the Evolve catalog with Morol in last or middle author position.

Papers in the Evolve catalog

• 2026-09-04

  AAMBERS-UAV: Acquisition-Aware Multimodal Backbone Evaluation and Ranking for UAV Weedy Rice Segmentation

• 2026-09-03

  IchthyoNoma: Nomenclature and Context Sensitivity of Zero-Shot Biological Vision--Language Models for Bangladeshi Freshwater Fish Recognition

• 2026-09-02

  CAPTURE: Disentangling Preference Drift from Memory Poisoning in Personalized LLM Agents

Frequent co-authors