Mohammed Rakib

Mohammed Rakib

Computer Science PhD Student

Oklahoma State University

My work focuses on designing machine learning architectures for real-world scientific and engineering applications, particularly in high-energy physics and precision agriculture. I also work on improving model efficiency through transfer learning and model-compression methods, including knowledge distillation, quantization, and pruning, to reduce computational cost and enable faster inference. Another area of my research is multimodal learning, with a focus on modality imbalance and the effective integration of information across different data sources. Alongside this research, my industry experience centers on turning models into reliable production systems through scalable APIs, automated deployment pipelines, forecasting services, and observability.

I currently work in rAIson (Reasoning & AI Laboratory) at Oklahoma State University under the supervision of Dr. Atriya Sen.

Résumé Academic CV
Email Google Scholar GitHub LinkedIn
Mohammed Rakib

About

About

Academic research

At Oklahoma State University’s rAIson (Reasoning & AI Laboratory), under the supervision of Dr. Atriya Sen, I develop machine learning methods for scientific and engineering applications. My work includes efficient learning and model compression, multimodal learning and modality imbalance, physics-guided models for high-energy physics, and multimodal methods for precision agriculture. I also teach and mentor students as a Graduate Teaching Assistant in Computer Science.

Industry experience

My industry work covers the full machine learning lifecycle, from data collection and model development to deployment and monitoring. At eBay, I built forecasting and hotspot-analysis systems for storage infrastructure and deployed them with FastAPI, Docker, Kubernetes, Prometheus, and automated reporting workflows. Earlier projects included distributed data ingestion, neural machine translation, and backend platforms on AWS.

Multimodal learning Machine learning systems Computer vision Natural language processing Model compression Distributed data systems
Résumé Academic CV

Updates

Recent News

Academic background

Education

2023 - Present

PhD, Computer Science

Oklahoma State University
Stillwater, Oklahoma

GPA 3.88

2023 - May 2026

MS, Computer Science

Oklahoma State University
Stillwater, Oklahoma

GPA 3.82 Big Data Analytics; Cloud Computing & Distributed Systems; Data Structure & Algorithms II; Database Systems

Sep 2017 - Sep 2021

BS, Computer Science & Engineering

North South University
Dhaka, Bangladesh

GPA 3.96 · Summa Cum Laude

Background

Experience

San Jose, California

eBay

Software Engineer Intern

  • Built an end-to-end ML system for the NuBlox platform using Python, Prometheus/PromQL, and CatBoost to forecast PVCs likely to exceed 90% utilization within 14 days.
  • Developed a hotspot-analysis pipeline across latency, I/O, memory, and capacity signals to identify imbalanced nodes and recommend workload rebalancing, reducing hotspot-related latency and improving storage efficiency.
  • Deployed a 34-endpoint FastAPI service with Docker and Kubernetes, automating four daily CronJobs for mapping, forecasting, reporting, and Slack alerts and cutting manual monitoring time by half.
PythonFastAPIPromQLCatBoostKubernetes
Stillwater, Oklahoma

rAIson (Reasoning & AI Laboratory), Oklahoma State University

Graduate Research Assistant

  • Spearheaded multimodal-learning research and developed G²D with PyTorch to mitigate modality imbalance, resulting in a first-author ICCV 2025 publication.
  • Engineered MIS-ME, a multimodal fusion model for soil-moisture estimation that reduced prediction error by more than 3% and led to a first-author DSAA 2024 publication.
  • Research spans knowledge distillation, computer vision, precision agriculture, and physics-guided learning.
PyTorchMultimodal AIKnowledge distillation
Stillwater, Oklahoma

Department of Computer Science, OSU

Graduate Teaching Assistant

  • Lead labs, grade assignments, review code, mentor students one-on-one, and support course websites and Canvas.
  • Courses include Operating Systems I, Object-Oriented Programming, Database Systems, and Discrete Mathematics, using C++, Java, MySQL, and shell scripting.
TeachingC++JavaDatabases
Dhaka, Bangladesh

Neovo Tech Ltd.

ML Intern → Data Scientist & ML Engineer

  • Built a distributed Celery and RabbitMQ web-crawling system for fintech client Eucaps, containerized with Docker, orchestrated on Kubernetes, and connected to downstream financial-news analytics.
  • Engineered a Hugging Face neural machine translation pipeline with Docker and AWS S3 for multilingual financial updates.
  • Co-developed more than 30 FastAPI endpoints for authentication, scheduling, and Stripe payments in a booking platform.
FastAPICeleryRabbitMQAWS
Dhaka, Bangladesh

Apurba-NSU R&D Lab

Research Assistant

  • Developed and deployed deep-learning models for NLP, QA, NER, ASR, OCR, and computer vision, contributing to publications at ICPR 2022 and ICLR 2023.
  • Researched pruning, quantization, and knowledge distillation; created the COLT iterative pruning algorithm published in IEEE Transactions on Artificial Intelligence.
  • Conducted this research with Dr. Nabeel Mohammed and the Apurba-NSU research team.
NLPComputer visionModel compression

Research output

Publications

Full Google Scholar profile ↗
  1. 2026
  2. 2025
  3. 2025

    PileUp Mitigation at the HL-LHC Using Attention for Event-Wide Context

    L. Vaughan, M. Rakib, S. Patel, F. Rizatdinova, A. Khanov, and A. Bagavathi · PAKDD 2025

  4. 2025

    COLT: Cyclic Overlapping Lottery Tickets for Faster Pruning of Convolutional Neural Networks

    M. I. Hossain, M. Rakib, M. M. L. Elahi, N. Mohammed, and S. Rahman · IEEE Transactions on Artificial Intelligence

  5. 2024

    MIS-ME: A Multi-modal Framework for Soil Moisture Estimation

    M. Rakib, A. A. Mohammed, C. Diggins, S. Sharma, J. M. Sadler, T. Ochsner, and A. Bagavathi · DSAA 2024

  6. 2024

    Exploiting Adaptive Contextual Masking for Aspect-Based Sentiment Analysis

    S. M. Rafiuddin, M. Rakib, S. Kamal, and A. Bagavathi · PAKDD 2024

  7. 2023

    Bangla-Wave: Improving Bangla Automatic Speech Recognition Utilizing N-gram Language Models

    M. Rakib, M. I. Hossain, N. Mohammed, and F. Rahman · ICSCA 2023

  8. 2023

    Automated Mapping of Healthcare Concepts to a Standardized Healthcare Taxonomy

    S. Mollah, M. Rakib, M. Wasek, A. S. A. Rabby, F. Rahman, and N. Mohammed · ICLR Tiny Papers 2023

  9. 2022
  10. 2022

    An Open Source Contractual Language Understanding Application Using Machine Learning

    A. Nawar, M. Rakib, S. A. Hai, and S. Haq · LREC Workshop 2022

  11. 2022

    IoT-Based Air Pollution Monitoring & Prediction System

    M. Rakib, S. Haq, M. I. Hossain, and T. Rahman · ICISET 2022

  12. 2022

Selected work

Research Projects

01KDD 2026

PhyGHT: Physics-Guided HyperGraph Transformer for Signal Purification at the HL-LHC

Designed a physics-guided hypergraph transformer for pileup mitigation, reaching R² = 0.932 for energy and R² = 0.836 for mass correction under extreme pileup.

40.4 ms latency8.7× faster than ParticleNet
View project
PhyGHT model architecture for physics-guided particle pileup mitigation
PhyGHT architecture · Paper
02ICCV 2025

G²D: Boosting Multimodal Learning with Gradient-Guided Distillation

Created an open-source knowledge-distillation framework and Sequential Modality Prioritization technique to counter modality imbalance.

18%+ accuracy gain0.5% memory overhead
View project
G squared D gradient-guided distillation architecture for multimodal learning
G²D architecture · Paper
03DSAA 2024

MIS-ME: A Multi-modal Framework for Soil Moisture Estimation

Fused soil-patch imagery with meteorological data, reducing MAPE by 3.25% over meteorological-only models, 2.15% over image-only models, and at least 1.5% over conventional fusion.

Read the paper
MIS-ME multimodal architecture combining soil images and meteorological data
MIS-ME multimodal architecture · Paper
04ICSCA 2023

Bangla-Wave: Improving Bangla Automatic Speech Recognition Utilizing N-gram Language Models

Fine-tuned wav2vec 2.0 on 399 hours of Bengali speech and added an n-gram post-processor, achieving 4.66% WER and 1.54% CER.

Read the paper
Bangla-Wave speech recognition and language-model correction pipeline
Bangla-Wave pipeline · Paper
05Open source

An Open Source Contractual Language Understanding Application Using Machine Learning

Co-led an end-to-end legal-contract review application, improving RoBERTa-base AUPR by 4% and reaching 20K monthly model downloads.

View project
System diagram for the contract-language review application from document upload to highlighted clauses
Contract-review system · Paper
06Distributed systems

Distributed k-NN with Hadoop MapReduce

Engineered a distributed k-NN classifier from scratch with a memory-efficient priority-queue reducer and benchmarked it on single- and multi-node Hadoop clusters.

Distributed k-NN pipeline using HDFS data blocks, parallel Hadoop mappers, a global top-k reducer, and majority-vote prediction
Distributed k-NN MapReduce pipeline
07PAKDD 2025

PUMiNet: PileUp Mitigation at the HL-LHC Using Attention for Event-Wide Context

Event-wide attention for HL-LHC pileup mitigation; R² of 0.912 for energy fraction and 0.720 for mass fraction, enabling improved Higgs-boson mass reconstruction.

Paper ↗
PUMiNet neural network architecture for event-wide pileup mitigation
PUMiNet architecture · Paper
08PAKDD 2024

Exploiting Adaptive Contextual Masking for Aspect-Based Sentiment Analysis

Gradient-learned masking for aspect term extraction and sentiment classification that outperformed comparison methods on SemEval benchmarks.

Paper ↗
Adaptive contextual masking threshold strategy for aspect-based sentiment analysis
Adaptive masking strategy · Paper
09IEEE TAI 2025

COLT: Cyclic Overlapping Lottery Tickets for Faster Pruning of Convolutional Neural Networks

Class-wise overlapping lottery tickets that required fewer pruning iterations than IMP and transferred across datasets without performance loss.

Paper ↗
COLT cyclic overlapping lottery ticket pruning method
COLT pruning method · Paper
10ICPR 2022

LILA-BOTI: Leveraging Isolated Letter Accumulations by Ordering Teacher Insights for Bangla Handwriting Recognition

Knowledge distillation for Bangla handwriting recognition, improving minor-class F1-Macro by up to 3.5% and overall word recognition by up to 4%.

Paper ↗
LILA-BOTI teacher-student knowledge-distillation pipeline for Bangla handwriting recognition
LILA-BOTI training pipeline · Paper
11Water 2022

Water Level Forecasting Using Spatiotemporal Attention-Based Long Short-Term Memory Network

Spatiotemporal attention LSTM for river forecasting in Bangladesh, improving Dhaka-station accuracy by 3.44%.

Paper ↗
Spatiotemporal attention LSTM architecture for water-level forecasting
Spatiotemporal attention LSTM · Paper
12ICISET 2022

IoT-Based Air Pollution Monitoring & Prediction System

Led a team of three building a sensor-to-cloud system and ARIMA model using 144 hourly observations for next-day forecasts with over 90% reported accuracy.

Paper ↗
Arduino Mega wiring diagram for the IoT air-quality system with dust, ammonia, carbon monoxide, temperature, humidity, and Wi-Fi sensors
IoT sensor wiring diagram · Paper
13Open source

My Reading Room

Django learning platform with document sharing, reading-time tracking, and OpenCV-based engagement monitoring.

My Reading Room use-case diagram showing teacher and student features for classes, reading materials, face detection, and reading analytics
Reading Room use-case diagram · Repository
14Computer vision

AgeDB Age Estimation

Fine-tuned ResNet-152 with CORAL loss to 9.07-year MAE versus the cited DEX result of 13.1, using 20× fewer samples; CORAL outperformed cross-entropy.

CORAL age-estimation architecture with a ResNet backbone, extended ordinal labels, and weight-shared binary tasks
CORAL age-estimation architecture · Repository

Recognition

Awards & Honors

  • 2026
    Google Cloud Research Credits

    $1,000 award supporting research computing.

  • 2024
    OSU Graduate College Student Travel Award

    $400 award supporting research dissemination.

  • 2022
    Bengali ASR Competition Runner-up

    DL Sprint, BUET CSE Fest.

  • 2021
    Summa Cum Laude

    Graduated from NSU with a 3.96 GPA.

  • 2021
    5th place, Project Showcasing

    MIST ICT Innovation Fest.

  • 2017
    Academic Excellence Award

    75% scholarship for the CSE bachelor’s degree at NSU.

Academic contributions

Service & Teaching

  • Peer reviewer
    BMVC 2026, CVPR 2025, ICCV 2025, and IJCNN 2024.

  • Workshop leader
    Led an OSU DataBytes workshop on multimodal learning and hands-on PyTorch and deep-learning workshops at NSU.

  • Student mentor
    Mentored teams at the OSU ACM Appathon 2025 ↗.

  • Volunteer
    Supported fundraising for SCARS to assist underprivileged communities.

Contact

Contact

mohammed.rakib@okstate.edu LinkedIn
Stillwater, Oklahoma, USAOpen to research and engineering conversations