Md Mahabub Uz Zaman

Aging and multi-omics Research

Achieved high-accuracy chronological age estimation from mouse brain metabolomic profiles using an Elastic Net regression model, which demonstrated superior performance with an MAE of 2.5, RMSE of 2.63, and an R2 of 0.925.
Developed and trained an explainable CNN-Transformer deep learning model on Raman spectroscopy data to classify exosomes from 11 mammal species with 97% accuracy, using the attention mechanism to identify key spectral regions.
Reviewed and validated the data processing and methodology for a large-scale metabolomics study comparing bowhead whales to terrestrial mammals, ensuring the scientific rigor of findings related to their unique longevity and physiological adaptations.
Currently developing and executing a comprehensive transcriptomics project on mouse testes, from conducting high-throughput omics experiments to performing detailed RNA-seq data analysis.

Studied RSV schemes for elections under bribery attacks in multi-agent distributed systems, showing strong inapproximability results for k-approval and k-veto, where k≥ 2. Designed greedy algorithm for BRSV-plurality with unit bribery costs and PTAS algorithms for BRSV-plurality and BRSV-veto for arbitrary bribery costs.
Formalized attack tolerance problem as a bi-level integer program against voter-deletion attack in multi agent consensus system and show that it is a difficult class problem. For constructive and destructive protection problem we present PTAS algorithms.
Formalized Honeypot deployment in production networks as a combinatorial optimization problem, proved its NP-hardness, provided a near-optimal PTAS algorithm, as well as analyzed against attacker’s reconnaissance cabability

Developed a multi modal learning model using ADNI & TADPOLE dataset to classify the diseases based on brain images, EHR data, image report, biomarker data etc combining LLM, Graph learning, knowledge representation, Knowledge fusion etc.

Designing a machine unlearning model with 86.3% accuracy, combining auxiliary learning in federated learning with selective pruning of neurons to comply with privacy protection and protect users from Byzantine attacks during training.

Built a transformer-based neural combinatorial optimization model equipped with multi-head attention along with pointer networks mechanisms to solve the network interdiction problem, leading to solutions falling within a 2OPT range
Modeling and working with reinforcement learning for combinatorial optimization (RL4CO) for classical algorithms falling in the domain of routing (tsp, vrp), scheduling, assignments (Knapsack), graph (maxcut, flp)etc.
β-robust stochastic scheduling using fixed parameter tractable (fpt) approximation scheme and dynamic programming
Designed a multi-echelon OPT distribution System with a Zone Skipping Concept using MIP for IISE Case Competition 2020
•Forecasting demand in analytics dashboard using ARIMA and LSTM, received $1000 travel grant, ASCM Case Competion 2020
•Created a multi-objective optimization problem formulation for a sequential decision-making processes as well as simulated a digital twin of 5 facilities and compared their efficiency using Python and Gurobi during INFORMS case competition in 2020.

Page updated

Google Sites

Report abuse