MOMENTA — Mixture-of-Experts for Multimodal Misinformation Detection
PyTorch framework combining modality-specific experts, cross-modal alignment, and temporal reasoning for fake-news detection.
- PyTorch
- multimodal
- mixture-of-experts
- misinformation
Things I've built, prototypes, and experiments in progress.
PyTorch framework combining modality-specific experts, cross-modal alignment, and temporal reasoning for fake-news detection.
Alignment-free nucleotide sequence analysis with 267 propensity scales and Lyapunov-based complexity estimation.
Lightweight multi-scale TCN with feature-level attention for robust, real-time EEG emotion recognition under LOSO.
Parameter-efficient LoRA + Mixture-of-Experts fusion of EEG, HRV, and EDA for calibrated fatigue estimation.
Multi-scale temporal convolution with multi-head attention for EEG emotion recognition, with extensive ablations.
Hybrid framework: mechanistic ODEs of nanoparticle-tumor transport, Bayesian MCMC, and LSTM/Neural-ODE validation.
Frequent-itemset mining over hospital CPT/ICD codes to surface treatment patterns and outcome disparities across 13 vulnerable groups.
Event-centric multimodal framework with cross-modal attention, trend-aware temporal modeling, and class-imbalance handling.
PDE/ODE tumor-nanoparticle model with Bayesian MCMC estimation and GPR/RF/GBM predictive comparison.
Data Envelopment Analysis of hospital efficiency combining HCAHPS survey and operational data.
CNN/VGG16 classifier for chest X-rays with Grad-CAM, Integrated Gradients, and adversarial robustness analysis.
Single- and multi-objective evolutionary algorithms for optimizing neural-network training, trading off accuracy and speed.
Classical ML pipeline (XGBoost, SVM, GBM) with feature engineering for ASD screening data.
LSTM neural networks forecasting stock prices from historical data — the intersection of ML and financial analysis.