spectraMR is a research framework for MRI reconstruction, super-resolution and generative modelling. It spans GAN, diffusion (including cold and latent diffusion), VAE/VQ-VAE, masked-autoencoder and self-supervised pretraining, reconstruction-only, domain-adaptation, physics-driven, disentangled, sensitivity-estimation (PINN), cycle-Bloch, flow, state-space (Mamba), magnetic-resonance-fingerprinting, fMRI and active/learnable-acquisition paradigms behind one configuration schema. Complex k-space is handled through centred, orthonormally-scaled FFT operators rather than raw transforms; configuration is nested, frozen and loaded once; component
Gdihi, Adnane
spectraMR is a research framework for MRI reconstruction, super-resolution and generative modelling. It spans GAN, diffusion (including cold and latent diffusion), VAE/VQ-VAE, masked-autoencoder and self-supervised pretraining, reconstruction-only, domain-adaptation, physics-driven, disentangled, sensitivity-estimation (PINN), cycle-Bloch, flow, state-space (Mamba), magnetic-resonance-fingerprinting, fMRI and active/learnable-acquisition paradigms behind one configuration schema. Complex k-space is handled through centred, orthonormally-scaled FFT operators rather than raw transforms; configuration is nested, frozen and loaded once; component
Gdihi, Adnane
This repository contains the official open-source verification framework, neural architecture implementations, and experiment replication suite for the research paper: "Trustworthy electricity theft detection integrating informative missingness learning and certified conformal audits" (Under review at Scientific Reports, Nature Portfolio). Key Components: 1. Missingness-Aware Multi-Representation Detector (MAMRD): A compact 0.97M-parameter deep learning architecture integrating selective state-space models (Mamba), zero-sum Laplacian differencing, and context-conditional missingness embeddings. 2. Leakage-Free Native-Prevalence Protocol: Stan
Akram, Hafiz Muhammad Azeem, Talatahari, Siamak, Jhangeer, Adil
Source code for DeepGuard-NR, a multimodal deep learning framework for detecting neural-rendering deepfakes. It combines four forensic branches (EfficientNet and DCT artifact analysis, MiDaS-based depth consistency, Mamba state space temporal modeling, and lighting/physics analysis) through a feature-fusion MLP, with Grad-CAM++ explainability. The repository includes the training and evaluation pipeline, a FastAPI backend, and a React frontend. Accompanies the manuscript submitted to Image and Vision Computing.
Prithviraj
CryptoMamba-Sentiment is a sentiment-conditioned selective state space framework for cryptocurrency time-series forecasting. The model extends the Mamba architecture by integrating engineered sentiment signals, including Sentiment Momentum, Sentiment Volatility, and Polarity Ratio, to dynamically modulate state-space parameters. The framework is evaluated on BTC, ETH, and XRP hourly data and demonstrates improved forecasting accuracy and directional prediction performance compared to strong deep learning baselines such as Temporal Fusion Transformer and vanilla Mamba.
hrdeyhimfar
