Readable implementation of Mamba 3 SSM model

Docker containers for AI models on NVIDIA DGX Spark (GB10, SM121, aarch64). TurboQuant KV cache compression + mamba-ssm aarch64 build.

Deploy Mamba-SSM on NVIDIA Jetson with TensorRT support

A physics-informed Deep Learning framework (Mamba/Swin-UNet) for Sentinel-1 SAR imagery denoising and speckle suppression. Features unsupervised refinement and multi-task learning.

MambaQuant - A production-ready stock prediction tool based on Mamba SSM. Features yfinance integration for global markets (US/TW/Crypto), sliding window inference for T+1 prediction, and optimized CUDA pipelines.

Listening Between the Lines: An explainable multimodal framework for MCI detection from spontaneous speech. Leverages Selective State Space Models (Mamba) and Gated Fusion to integrate linguistic disfluencies and eGeMAPS biomarkers across multi-corpus benchmarks (Pitt, ADReSS, TAUKADIAL)

Computational phenomenology study of semantic satiation in neural networks. Comparing how GPT-2, BERT, and Mamba handle extreme repetition reveals causal models drift into hallucination while bidirectional models stay stable—suggesting attention directionality preserves semantic identity.