Brain-Inspired Computing - CAM-HDC Project

Project repository for the Brain-Inspired Computing course at the Technical University of Munich (TUM).

What This Repo Contains

• Main implementation: Task7/ (core project code)
• Dataset: language_recognition_dataset/
• Earlier coursework tasks: Task1-Task6, Task4-alternative (kept for reference)

This project studies a CAM-cell-based Hyperdimensional Computing (HDC) pipeline for language recognition, focusing on the accuracy-energy trade-off under different hardware and software hyperparameters.

Core Idea

• Hardware side: sweep CAM-relevant parameters such as supply voltage, block size, and precision.
• Software side: run probabilistic inference with hardware-informed confusion matrices.
• Goal: find Pareto-optimal operating points (higher accuracy, lower energy).

Inference Strategies (Task7)

Task7/strategies includes multiple accuracy/efficiency strategies, including:

• Adaptive precision
• Bit pruning
• Error masking
• Optimal configuration search
• Multi-centroid representation
• Hierarchical early-exit inference
• Voltage compensation
• Approximate CAM modeling
• MRMR feature selection

These strategies improve robustness and/or efficiency by adapting computation to noise, confidence, and feature relevance.

Quick Start

cd Task7
python main_runner.py --quick
python run_all_strategies.py --quick

Notes

• If you want, Task7 can be renamed to something clearer (for example, cam-hdc), but path references in scripts/config must then be updated consistently.
• Top-level .gitignore is configured to ignore non-Task7 task folders for cleaner repo views.