Apppublic

recursiverall/noiseminer-demo

sourceHugging Facemitupdated 2mo agoView on Hugging Face

0likes

App README

NoiseMiner — Exact Compression-Mining for Structured Noise

Interactive benchmark visualization for the NoiseMiner compression system. Shows AIT 2026 Challenge results across 8 files, comparing Phase 1 (sequential greedy) vs Phase 2 (recursive re-submission) pipelines.

About

NoiseMiner is an exact compression-mining system for structured noise — data that appears random but contains hidden algorithmic structure. Unlike statistical compressors (gzip, zstd), NoiseMiner finds and encodes the deterministic algorithm that generated the data.

What This Demo Shows

—Benchmark Results Table: 8 AIT challenge files, file-by-file compression ratios
—Phase 1 vs Phase 2 Comparison: Honest accounting of when recursive re-submission helps vs hurts
—Miner Registry: All 33 miners (25 v0 + 8 v1) with cost ranks
—File Upload: Run NoiseMiner on your own binary files

Key Findings

—Phase 1 alone is optimal for structured data — Phase 2 recursive re-submission adds overhead on already-compressible content
—All 8 AIT files match exact round-trip verification (zero information loss)
—Net-positive compression on structured files: A (+7.7%), G (+25.4%), H (+1.1%)

Local Setup

bash

pip install -e .  # Install noiseminer package from parent directory
pip install gradio pandas
python app.py