---
title: "COMPASS-Inspired Semantic Sampling for Sudanese Arabic Dialect Understanding"
emoji: "🎯"
colorFrom: "blue"
colorTo: "indigo"
sdk: gradio
sdk_version: 4.36.0
app_file: app.py
pinned: false
license: mit
---

# COMPASS-Inspired Semantic Sampling Demo

This Space demonstrates adaptive semantic sampling for low-resource dialect understanding, inspired by the COMPASS paper (2604.20720). It clusters Sudanese Arabic text by semantic meaning and prioritizes under-represented clusters for more efficient learning.

## Hypothesis
Semantic clustering with distribution-aware sampling improves Sudanese Arabic dialect coverage compared to random sampling, requiring fewer examples to achieve equivalent semantic diversity.

## Method
1. Generate/embed sample Sudanese dialect sentences
2. Cluster by semantic similarity
3. Compare random vs COMPASS-style adaptive sampling
4. Measure cluster coverage and semantic diversity

## Paper Reference
COMPASS: COntinual Multilingual PEFT with Adaptive Semantic Sampling (arXiv:2604.20720)