github.com/dovrine/neuralink-compression-challenge

Neuralink Compression Challenge

content.neuralink.com/compression-challenge/data.zip is one hour of raw electrode recordings from a Neuralink implant.

This Neuralink is implanted in the motor cortex of a non-human primate, and recordings were made while playing a video game, like this.

Compression is essential: N1 implant generates ~200Mbps of eletrode data (1024 electrodes @ 20kHz, 10b resolution) and can transmit ~1Mbps wirelessly. So > 200x compression is needed. Compression must run in real time (< 1ms) at low power (< 10mW, including radio).

Neuralink is looking for new approaches to this compression problem, and exceptional engineers to work on it. If you have a solution, email [e-mail] Leaderboard Name Compression ratio Compressed size ./encode size ./decode size zip 2.2 63M 231K 480K Task

Build executables ./encode and ./decode which pass eval.sh. This verifies compression is lossless and measures compression ratio.

Your submission will be scored on the compression ratio it achieves on a different set of electrode recordings. Bonus points for optimizing latency and power efficiency

Submit with source code and build script. Should at least build on Linux. Data $ ls -lah data/ total 143M 193K 0052503c-2849-4f41-ab51-db382103690c.wav 193K 006c6dd6-d91e-419c-9836-c3f320da4f25.wav ...

Uncompressed monochannel WAV files.
5 seconds per file.

NOTE: eval.sh is modified from the original to include the bin folder when calling the executables

tested on arch linux

requirements

• golang 1.21+

setup

chmod +x eval.sh
chmod +x setup.sh
./setup.sh

run

./eval.sh