Why FlyHash is not
a cryptographic hash

FlyHash preserves similarity. Proof of work needs additional properties.

Inspired by the fly olfactory circuit.

FlyHash expands inputs with sparse binary projections, then retains a small set of winners. Similar inputs tend to overlap, which supports approximate similarity search.

Proof of work asks a different question.

PoW lets a node check that a result belongs to a fixed challenge and meets a target. Similarity codes alone neither provide cryptographic security nor prevent precomputation and task-selection bias.

How flybrain combines them.

FlyHash fingerprints connectivity sketches. A challenge binds the full data root, index root and template. Each nonce performs 256 dependent accesses, followed by a domain-separated, length-prefixed SHA-256 transcript.

connectome → sketch → FlyHash fingerprint
challenge + nonce → 256 data-dependent accesses
access trace + FlyHash code → SHA-256 → target check

FlyHash supplies the sparse representation, data accesses contribute work, and SHA-256 supplies the final hash. This combination still requires independent analysis; memory-hardness and ASIC resistance are not automatic.

Utility still needs measurement.

Retrieval needs Recall@K and mAP comparisons against baselines. Verifying computation does not establish scientific retrieval quality.

Reference: Dasgupta, Stevens and Navlakha, A neural algorithm for a fundamental computing problem，2017。