Implementing Neural Networks with SQL
alxmrs · hn · 2026-07-14
The author demonstrated how to "build neural networks in SQL" using Xarray-SQL:
- By mapping N-dimensional arrays to two-dimensional table structures, many geographic/climate calculations can be treated as relational operations.
- They also used Claude Code to design a benchmark, arguing that common 100TB-scale geo/climate operations can fundamentally be expressed as relational queries.
- Most notably, the author suggests that regridding can be formulated as sparse matrix-vector multiplication, and matrix multiplication can be represented using SQL syntax like SUM(val val) ... JOIN ... GROUP BY.
- They further propose that if even these array/physical post-processings can be pushed down to the database, the applicability of databases might be far broader than traditionally understood.
More from Research
- ArtiFixer to appear in SIGGRAPH Reconstruction session on Wednesday — ZGojcic · 2026-07-21
- NeurIPS 2026 workshop will focus on on-device intelligence and local execution — YiMaTweets · 2026-07-21
- NeurIPS 2026 workshop calls papers on on-device intelligence — YiMaTweets · 2026-07-21
- AI Security Institute says every tested model tried to cheat in cyber evaluations — connoraxiotes · 2026-07-21
- AI companies are buying old books to avoid training on AI-generated slop — CackleRooster · 2026-07-21
- Sakana says multiple diffusion models plus MCTS beat test-time scaling on coding and math — SakanaAILabs · 2026-07-21