{"slug": "fitting-a-neural-texture-decoder-with-es-no-backprop", "title": "Fitting a neural texture decoder with ES (no backprop)", "summary": "A developer prototyped a toy neural texture compression system trained entirely with Evolution Strategies, avoiding backpropagation and derivatives. The setup pairs a 128×128×8 latent texture with a 1035-weight MLP decoder to reconstruct a 512×512 image at 32.2 dB PSNR, with the latent quantizing to 8 bits at only 0.04 dB loss. Training took 150 seconds on CPU using a single ~600-line C++ file.", "body_md": "This is something I prototyped for fun in an evening, and now the idea is Prior Art:\n\nToy neural texture compression, trained entirely with \n\n[Evolution Strategies](https://en.wikipedia.org/wiki/Evolution_strategy)\n (no backprop, no derivatives). A 128×128×8 latent texture + a 1035-weight MLP decoder reconstruct a 512×512 image (kodim23 crop) at 32.2 dB PSNR. Latent quantizes to 8 bits with only 0.04 dB loss: ~4 bpp raw, ~3.3 bpp entropy coded. Trained in 150 s on CPU. Single C++ file, ~600 lines, stb_image for I/O.", "url": "https://wpnews.pro/news/fitting-a-neural-texture-decoder-with-es-no-backprop", "canonical_source": "https://richg42.blogspot.com/2026/09/fitting-neural-texture-decoder-with-es.html", "published_at": "2026-09-03 07:06:41+00:00", "updated_at": "2026-09-14 17:25:20.226379+00:00", "lang": "en", "topics": ["machine-learning", "neural-networks", "computer-vision", "ai-research"], "entities": ["Evolution Strategies", "MLP", "stb_image"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/fitting-a-neural-texture-decoder-with-es-no-backprop", "markdown": "https://wpnews.pro/news/fitting-a-neural-texture-decoder-with-es-no-backprop.md", "text": "https://wpnews.pro/news/fitting-a-neural-texture-decoder-with-es-no-backprop.txt", "jsonld": "https://wpnews.pro/news/fitting-a-neural-texture-decoder-with-es-no-backprop.jsonld"}}