The rsync algorithm (1996) [pdf]
16 by vortex_ape | 1 comments on Hacker News.
Friday, January 2, 2026
Thursday, January 1, 2026
New top story on Hacker News: Show HN: Feature detection exploration in Lidar DEMs via differential decomp
Show HN: Feature detection exploration in Lidar DEMs via differential decomp
3 by DarkForestery | 0 comments on Hacker News.
I'm not a geospatial expert — I work in AI/ML. This started when I was exploring LiDAR data with agentic assitince and noticed that different signal decomposition methods revealed different terrain features. The core idea: if you systematically combine decomposition methods (Gaussian, bilateral, wavelet, morphological, etc.) with different upsampling techniques, each combination has characteristic "failure modes" that selectively preserve or eliminate certain features. The differences between outputs become feature-specific filters. The framework tests 25 decomposition × 19 upsampling methods across parameter ranges — about 40,000 combinations total. The visualization grid makes it easy to compare which methods work for what. Built in Cursor with Opus 4.5, NumPy, SciPy, scikit-image, PyWavelets, and OpenCV. Apache 2.0 licensed. I'd appreciate feedback from anyone who actually works with elevation data. What am I missing? What's obvious to practitioners that I wouldn't know?
3 by DarkForestery | 0 comments on Hacker News.
I'm not a geospatial expert — I work in AI/ML. This started when I was exploring LiDAR data with agentic assitince and noticed that different signal decomposition methods revealed different terrain features. The core idea: if you systematically combine decomposition methods (Gaussian, bilateral, wavelet, morphological, etc.) with different upsampling techniques, each combination has characteristic "failure modes" that selectively preserve or eliminate certain features. The differences between outputs become feature-specific filters. The framework tests 25 decomposition × 19 upsampling methods across parameter ranges — about 40,000 combinations total. The visualization grid makes it easy to compare which methods work for what. Built in Cursor with Opus 4.5, NumPy, SciPy, scikit-image, PyWavelets, and OpenCV. Apache 2.0 licensed. I'd appreciate feedback from anyone who actually works with elevation data. What am I missing? What's obvious to practitioners that I wouldn't know?
Wednesday, December 31, 2025
Tuesday, December 30, 2025
Monday, December 29, 2025
New top story on Hacker News: Show HN: Per-instance TSP Solver with No Pre-training (1.66% gap on d1291)
Show HN: Per-instance TSP Solver with No Pre-training (1.66% gap on d1291)
5 by jivaprime | 0 comments on Hacker News.
OP here. Most Deep Learning approaches for TSP rely on pre-training with large-scale datasets. I wanted to see if a solver could learn "on the fly" for a specific instance without any priors from other problems. I built a solver using PPO that learns from scratch per instance. It achieved a 1.66% gap on TSPLIB d1291 in about 5.6 hours on a single A100. The Core Idea: My hypothesis was that while optimal solutions are mostly composed of 'minimum edges' (nearest neighbors), the actual difficulty comes from a small number of 'exception edges' outside of that local scope. Instead of pre-training, I designed an inductive bias based on the topological/geometric structure of these exception edges. The agent receives guides on which edges are likely promising based on micro/macro structures, and PPO fills in the gaps through trial and error. It is interesting to see RL reach this level without a dataset. I have open-sourced the code and a Colab notebook for anyone who wants to verify the results or tinker with the 'exception edge' hypothesis. Code & Colab: https://ift.tt/BNY5jbS Happy to answer any questions about the geometric priors or the PPO implementation!
5 by jivaprime | 0 comments on Hacker News.
OP here. Most Deep Learning approaches for TSP rely on pre-training with large-scale datasets. I wanted to see if a solver could learn "on the fly" for a specific instance without any priors from other problems. I built a solver using PPO that learns from scratch per instance. It achieved a 1.66% gap on TSPLIB d1291 in about 5.6 hours on a single A100. The Core Idea: My hypothesis was that while optimal solutions are mostly composed of 'minimum edges' (nearest neighbors), the actual difficulty comes from a small number of 'exception edges' outside of that local scope. Instead of pre-training, I designed an inductive bias based on the topological/geometric structure of these exception edges. The agent receives guides on which edges are likely promising based on micro/macro structures, and PPO fills in the gaps through trial and error. It is interesting to see RL reach this level without a dataset. I have open-sourced the code and a Colab notebook for anyone who wants to verify the results or tinker with the 'exception edge' hypothesis. Code & Colab: https://ift.tt/BNY5jbS Happy to answer any questions about the geometric priors or the PPO implementation!
Subscribe to:
Comments (Atom)