Comparing Elixir and Python when working with Simple Neural Networks
summary
Comparison of Elixir using Nx and Axon versus Python with Keras trains MNIST in ~20s (Elixir) vs 15s (Python) and CIFAR-10 in ~28s vs 23s; raw Elixir without Nx previously took ~8.5 hours. A simple benchmark tool (Unix-stats) records CPU/GPU/RAM/VRAM. Nx + Axon makes Elixir viable for small CNNs but averages ~20% slower and often uses more memory; GPU usage behavior depends on the dataset and needs deeper profiling.
AI-written summary · about
chapters (12)
- 0:00Elixir vs Python comparison
- 3:01Neural networks theory
- 9:25Project overview and datasets
- 11:36NX and Axon definitions
- 14:02Benchmarking setup
- 15:51MNIST performance
- 18:20CIFAR-10 performance
- 20:53CPU and RAM usage
- 23:44GPU and VRAM usage
- 28:25Conclusions and future work
- 31:55Acknowledgments
- 33:25Q&A
related talks
25mApplied Machine Learning in Elixir
45mKeynote: Panel(...)
50mCreating a scalable (...) with Elixir, Python & Tensorflow
32mLearn Elixir: Building a Neural Network from Scratch
28mNetworkX is Fast Now: Graph Analytics Unleashed
42mMachine Learning with Elixir and Phoenix
20mScaling Edge AI with Elixir and Nerves
39mAccelerate large-scale machine learning with NP on MXNet