Fast Neural Networks… a no brainer!
summary
Clear claim: a working neural net implemented in F# that maps neurons to agent-based neurons and composes layers as functions, then parallelizes the matrix-heavy work with a MapReduce-like pattern and optional CUDA. Demonstrations include an elastic‑ring approach to TSP and a stock‑price estimator, with code on GitHub and measured gains (example run drops from ~3s to ~400ms, roughly 3x speedup). Slides and runnable source accompany the results.
AI-written summary · about
chapters (29)
- 0:00Start
- 0:13Agenda
- 0:46Introduction - Riccardo Terrell
- 3:19Machine Learning Algorithms
- 4:12Unsupervised & Supervised Learning
- 5:00Neural Networks are multipurpose
- 5:28Microsoft Azure ML Cheat Sheet
- 6:05Problems NN solve
- 6:34Recommendation systems
- 7:11NN Prediction Example
- 7:59What is this Neural Network?
- 8:37Is about learning/growing
- 9:09What is this?
- 10:31What is Neural Network
- 12:28Mathematical Background
- 12:50Some vocabulary
- 13:33Neural Network Classification
- 15:19What's going on?
- 17:22The perceptron
- 19:19Activation functions
- 20:07Neural Network Training
- 20:58NN Tips
- 24:51Neural Network Agent based
- 29:25Neural Network is Functional
- 29:53Neural Network in code
- 34:10Why a Parallel Neural Network?
- 35:19Parallel NN as Map Reduce
- 38:17The Traveling Salesmen Santa Problem
- 42:51Summary
related talks
37mNeural Networks are Programs, Too
01h 02mMachine Learning with F#, Redux
33mLearn Neural Networks With Go - Not Math!
01h 51mUnderstanding Neural Networks (Somewhat)
25mConvergence of AI, Simulations and HPC
41mUnveiling the simple math that powers AI
36mBrain's Neural Network as a service
30mAccelerating Neural Networks for Vision Systems via FPGAs