......................................................................................298 Chapter 50. PyTorch Release 19.07............................................................................ paper is in the backbone. Specifically, the VGG model is obsolete and has been replaced by the ResNet50 model. This model script is available on GitHub and NGC. ‣ Neural Collaborative Filtering (NCF) model: paper. This model script is available on GitHub and NGC. ‣ ResNet50 v1.5 model: This model is a modified version of the original ResNet50 v1 model. This model script is available on GitHub and NGC. ‣

1/50 C++ Modules - Getting Started Today Andreas Weis Woven by Toyota CppCon 20232/50 Introduction3/50 About me - Andreas Weis (he/him) ComicSansMS Co-organizer of the Munich C++ User Group Currently Currently working as a Vehicle Architect for Woven by Toyota4/50 A familiar example... — File: my function.hpp char const* my_function (); — File: my function.cpp char const* my_function () { return println("{}", my_function ()); }5/50 #include can happen anywhere — File: a.hpp inline int my_function #include — File: a impl.hpp { return 42; }6/50 Including files twice does not

machine learning on embedded devices Adrian Stanciu adrian.stanciu.pub@gmail.com CppCon, 2023 1 / 50About me ◮ I am a software engineer from Romania ◮ I have a bachelor’s degree in computer science from global leader in cybersecurity, where I develop security software solutions which run on routers 2 / 50Disclaimer ◮ This presentation is based on a personal project I worked on in my spare time and it is proof of concept ◮ Any mistakes are mine 3 / 50Agenda ◮ Motivation ◮ Image classification ◮ Hand gesture recognition ◮ Summary ◮ Q&A 4 / 50Motivation 5 / 50Machine Learning (ML) ◮ Subfield of Artificial

expertise? Answered: 240 Skipped: 2 Total 240 Beginner Intermediate Expert 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Answer Choices Responses Beginner Intermediate Expert 2 / 37 2015 Jenkins Architect QA Executive IT Group Manager / Team Lead Other (please specify) 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Answer Choices Responses Software Developer DevOps Build Manager Project 240 Skipped: 2 Total 240 1-10 11-49 50-99 100-499 500-999 1000+ 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Answer Choices Responses 1-10 11-49 50-99 100-499 500-999 1000+ 4 / 37 2015

Source: Company disclosures Details on Page 55 6MM 2005 2025 Number of Developers, MM 0% 50% 100% Internet LLM 33 Years In 90% @ Year 3 90% @ Year 23 10/22 4/25 800MM Big Six* USA Technology Number of New Models Released Each Year AI Technology Compounding = Numbers Behind The Momentum 0 50 100 2017 2018 2019 2020 2021 2022 2023 2024 Includes models from • xAI • Anthropic • Meta • NVIDIA University & Stanford Law School sources, iRobot, TechCrunch, BBC, OpenAI. Data aggregated by BOND. 10/50: Alan Turing creates his Turing Test to measure computer intelligence, positing that computers

-O3 ×30 ×30 0% 1% 2% 3% 4% 0 25 50 75 100 Time (s) Percent of Observed Runtimes Version ⌧O2 ⌧O3 Comparing optimizations-O2 -O3 0% 1% 2% 3% 4% 0 25 50 75 100 Time (s) Percent of Observed -O2 -O3 0% 1% 2% 3% 4% 0 25 50 75 100 Time (s) Percent of Observed Runtimes Version ⌧O2 ⌧O3 Is faster than ?-O2 -O3 0% 1% 2% 3% 4% 0 25 50 75 100 Time (s) Percent of Observed 0% 1% 2% 3% 4% 0 25 50 75 100 Time (s) Percent of Observed Runtimes Version ⌧O2 ⌧O3 Is faster than ?If = -O2 -O3 0% 1% 2% 3% 4% 0 25 50 75 100 Time (s) Percent

4m50s echo-b-host-67446447f7-chvsp 1/1 Running 0 4m50s host-to-b-multi-node-clusterip-78f9869d75-l8cf8 1/1 Running 0 4m50s host-to 4m50s pod-to-a-59b5fcb7f6-gq4hd 1/1 Running 0 4m50s pod-to-a-allowed-cnp-55f885bf8b-5lxzz 1/1 Running 0 4m50s pod-to- 4m48s pod-to-a-l3-denied-cnp-64c6c75c5d-xmqhw 1/1 Running 0 4m50s pod-to-b-intra-node-845f955cdc-5nfrt 1/1 Running 0 4m49s pod-to-

output below, the sparsified compressed matrix is smaller than the regular compressed matrix by nearly 50%. weights = np.random.normal(size=(100, 100)).astype(np.float32) sparsity_rate = 0.4 # The percentage scores. Saliency Scores In the paper Optimal Brain Damage1, LeCun et al. suggested that as much as 50% of the connections (weights) from a large network could be safely removed with minimal performance for sparse training using TFMOT (Tensorflow Model Optimization) library. In this case, we prune the 50% of the weights in each prunable block using magnitude-based pruning. Note that the below code is in

align: "center" }; 20 21 var h = this.game.add.text(this.game.width/2, this.game.height/2 + 50, text, style); 22 h.anchor.set(0.5); 23 }, 24 update: function() { 25 if(this.game.input "#fff", align: "center" }; 5 this.scoreLabel = this.game.add.text(this.game.width-50, this.game.height - 50, text, style); 6 this.scoreLabel.fixedToCamera = true; 7 } 1 this.scoreLabel.fixedToCamera tileHeight; 45 result.push(element); 46 } 47 }); 48 return result; 49 }, 50 //create a sprite from an object 51 createFromTiledObject: function(element, group) { 52 var