PacSec2020 ai-security_overview
0 likes60 views
This document provides an overview of artificial intelligence and security topics. It discusses the history of AI including important early works and algorithms developed from 1958 to 2014. It also summarizes different types of neural networks like convolutional neural networks, generative adversarial networks, and recurrent neural networks. The document outlines common machine learning techniques including supervised learning, unsupervised learning, and reinforcement learning. It briefly describes different AI hardware options and frameworks used for development. Finally, it mentions two examples of AI interacting with other AI systems: a one pixel attack and the DARPA Cyber Grand Challenge competition.
PacSec2020 ai-security_overview
- 1. AI and Security panel PacSec2020 Nov 06, 2020
- 2. The Classics • Perceptron 1958 • Back Propagation 1986 • Image recognition with Convolutional Neural Network 1989 • Recommender System with Collaborative Filtering 1994 • AlexNet 2012 • Word2Vec 2013 • Generative Adversarial Networks 2014 • Deep Q Learning 2014 • etc -
- 3. Artificial Neural Network (in text books) Gohsuke Takama https://towardsdatascience.com/
- 5. Artificial Neurons but Gohsuke Takama - Multi Layer Perceptron (MLP) Neurons
- 6. Neural Network types which? • Convolutional Neural Network (CNN) • AutoEncoder (Convolution-DeConvolution) • Generative Adversarial Network (GAN) • Graph Neural Network • Recurrent Neural Network (RNN) • Long Short Term Memory (LSTM) • Attention Network • Transformers • Performers • etc -
- 7. Learning which? • Supervised Learning • Un-Supervised Learning • Reinforcement Learning • Self Supervised Learning • Transfer Learning • One Shot Learning • etc -
- 8. AI Hardware which? • CPU • GPU • FPGA • ASIC • Neuromorphic chip (coming soon?) • Bio/Silicon Hybrid (?because Neuralink ) -
- 9. AI Hardware which? CPU, GPU, FPGA, ASIC Gohsuke Takama - CPU GPU FPGA ASIC ○ → →→ →→→ by Software by Software Hardware Logic Circuit Hardware Logic Circuit Programming Languages Programming Languages on Logic Definition Fixed by Design ○ ○ FeedForward Network oriented Fixed by Design x86 ARM NVIDIA (CUDA) AMD Xilinx (AMD), Lattice, Altera (Intel) Google TPU speed execution programming design flexibility chips
- 10. Development Frameworks which? • TensorFlow : Google • PyTorch : Facebook • MXNet : Amazon, Apple? • CNTK : Microsoft (Windows) • DL4J (Java, Android) • CoreML (iOS) • ONNX : Fb + MS (interoperability oriented) • Keras (entry level friendly) • Ludwig : Uber (code-free framework) • Azure ML Studio : Microsoft (code-free framework) -
- 11. AI vs AI One Pixel Attack Gohsuke Takama https://youtu.be/SA4YEAWVpbk https://github.com/Hyperparticle/one-pixel-attack-keras
- 12. AI vs AI DARPA Cyber Grand Challenge 2016 Gohsuke Takama https://www.darpa.mil/program/cyber-grand-challenge