¿Que paso en Machine Learning en los últimos 6 meses?

Todo lo que paso en Machine Learning en el ultimo semestre del 2015 y principios del 2016.

Transcript

1. ¿Que paso en Machine Learning en los últimos 6 meses? Edgar Marca @matiskay
2. Machine Learning en 5 minutos
3. Aprendizaje No Supervisado (Unsupervised Learning) • No Knowledge of output class or value. • Data is unlabelled or value unknown. • Goal: Determinate data patterns/groupings. • Algorithms: K-means, genetic algorithms, clustering approaches, etc.
4. Aprendizaje Supervisado (Supervised Learning) • Knowledge of output: Learning with the presence of an “expert” / teacher. • Data is labelled with a class or value. • Goal: Predict class or value label. • Algorithms: Neural Networks, Support Vector Machines, Decision Trees, Bayesian Classifier, etc. C1 C1 C1 C1 C2 C2 C2 C2 C2 C2
5. Aprendizaje por Refuerzos (Reinforcement Learning) Reinforcement Learning: An Introduction, Richard S. Sutton and Andrew G. Barto. https://webdocs.cs.ualberta.ca/~sutton/book/ebook/the-book.html Reinforcement Learning is the area of Machine Learning concerned with the actions that software agents ought to take in a particular environment in order to maximize rewards.
6. ¿Que paso en los ultimos 6 meses?
7. DEEP LEARNING
8. Software
9. TensorFlow Features • Deep Flexibility: TensorFlow isn't a rigid neural networks library. • True Portability: runs on CPUs or GPUs, and on desktop, server, or mobile computing platforms. • Auto-Differentiation. • Language Options: Python and C++ • Maximize Performance. TensorFlow: https://www.tensorflow.org/
10. MxNET • Lightweight, memory efficient and portable to smart devices • MXNet is a deep learning framework designed for both efficiency and flexibility. • Support for python, R, C++ and Julia • Cloud-friendly and directly compatible with S3, HDFS, and Azure • Training Deep Net on 14 Million Images by Using A Single Machine.
11. CuDNN 4 • The NVIDIA CUDA Deep Neural Network library (cuDNN) is a GPU-accelerated library of primitives for deep neural networks. • Train neural networks up to 14x faster using Google’s Batch Normalization technique • Increase training and inference performance for convolutional layers up to 2x faster with new 2D tiled FFT algorithm. • Accelerate inference performance for convolutional layers on small batch sizes up to 2x on Maxwell-architecture GPUs. CuDNN4: https://developer.nvidia.com/cudnn
12. Aplicaciones
13. A Neural Algorithm of Artistic Style Paper: http://arxiv.org/abs/1508.06576 Codigo: https://github.com/jcjohnson/neural-style “The key finding of this paper is that the representations of content and style in the Convolutional Neural Network are separable.”
14. A Neural Algorithm of Artistic Style
15. Large Scale Visual Recognition Challenge 2015 (ILSVRC2015) and COCO 2015 Paper: Deep Residual Learning for Image Recognition http://arxiv.org/abs/ 1512.03385 • ImageNet Classification: “Ultra-deep” 152-layers net. • ImageNet Detection: 16% better than 2nd. • ImageNet Localization: 27% better than 2nd. • COCO Detection: 11% better than 2nd. • COCO Segmentation: 12% better than 2nd.
16. Mastering the game of Go with deep neural networks and tree search Link: http://googleresearch.blogspot.pe/2016/01/alphago-mastering-ancient-game- of-go.html
17. Predict the destination of taxi trips based on initial partial trajectories Link: http://blog.kaggle.com/2015/07/27/taxi-trajectory-winners-interview-1st-place- team-%F0%9F%9A%95/
18. ¿Como Mantenerse al día en Machine Learning? • Reddit: http://www.reddit.com/r/machinelearning • Data Tau: http://www.datatau.com/ • Kaggle Blog: http://blog.kaggle.com/ • R Bloggers: http://www.r-bloggers.com/ • NLP People: https://nlppeople.com/category/blog/
19. Preguntas
20. Gracias