The Path Ahead
This fast-track introductory guide was designed to get you acquainted with the field of DL using Keras in the fastest yet most effective way. I hope you had a great time on this journey. In this final chapter, we will take a brief look at the path ahead.
- PDF / 197,565 Bytes
- 16 Pages / 439.37 x 666.142 pts Page_size
- 33 Downloads / 208 Views
The Path Ahead This fast-track introductory guide was designed to get you acquainted with the field of DL using Keras in the fastest yet most effective way. I hope you had a great time on this journey. In this final chapter, we will take a brief look at the path ahead. We will try to answer the following question: what additional topics are important for a data scientist to ace the DL journey? Let’s get started.
What’s Next for DL Expertise? We have covered the fundamentals in DL with DNNs for classification and regression. The most interesting part and in fact the major reason why DL gained its popularity and momentum in 2012 was DL for computer vision. A few years back, designing an algorithm that could help a computer in making sense out of an image was almost impossible. The idea of using algorithms to extract meanings from an image or classifying the image into a particular class was unimaginable. As time passed, ML became popular and the approach of using handcrafted features in images and then using a classifier for training the algorithm showcased improved results, but this was nowhere what we would want it to be. In 2012, Alexnet (an architecture developed by Alex Krizhevsky, Ilya Sutskever, and Geoffrey Hinton) was used to compete in the “ImageNet Large Scale Visual Recognition Challenge.” This was a competition hosted for developing algorithms that could learn and predict on classifying images © Jojo Moolayil 2019 J. Moolayil, Learn Keras for Deep Neural Networks, https://doi.org/10.1007/978-1-4842-4240-7_6
161
Chapter 6
The Path Ahead
into a defined set of classes. Alexnet achieved a top-five error of 15.3%; this was almost 11% lower than the previous best score and set a historic record in the challenge. The architecture was a type of DNN architecture especially used for image classification. That is when DL got noticed and immediately became a hot topic for research. The journey of DL from there onward skyrocketed. With more research and experiments on DL, the field got extended to video, audio, text, and pretty much any form of data. Today, DL is ubiquitous. Almost every major tech company has embraced DL in its entire stack of offerings. A small step for you as a DL enthusiast in exploring advanced DL topics would be to first start with DL for computer vision. This is where you will explore convolutional neural networks (CNNs).
CNN CNNs are the class of DL algorithms used for computer vision use cases like classifying an image or a video and detecting an object within an image or even a region within an image. CNN algorithms were a huge breakthrough in the field of computer vision, as it required a bare minimum of image processing compared to the other prevalent techniques of the time and also performed exceptionally well. The performance improvement with CNN for image classification was phenomenal. The process of building CNN is also simplified in Keras, where all the logical components are neatly abstracted. Keras provides CNN layers, and the overall process of developing CNN models is qu
Data Loading...
