Applying Transfer Learning on Your Data
Amita Kapoor (~amita) |
Humans have great ability to generalize; we can very efficiently apply the knowledge we learned in classrooms to real world problems. Transfer learning provides a similar capability to artificial neural networks. The Workshop will introduce the concept of 'Transfer Learning' described by Andrew Ng, the leading expert in Machine Learning as the "next driver of ML commercial success." Transfer Learning is important because training deep neural networks from scratch have two key requirements. First, we need a large labeled dataset; the second one requires computationally efficient hardware (GPUs). While such immensely large data exists for some tasks and domains, in most cases the data are usually proprietary or expensive. Transfer Learning, the technique to use models pre-trained on one domain for another problem domain, provides the ability to use DNNs even when the dataset is small. Moreover, Transfer learning requires less computation and thus can be done in respectable time using CPUs as well.
The workshop will cover following topics:
- Introduction to Transfer Learning
- Application of Transfer Learning
- Transfer Learning Scenarios
- Applying Transfer using Keras and Tensorflow
And finally, we will have hands on session demonstrating how to use Xception and Inception networks for Dog breed Recognition.
- Anaconda installed (Python =3.5)
- Tensorflow 1.x
- For the sake of convenience and due to limited time, Speaker, will also provide environment 'yml' files (Windows10, Ubuntu14.04/16.04, Mac OS X)
To make best use of the workshop it would be appreciated if participants are well versed with Anaconda, Python and understand Convolution Neural Networks. The information necessary can be accessed via following links
- How to manage Anaconda Environments: https://conda.io/docs/user-guide/tasks/manage-environments.html
- Convolutional Neural Networks here and here
Amita Kapoor: Amita Kapoor is Associate Professor in the Department of Electronics, SRCASW, University of Delhi. She has been actively teaching neural networks for last twenty years. She did her Masters in Electronics in the year 1996, and her PhD in the year 2011. During the course of her PhD, she was awarded prestigious DAAD fellowship to pursue a part of her research work in Karlsruhe Institute of Technology, Karlsruhe, Germany. She had been awarded best Presentation Award at International Conference Photonics 2008 for her paper. She is a member of professional bodies like OSA (Optical Society of America), IEEE (Institute of Electrical and Electronics Engineers), INNS (International Neural Network Society), ISBS (Indian Society for Buddhist Studies). She has more than 40 publications in the international journals and conferences. Her present research areas include Machine Learning, Artificial Intelligence, Neural Networks, Photonics and Robotics.
Narotam Singh: Narotam Singh has been with India Meteorological Department, Ministry of Earth Sciences, India since 1996. He has been actively involved with various technical programs and training of officers of GOI in the field of Information Technology and Communication. He did his post-graduation in the field of Electronics in 1996 and both Post graduate diploma and Diploma in the field of Computer Engineering, in 1997 and 1994 respectively. He is currently working in the enigmatic field of Neural Networks.