Humans learn new things with a very small set of examples — e.g. a child can generalize the concept of a “Dog” from a single picture but a machine learning system needs a lot of examples to learn its features. In this session, we will go over recent advanced Deep Learning approaches on training a model in data scarce situations.
Human brains have capacity to learn new things within small number of examples, in particular, when presented with stimuli, people seem to be able to understand new concepts quickly and then recognize variations on these concepts in future percepts. Machine learning as a field has been highly successful at a variety of tasks such as classification, web search, image, and speech recognition. Often times, however, these models do not do very well in the regime of low data. This is the primary motivation behind Few-Shot Learning; to train a model with fewer examples but generalize to unfamiliar categories without extensive retraining. In this session, we will go over recent advances made in the field of few-shot learning using deep learning and learn how it can be modified for applications in industries like medicine, agriculture, and automation.
Key Features of this session: 1. Learn how you can speed up the deep learning process with one-shot learning 2. Use Python and PyTorch to build state-of-the-art one-shot learning models 3. Explore architectures such as Siamese networks, memory-augmented neural networks, and model-agnostic meta-learning.
This session will include learning about 3 form of deep learning approaches for few-shot learning: 1. Metrics Based Methods, (Code Link: https://github.com/shruti-jadon/Hands-on-One-Shot-Learning/tree/master/Ch02-MetricsBasedMethods) 2. Models Based Methods, and (Code Link: https://github.com/shruti-jadon/Hands-on-One-Shot-Learning/tree/master/Ch03-ModelsBasedMethods) 3. Optimization Based Methods. (Code Link: https://github.com/shruti-jadon/Hands-on-One-Shot-Learning/tree/master/Ch04-OptimizationBasedMethods)
Book Link: https://www.amazon.com/Hands-One-shot-Learning-Python-implementing-ebook/dp/B07S9QWNG2
What you will learn 1. Get to grips with the fundamental concepts of one- and few-shot learning 2. Work with different deep learning architectures for one-shot learning 3. Understand when to use one-shot and transfer learning, respectively 4. Implement one-shot learning approaches based on metrics, models, and optimization in Jupyter Notebook using PyTorch 5. Discover different optimization algorithms that help to improve accuracy even with smaller volumes of data 6. Explore various one-shot learning architectures based on classification and regression
Who's this session for? If you're an AI researcher or a machine learning or deep learning expert looking to explore few-shot learning, this session is for you. It will help you get started with implementing various few-shot techniques to train models faster. Some Python programming experience is necessary to understand the concepts covered in this session.