Page 263 - AWSAR 2.0

comparison with humans?
RQ.2. What are the exact challenges of teaching machines to learn from few examples of visual concepts?
RQ.3. How can the deep generative models be used to learn concepts end to end rather than only feature learning?
The answers to the above questions will enable us to improve FSL because we can incorporate human-like high cognitive processing in machines. We are
also working on constructing a framework that organizes the machine learning tasks as per their cognitive difficulty. This framework will also provide a structured comparison of human and machine cognition based on the implementation of cognitive processes such as memory and attention.
Principle ideas and research progress so far
The exploratory literature
review conducted has
provided initial understanding
of challenges in FSL (RQ.2).
Thus far we have published
a comprehensive literature
review and analysis of diverse
few shot learning approaches Kadam and Vaidya [2019] and have answered the RQ.2 to some extent. For RQ.1, we are proposing a framework to assess machine cognition by evaluating cognitive level of the tasks it can perform. We are addressing RQ.3 by first understanding the role of generative deep models as generic learner. We intend to capture the generic learning capability of
Ms. Suvarna Kadam || 239
humans in machines. Generic learning is not dependent on any specific task, for example, when we learn to write in one language, we can practically draw alphabets of any language.
Applications of proposed research
Just the way we say, “A picture worth of thousand words!” we can say, ”A real example worth a thousand claims!” about any new research. We believe the best way
to communicate the research usefulness is by discussing some simple application cases.
Conclusion and future direction
In this article, we discussed why we are passionate about improving the learning in machines. We also discussed the challenges we face while teaching machines. Learning real world visual concepts with just a few samples is hard even for humans. But we learn to generalize when we don’t have enough information. Our research focus on implementing this challenging human trait into machines. We briefly discussed the literature review for FSL and how to improve deep learning
based FSL with cognitive approach. We then discussed our experimentation using deep generative models for simple visual concept learning with only few examples. We believe that enabling powerful machine learning methods, such as deep learning to learn from few examples is a high impact research with several applications across domains.
For use as a source of biofuel, many universities and research organizations are working on a variety of easy-to-manipulate microalgae, for increased lipid production. Our research work was focussed on improving the various steps involved in microalgal biofuel production and, thereby, enhancing the productivity. Microalgal biofuel production consists of four major steps: cultivation, harvesting, lipid extraction and transesterification.