© Pint of Science, 2021. All rights reserved.
The future of technology aims for artificial general intelligence (AGI), but how do we get there? Now that machines can identify objects in images, write stories, and fold proteins, we must teach them how to learn to learn. Join our discussion with three researchers on the latest developments in the transfer of knowledge.
Overview of AI and Future Directions
Machine learning is widely applied in the world today with great success, but how will it continue to improve in the years to come? We start this event with an overview of AI and its applications to date, so every attendee gains intuition on how and why AI is used. Then our host, Kate Highnam, will introduce future directions such as multiple agents (federated learning), hierarchy/pipelining models, and meta-level analysis, focusing on the transfer of knowledge with experts in the field.
Learning to Transfer with Janith Petangoda
The transfer of knowledge between humans is continuously developing and improving; the same goes for transferring knowledge between AI systems. Janith Petangoda, a PhD student at Imperial College London, shall highlight what is currently done in the field to transfer knowledge and the challenges these methods must overcome. His work primarily targets methods to quantify transferability between tasks.
Learning to Learn with Isak Falk
When given an unfamiliar task, we (as humans) extrapolate from past experiences and other sources of knowledge to perform decently in the unfamiliar setting. But how would an AGI system do this? Isak Falk, a 2nd year PhD student at UCL studying how AI systems can learn to learn, will take advanced concepts in this area to a simpler level. His diverse experience in this field will provide much needed context on learning to learn as it relates to both the task at hand and the series of tasks to transfer knowledge across.