LSLT: Saahiti Potluri (MATH major)

Lunch will be served by 12:15. Please arrive early to get your food so we can start the talk on time!

Comparison of Solutions to Catastrophic Forgetting

Abstract: Continuous learning is the ability to learn tasks sequentially. Humans are excellent at this, and importantly, humans can retain previously learned information while integrating new knowledge. Artificial neural networks generally cannot, and experience catastrophic forgetting in which performance on previous tasks deteriorates when new tasks are learned. We examine two solutions to the catastrophic forgetting problem and how they can be interpreted in the architecture of the neural network.