Adapting to Change: How Machine Intelligences Adapt to a Changing World • Robert Crowe • GOTO 2022

This presentation was recorded at GOTO Amsterdam 2022. #GOTOcon #GOTOams Robert Crowe - TensorFlow Developer Advocate at Google ABSTRACT Change is perhaps the only constant in our world, and it requires us to adapt to new realities. Depending on the rate of change and the amount of change, the requirement to adapt often results in our brains increasing levels of stress hormones as they try to adjust. To adapt to changes in the world that it lives in, our brain needs to make changes in what it’s learned, which means making changes in the neural structures that have captured that learning. Sometimes in fact it can be more difficult for our brains to relearn something than it is for them to learn it for the first time. Failing to adapt to change has been the downfall of many people, and even whole species. While this is true for human intelligence, it’s also true for machine intelligence. But because of the isolated nature of machine intelligences it’s even more difficult for them to detect and adapt to change. In the vast majority of cases these intelligences are very limited in their ability to perceive and adapt to change, even in very limited, specialized areas of learning, which requires humans to design and direct their adaptation to change. Failing to adapt to change can lead to poor decision making, and even catastrophic consequences. This talk will explore: • How both human and machine intelligences adapt to change, including the state of the art and industry best practices for adapting machine intelligence to change • Techniques for making change less stressful for both machines and their humans [...] TIMECODES 00:00 Intro 00:55 Change can be stressful 01:55 Do I need to adapt? 02:51 Mental adaptation 04:01 Neocortex 05:47 Cortical columns 06:45 Reference frames 09:54 Machine intelligence 10:05 Artificial neural networks 11:00 Supervised learning 12:32 Prediction aka inference 14:26 Expectations 15:27 Machine expectations 16:45 Labeling data 17:50 Consequences 18:17 MLOps 19:41 Production ML 21:35 Tales from the trenches 21:53 Production ML research 22:29 Process 22:40 What is MLOps? 23:10 CI, deployment & training 24:35 Training & deploying models 25:29 TFX production components 26:12 MLOps Level 0 29:46 Outro Read the full abstract here: RECOMMENDED BOOKS Phil Winder • Reinforcement Learning • Kelleher & Tierney • Data Science (The MIT Press Essential Knowledge series) • Lakshmanan, Robinson & Munn • Machine Learning Design Patterns • Lakshmanan, Görner & Gillard • Practical Machine Learning for Computer Vision • Aurélien Géron • Hands-On Machine Learning with Scikit-Learn, Keras & TensorFlow • #TensorFlow #TFX #AI #ML #MLOps #DeepLearning #ArtificialIntelligence #MachineLearning #ReinforcementLearning #DataScience #MachineIntelligence #Programming #Change #NeuralNetworks #ArtificialNeuralNetwork #Inference #Data #CI #ContinuousIntegration #Deployment Looking for a unique learning experience? Attend the next GOTO conference near you! Get your ticket at Sign up for updates and specials at SUBSCRIBE TO OUR CHANNEL - new videos posted almost daily.