This is the edited chat log from the CSIAC Webinar, “Leveraging Machine Learning: How to Achieve the Right Balance Between Humans and Automation to Optimize Outcomes“.
Chris R: Progress from machine learning to deep learning, how does reinforcement learning differ from deep learning? Thank you.
Yogesh M: Reinforcement Learning differs from Supervised and Unsupervised Learning in terms of learning in interactions with environment… based upon reward-response… akin to reinforcement learning in psychology of learning and behavior.
Yogesh M: Deep Learning in all forms listed above depends upon ‘deep’ Neural Networks.
Chris R: Thanks!
Yogesh M: Deep Neural Networks including CNNs, RNNs, and LSTMs, typically have multiple layers of networks… with number of layers represented by “depth” in “deep”.
Yogesh M: Kind of “Nudge”…
Yogesh M: ?
Yogesh M: by Richard H. Thaler
Chris R: Is that how the deep learning trains itself through the validation phase on the sample data set? Reinforcement will “plug and play?”
Yogesh M: Nudge is external influence on behavior as we saw in the “Kale” example… influencing behavior.
Chris R: Thanks, I’ll check it out!
Yogesh M: Separately, DL uses Training Set to Train, Test Set to Check on new instances, and Validation is the next step.
Chris R: Thereby taking the human trainer out of the MBE loop?
Yogesh M: Humans are always in the loop unless it is a complete autonomous AI system.
Chris R: And does RL achieve that or are they still required for training/reinforcement? Thanks!
Yogesh M: Alpha Go Zero which advances on Alpha Go uses RL, here is a paper:
Yogesh M: AlphaGo becomes its own teacher: a neural network is trained to predict AlphaGo’s own move selections and also the winner of AlphaGo’s games. This neural network improves the strength of the tree search, resulting in higher quality move selection and stronger self-play in the next iteration. Starting tabula rasa, our new program AlphaGo Zero achieved superhuman performance, winning 100–0 against the previously published, champion-defeating AlphaGo.
Chris R: Excellent, thanks!
Frank S: Any reading recommendations?
Yogesh M: The notion of “Meaning” seems missing from the Persona graph: here is a related paper: Malhotra, Y., Expert Systems for Knowledge Management: Crossing the Chasm between Information Processing and Sense Making, Expert Systems with Applications: An International Journal, 20(1), 7-16, 2001. http://www.brint.org/expertsystems.pdf.
Yogesh M: ML is a very wide topic, lot of depth and breadth… here is one starting intro article I use for industry execs new to ML: https://qz.com/1046350/the-quartz-guide-to-artificial-intelligence-what-is-it-why-is-it-important-and-should-we-be-afraid/ ; https://www.mckinsey.com/business-functions/mckinsey-analytics/our-insights/an-executives-guide-to-ai
Yogesh M: For the latest book on ML/DL which Elon Musk recommends: MIT Press: Full Text online: http://www.deeplearningbook.org/
Frank S: Thank You,
Thomas S: Excellent…timely
Chris R: Yes, thanks again!
Yogesh M: Thoughts about Alpha Go Zero which beat Alpha Go using Reinforcement Learning?
Yogesh M: Thoughts on current ongoing shift of Machine Learning to Unsupervised Learning using Neural Evolution Search instead of the Back Propagation (Gradient Descent) Algorithm? Latest from Google and Geoffrey Hinton in focus on Capsule Networks?
Yogesh M: Nature: Prediction of cardiovascular risk factors from retinal fundus photographs via deep learning
Yogesh M: Major problems with current ML/DL are often associated with Back Propagation (Gradient Descent) Algorithm – that’s where key future advances are hoped to be.
Yogesh M: Thank you so much!
Del H: What would be your one question to a vendor who is marketing a product with machine learning capabilities?
Chris R: Excellent, thanks!
Yogesh M: Related question: on the issue of “interpretability” where it is difficult to attribute specific [cause-effect] rationale to the ‘black box’ nature of AI/ML that can benefit health concerns?
Yogesh M: Related presentation on Model Risk Management of such Autonomous ISR systems: https://ssrn.com/abstract=3111837.
Yogesh M: Thank you!!
Edwin H: thank you
Will J: Thank you all @Metaversial and I will get answers from my bigger brained colleagues
Gregory H: Thanks a lot Richard.
William M: Thanks Richard!
Alex O: Thank You.