Reinforcement Learning at 65,000 Feet with Marlos C. Machado | Approximately Correct Podcast

Discover the secret to training AI with less data! On this episode of Approximately Correct, we talk with Amii Fellow and Canada CIFAR AI Chair Lili Mou about the challenges of training large language models and how his research on Flora addresses memory footprint concerns. 

AI-powered prosthetics are changing lives, but it takes more than just technology. In this episode, Amii Fellow and Canada CIFAR AI Chair is here to talk about the work going on in his lab designing AI controls for bionic limbs and explains the unique partnership between researchers and users that's creating AI learning that is made with humans in mind. Approximately Correct: An AI Podcast from Amii is hosted by Alona Fyshe and Scott Lilwall. It is produced by Lynda Vang, with video production by Chris Onciul.

How can machine learning revolutionize healthcare? In this episode, Amii Fellow and Canada CIFAR AI Chair Russ Greiner explores how AI is transforming survival prediction, giving doctors and patients personalized health insights that were never before possible. From creating tailored survival curves to improving treatment decisions, Greiner reveals the groundbreaking potential of AI in medicine.

In this episode, reinforcement learning legend Rich Sutton argues the focus on non-continual learning over the past 40 years is now holding AI back. Listen to one of the leading minds in machine learning explain what needs to change.  Approximately Correct: An AI Podcast from Amii is hosted by Alona Fyshe and Scott Lilwall, produced by Lynda Vang, with video production by Chris Onciul

Artificial intelligence has been a part of video games for decades. But new advances in machine learning and generative AI could radically change how we make — and play — games. In this episode, Andrew Butcher, co-founder of Artificial Agency, talks about how the company is creating technology to make games more playful, intelligent, and fun. Tune in to discover how AI is changing game development and what that means for developers and players alike.

A team of Amii researchers recently  published a paper in the prestigious scientific journal Nature investigating a mysterious problem in deep learning that can hinder long-term continual learning. In this Sidebar episode, we talk to two of the co-authors about the problem of Loss of Plasticity, what their findings mean for advanced AI, and the journey from idea to Nature. Read more about Loss of Plasticity: https://www.amii.ca/latest-from-amii/amii-researchers-investigate-ai-mystery-new-nature-paper-loss-plasticity/

Reinforcement learning is being used to make water treatment more efficient. On this week’s episode, we’re joined by Amii Fellow and Canada CIFAR AI Chair Martha White to talk about co-founding RL Core Technologies, which is exploring how RL can be used in to increase efficiency in water treatment plants and other industrial control systems. Find out more about how AI is being used in the real world and its potential large-scale impact.

How do we get the best results when AI and human beings work together? In this episode of Approximately Correct, we’re looking into Human-In-The-Loop (HITL) AI with Matt Taylor. The Amii Fellow and Canada CIFAR AI Chair talks about the importance of human input in AI decision making, the need to recognize the strengths and weaknesses of both natural and artificial intelligence, and how he thinks HITL will be vital if people are to trust AI in their lives. Production Credits: Lynda Vang, - Producer, Chris Onciul - Video Production,  Music Credits: Main Theme - Brooklyn Bridge by Lunareh

Medicine is built on data. Clinical studies, patient charts, test results, X-rays - data helps diagnose us when we’re unhealthy and understand how to treat us. Amii Fellow and Canada CIFAR AI Chair Ross Mitchell explores how machine learning can use this data, pulling out insights to help medical professionals work better and provide better patient outcomes. In this Sidebar episode, we present a conversation between Mitchell and machine learning scientist Jubair Sheik about AI's massive potential in medicine and the increasing accuracy of medical AI models. Production Credits: Lynda Vang, - Producer, Chris Onciul - Video Production,  Music Credits: Main Theme - Brooklyn Bridge by Lunareh

This week’s episode of Approximately Correct looks at how the work of Marlos C. Machado is taking reinforcement learning to new heights. Marlos sits down with hosts Alona Fyshe and Scott Lilwall to share the insights from his work using AI to control balloons in Earth's stratosphere, and what it teaches us about how reinforcement learning works in the real world. Production Credits: Lynda Vang - Producer, Chris Onciul - Video Production Music Credits: Main Theme - Brooklyn Bridge by Lunareh