Anecdotes from 11 Role Models in Machine Learning
Human-in-the-Loop Machine Learning: Active Learning and Annotation for Human-Centered AI. The book features anecdotes from 11 machine learning experts. Each expert shared an anecdote about data-related problems they encountered building and evaluating machine learning models in real-world situations. If you are early in your career, then I hope you can relate to many of the anecdotes in the book, which are shared here.
DeepMind RL method promises better co-op between AI and humans
AI researchers at DeepMind present a new technique to improve the capacity of reinforcement learning agents to cooperate with humans at different skill levels. The technique is called Fictitious Co-Play (FCP) and it does not require human-generated data to train the RL agents. FCP created RL agents that provided better results and caused less confusion when teamed up with humans. The findings can provide important directions for future research in human-AI systems.
Comparing Hinton’s GLOM Model to Numenta’s Thousand Brains Theory
Geoffrey Hinton recently published a paper “How to Represent Part-Whole Hierarchies in a Neural Network” and presented a new theory called GLOM. The Thousand Brains Theory is a sensorimotor theory that models the common circuit in the neocortex and suggests a new way of thinking about how our brain works. We learn a model of the world by observing how our sensory inputs change as we move.
Q&A with Niels Leadholm, Numenta Visiting Scholar 2020
Niels Leadholm is a PhD student at the Oxford Lab for Theoretical Neuroscience and Artificial Intelligence. He spent 12 weeks with Numenta as a visiting research scholar. His interest is in understanding primate vision at a computational level, and using this understanding to improve artificial systems. His research looks at what is known about how our brains process visual information at both a high level and a low level.
New tech aimed at improving public health and accessibility
In the era of smart cities, taking on public health and the accessibility of healthcare is a different game. From transportation to healthcare, tech is here to help streamline accessible services and promote better public health. As a pandemic ravages public health, how can technology make a real and measurable difference? It all comes down to the tools and practices health professionals, business leaders, and public officials employ when integrating advancing technologies with accessible public services.
Supporting COVID-19 policy response with large-scale mobility-based modeling
Mobility restrictions, from stay-at-home orders to indoor occupancy caps, have been utilized extensively by policymakers during the COVID-19 pandemic. These reductions in mobility help to control the spread of the virus, but they come at a heavy cost to businesses and employees. To balance these competing demands, policymakers need analytical tools that can evaluate the tradeoffs between mobility and infections. Such tools should be fine-grained, able to test out heterogeneous plans.
Inside Boston Dynamics’ project to create humanoid robots
The video of a robot that jumps over obstacles has been released on YouTube. It is the latest in a series of videos of robots that have gone viral. The video shows some of the challenges of creating robots that can be taught to be successful. The videos have been viewed more than 1,000 times on social media in the past year. The YouTube video shows the challenges faced by engineers and their creations.
Is DeepMind’s new reinforcement learning system a step toward general AI?
DeepMind researchers claim to have taken the “first steps to train an agent capable of playing many different games without needing human interaction data. The new project includes a 3D environment with realistic dynamics and deep reinforcement learning agents that can learn to solve a wide range of challenges. The researchers say the new system is an ‘important step toward creating more general agents with the flexibility to adapt rapidlyâ€
What Matters in Learning from Offline Human Demonstrations for Robot Manipulation
Imitation Learning is a promising approach to endow robots with various complex manipulation capabilities. By allowing robots to learn from datasets collected by humans, robots can learn to perform the same skills that were demonstrated by the human. However, a lack of open-source human datasets and reproducible learning methods make assessing the state of the field difficult. The study analyzes the most critical challenges when learning from offline human data for manipulation.
A look at industry demand for data scientists
Data scientists wrangle and analyze structured and unstructured data sets to produce actionable plans that businesses can apply to greater success. Glassdoor ranked data science as the #2 job in America for 2021. By 2025, experts predict that by 2025, the field of data science will be on an upward trend. By analyzing this trend, we can observe where opportunities may be most widely available and how data scientists can make the most of these shifts.
Nvidia, University of Toronto are making robotics research available to small firms
A robotic hand that could manipulate objects as we do would be enormously useful in factories, warehouses, offices, and homes. Despite tremendous progress in the field, research on robotics hands remains extremely expensive and limited to a few very wealthy companies and research labs. New research promises to make robotics research available to resource-constrained organizations. In a paper published on arXiv, researchers at the University of Toronto, Nvidia, and other organizations have presented a new system that leverages highly efficient deep reinforcement learning techniques.
Ever thought of tracking your metabolism? Let’s talk Lumen
Lumen gathers data from the gases of your breath, translates the information through an app into plain English for you and me. Lumen is a device that “hacks” your body data to provide unique information about what fuel your body is employing. The Lumen device is based on determining an individual’s metabolic fuel usage according to a measurement called Respiratory Quotient, or RQ.
Machine Learning Safety: Unsolved Problems
There remain critical challenges in machine learning that, if left resolved, could lead to unintended consequences and unsafe use of AI in the future. As an important and active area of research, roadmaps are being developed to help guide continued ML research and use toward meaningful and robust applications. We provide a new roadmap for ML Safety and refine the technical problems that the field needs to address.
OpenAI’s Approach to Solve Math Word Problems
Yesterday’s edition of The Sequence highlighted OpenAI’s latest research to solve math word problems. Mathematical reasoning has long been considered one of the cornerstones of human cognition. Solving this problem does not only entail reasoning through the text but also orchestrating a sequence of steps to arrive at the solution. OpenAI models are based on GPT-3 for language understanding and use two fundamental methods to optimize its capabilities for math problems.
Numenta Demonstrates 100x Speed-up of Deep Learning Networks | June 2021
Jeff is a keynote speaker at the 2021 Beijing Academy of Artificial Intelligence Conference. He will discuss the key components of the Thousand Brains Theory and his insights on the current and future AI landscape. We recently announced that we achieved 100x performance improvements on inference tasks in deep learning networks without any loss in accuracy, using sparse-sparse implementations. We expect to see additional benefits in generalization, robustness, and sensorotor behavior.
Reinforcement learning frustrates humans in teamplay, MIT study finds
AI systems that use reinforcement learning algorithms have outperformed human world champions in games such as chess and StarCraft. MIT researchers at MIT Lincoln Laboratory studied cooperation between humans and AI agents in the card game Hanabi. The findings highlight some of the underexplored challenges of applying reinforcement learning to real-world situations and can have important implications for future development of AI systems meant to cooperate with humans.
Google’s new deep learning system can give a boost to radiologists
Deep learning can detect abnormal chest x-rays with accuracy that matches that of professional radiologists. It can also serve as a first response tool in emergency settings where experienced radiologists are not available. The findings show that, while deep learning is not close to replacing radiologists, it can help boost their productivity. The paper also shows how far the AI community has come to build processes that can reduce the risks of deep learning models.
A Conversation on A Thousand Brains and Pedagogy: Exploring Narrative Argument with Big Ideas
In A Thousand Brains, Jeff Hawkins describes the notion of reference frames that are built and refined by the neocortex as central to learning. In this video series, we were joined by Dr. Michael Riendeau and his two students, Ranger Fair and Jacob Shalmi from Eagle Hill School. We explored how the ideas of the theory can be beneficial to educators and students, and the possible pedagogical implications of those ideas.