Teaching an artificial intelligence system to make meaningful decisions is not easy. Reinforcement learning models, which support these systems, sometimes fail when faced with minor variations in tasks they’ve been trained to carry out. To improve the reliability of these models for complicated tasks, researchers at MIT have developed a better algorithm to train them.
The algorithm has been designed to choose the best tasks to train an artificial intelligence agent on so it can perform every task effectively.
For their technique, the researchers selected some tasks then independently trained an algorithm for every task. They chose tasks that would improve the overall performance of the algorithm on all tasks based on an algorithm dubbed Model-Based Transfer Learning.
This algorithm demonstrates how well every algorithm would perform if it was independently trained for a single task as well as how the performance of every algorithm would diminish if its tasks were changed. This allowed model-based transfer learning to estimate the value of training on new tasks.
The researchers then used zero-shot transfer learning to gauge how well the model performed in new tasks without any additional training. Once this was done, they tested their method on simulated tasks, which included managing speed advisories in real-time and controlling traffic signals.
They discovered that their method was five to fifty times more efficient than standard techniques on various simulated tasks. This improvement in efficiency allows the algorithm to learn better solutions in a timely manner, ultimately enhancing the AI agent’s performance.
Prof. Cathy Wu, the senior author of the study, explained that an algorithm that wasn’t complex stood a better chance of adoption as it was easier for others to understand and easier to implement. Wu is also a member of the Laboratory for Information and Decision Systems.
The researchers now plan to design model-based transfer learning algorithms that can be used to handle more complex issues. This is in addition to applying their approach to real-world issues, particularly in next-gen mobility systems.
The lead author of the study is CEE graduate student, Jung-Hoon Cho. Other researchers involved include IDSS graduate student Sirui Li; and Vindula Jayawardana, a graduate student in the Electrical Engineering and Computer Science department. Their findings will be presented at the Conference on Neural Information Processing Systems.
The study was partly funded by an Amazon Robotics PhD Fellowship, the Kwanjeong Educational Foundation PhD Scholarship Program, and a National Science Foundation CAREER Award.
With many companies like Intel Corp. (NASDAQ: INTC) also engaged in the development of AI solutions to different challenges, the world could soon see a lot more breakthrough approaches to handling existing problems.
About AINewsWire
AINewsWire (“AINW”) is a specialized communications platform with a focus on the latest advancements in artificial intelligence (“AI”), including the technologies, trends and trailblazers driving innovation forward. It is one of 70+ brands within the Dynamic Brand Portfolio @ IBN that delivers: (1) access to a vast network of wire solutions via InvestorWire to efficiently and effectively reach a myriad of target markets, demographics and diverse industries; (2) article and editorial syndication to 5,000+ outlets; (3) enhanced press release enhancement to ensure maximum impact; (4) social media distribution via IBN to millions of social media followers; and (5) a full array of tailored corporate communications solutions. With broad reach and a seasoned team of contributing journalists and writers, AINW is uniquely positioned to best serve private and public companies that want to reach a wide audience of investors, influencers, consumers, journalists, and the general public. By cutting through the overload of information in today’s market, AINW brings its clients unparalleled recognition and brand awareness.
AINW is where breaking news, insightful content and actionable information converge.
To receive SMS alerts from AINewsWire, text “AI” to 888-902-4192 (U.S. Mobile Phones Only)
For more information, please visit www.AINewsWire.com
Please see full terms of use and disclaimers on the AINewsWire website applicable to all content provided by AINW, wherever published or re-published: https://www.AINewsWire.com/Disclaimer
AINewsWire
Los Angeles, CA
www.AINewsWire.com
310.299.1717 Office
Editor@AINewsWire.com
AINewsWire is powered by IBN