In the Laboratory of Intelligent Systems we create novel AI engines as well as build robust and adaptive intelligent systems. Current AIs can solve 19×19 versions of Go but behave poorly on easier 9×9 versions of the same game. Similarly, image recognition algorithms can reach 96% accuracy (supra-human) on tests and be fooled by only one pixel change. In other words, current AI lacks the robustness and adaptation present in even simple living beings. AI is based on engines that allows it to learn and reason over things, this lab builds novel engines based on different paradigms to reach high levels of robustness and adaptiveness intrinsically. Interestingly, by increasing the robustness and adaptiveness, other problems like Transfer Learning, One-Shot Learning would also be solved at the same time, igniting, possibly, a new age of intelligent systems.
Member
The Main Research Topics
Next Generation AI
Bioinspired Adaptive and Robust AI
Action/Image Recognition ・Machine Learning・Reinforcement Learning・Self-Learning
Applications：AI Benchmarks（Games included), Humanoid Robots, Autonomous Driving