The National Science Centre in Poland has granted deepsense.ai data scientist and University of Warsaw researcher, Piotr Miłoś, 500,000 USD. The project “Reinforcement Learning – contemporary challenges” is designed to help resolve long-standing open problems in training robots and game agents in difficult environments.
The research is a continuation of deepsense.ai’s work on reinforcement learning, which was recognized by the General Chairs of the Conference on Robot Learning (CoRL) as one of 11 papers in the reinforcement learning and robotics category. In that project, members of deepsense.ai’s machine learning team trained a robotic arm to grip a can of coke using reinforcement learning. Recognizing that numerous fundamental ideas have yet to be discovered and understood, the company intends to invest in research and develop its reinforcement learning competences.
The grant, SONATA BIS, is extended to scientists with a strong academic background seeking to establish a new scientific group. Miłoś’ team will comprise students and postdocs from the University of Warsaw, and will be supported by deepsense.ai. Together they will address issues including hierarchical learning, curriculum learning, meta learning, metamemory, model-free and model-based methods.
According to Miłoś, “these are believed to be important concepts in developing reinforcement learning. One of the challenges in AI is the ability to plan hierarchically. deepsense.ai’s intention is to establish a group at the University of Warsaw that is engaged in world-class reinforcement learning research and will bring significant advancement in the field.”
deepsense.io trademarks at boilerplate