Скачать с ютуб Unlocking the Future of AI: Multi-Task Reinforcement Learning Explored with Ahmed Hendawy в хорошем качестве

Unlocking the Future of AI: Multi-Task Reinforcement Learning Explored with Ahmed Hendawy

Multi-task reinforcement learning allows an agent to solve multiple problems in one shot. A critical question is how to ensure the diversity of the learned skills. Ahmed tells us about his new method, which uses orthogonalisation to ensure diversity. #reinforcementlearning #agi #ai #machinelearning #gpt4 Paper discussed: https://arxiv.org/pdf/2311.11385v1.pdf GPT4's Poem: In the realm of machines, a new dawn unfurls, Multi-task learning, a world it swirls. Agents learning more than just one task, In the light of knowledge, they bask. Across varied domains, they stride, In their artificial minds, insights reside. Reinforcement learning, their guiding light, Balancing challenges, with all their might. From gaming realms to real-world quests, They adapt, improve, and face all tests. A symphony of algorithms, in harmony they blend, Towards smarter futures, their efforts bend. Multi-tasking learners, in data they dive, From each experience, they learn, they thrive. A dance of metrics, rewards, and time, In the world of AI, they gracefully climb. 00:00 What is Multi-Task Reinforcement Learning 02:25 Related Works 07:26 Mutli-Task Reinforcement Learning with Orthogonal Representations 13:59 Experiments and Results 38:28 Questions and Discussion 47:03 Ahmed's View on the Review Process