AI
Generalizing from simulation
Researchers have developed a way for robot controllers trained entirely in simulation to adapt to unexpected changes in their physical environment. Th...
AI
AI
Asymmetric actor critic for image-based robot learning
Researchers at OpenAI have developed an asymmetric actor critic (A2C) algorithm for image-based robot learning. This method allows robots to learn fro...
AI
Sim-to-real transfer of robotic control with dynamics randomization
Researchers have developed a method to improve the performance of robots by transferring skills learned in simulation to real-world environments. This...
AI
Domain randomization and generative models for robotic grasping
Researchers at OpenAI have developed a method to improve robotic grasping using domain randomization and generative models. The approach involves trai...
AI
Learning Word Embedding
Word embedding is a technique used to transform free-text words into numeric values that can be processed by machine learning models. One-hot encoding...
AI
Meta-learning for wrestling
Researchers have developed a meta-learning algorithm that enables an artificial intelligence (AI) to quickly learn and adapt in simulated robot wrestl...
AI
Competitive self-play
Researchers at OpenAI have found that simulating AIs playing against themselves can help them learn physical skills. In one example, they used this me...
AI
Nonlinear computation in deep linear networks
Researchers have found a way to perform nonlinear computations using deep linear networks. This is significant because it challenges the conventional ...
AI
Anatomize Deep Learning with Information Theory
Professor Naftali Tishby's concept of Information Bottleneck (IB) method applies information theory to deep neural networks. He proposed a new learnin...
AI
Learning to model other minds
Researchers at OpenAI have developed an algorithm called LOLA (Learning with Opponent-Learning Awareness) that takes into account the fact that other ...
AI
Learning with opponent-learning awareness
OpenAI researchers have proposed a new approach to learning called 'opponent-learning awareness'. This method involves training models to be aware of ...
AI
From GAN to WGAN
Generative adversarial networks (GANs) have shown promise in generating realistic content such as images and music by competing between a generator an...