Online Multi-Task Learning Using Biased Sampling
نویسندگان
چکیده
One of the long-standing challenges in Artificial Intelligence for goal-directed behavior is to build a single agent which can solve multiple tasks. Recent progress in multi-task learning for learning behavior in many goal-directed sequential tasks has been in the form of distillation based learning wherein a single student network learns from multiple task-specific teacher networks by mimicking the task-specific policies of the teacher networks. There has also been progress in the form of Progressive Networks which seek to overcome the catastrophic forgetting problem using gating mechanisms. We propose a simple yet efficient Multi-Tasking framework which solves many tasks in an online or active learning setup.
منابع مشابه
Enhanced representation and multi-task learning for image annotation
In this paper we evaluate biased random sampling as image representation for bag of words models in combination with between class information transfer via output kernel-based multi-task learning using the ImageCLEF PhotoAnnotation dataset. We apply the mutual information measure for measuring correlation between kernels and labels. Biased random sampling improves ranking performance of classif...
متن کاملOnline Multi-Task Learning Using Active Sampling
One of the long-standing challenges in Artificial Intelligence for goal-directed behavior is to build a single agent which can solve multiple tasks. Recent progress in multi-task learning for goal-directed sequential tasks has been in the form of distillation based learning wherein a single student network learns from multiple task-specific expert networks by mimicking the task-specific policie...
متن کاملDiscover Relevant Environment Feature Using Concurrent Reinforcement Learning
In order to compare the policies more efficiently, we introduce a new reinforcement learning method called concurrent biased learning. This is a multi-thread learning method, in which each learning thread refers to one feature of the environment. If an agent intentionally focuses on part of these environmental features to learn a policy of a task, we call this method a biased learning; otherwis...
متن کاملOnline Reinforcement Learning Using a Probability Density Estimation
Function approximation in online, incremental, reinforcement learning needs to deal with two fundamental problems: biased sampling and nonstationarity. In this kind of task, biased sampling occurs because samples are obtained from specific trajectories dictated by the dynamics of the environment and are usually concentrated in particular convergence regions, which in the long term tend to domin...
متن کاملUsing Free Energies to Represent Q-values in a Multiagent Reinforcement Learning Task
The problem of reinforcement learning in large factored Markov decision processes is explored. The Q-value of a state-action pair is approximated by the free energy of a product of experts network. Network parameters are learned on-line using a modified SARSA algorithm which minimizes the inconsistency of the Q-values of consecutive state-action pairs. Actions are chosen based on the current va...
متن کامل