离线奖励塑造与扩展人类偏好反来进行深度强化学习
Jinfeng Li1, Biao Luo1, Xiaodong Xu1
1School of Automation, Central South University, Changsha 410083, China.
概括
本研究引入了基于偏好的强化学习 (PbRL) 的新反方法. 这些方法捕捉了微妙的人类偏好,导致人工智能系统中更准确的奖励函数学习.
科学领域:
- 人工智能的人工智能
- 机器学习 机器学习
- 机器人技术 机器人技术 机器人技术
背景情况:
- 奖励函数设计对于将AI行为与人类意图对齐至关重要.
- 基于偏好的强化学习 (PbRL) 使用人类的偏好来奖励学习.
- 目前的PbRL方法往往缺乏偏好强度和动态反.
研究的目的:
- 开发新的反方法,以便在PbRL中更准确地学习人类偏好.
- 改进学习奖励函数与人类意图的调整.
- 解决现有的PbRL中非动态和强度不可知反的局限性.
主要方法:
- 提出了缩放偏好 (SP) 反方法.
- 引入了定性和定量缩放偏好 (Q2SP) 反方法.
- 使用对基准控制和机器人任务的离线数据实施和评估方法.
主要成果:
- SP和Q2SP方法使人类能够表达轨迹之间的偏好程度.
- 更详细的反显著提高了学习奖励函数的准确性.
- 实验结果显示,与基线和最先进的方法相比,其性能具有竞争力.
结论:
- 新的反方法提高了PbRL中奖励学习的准确性.
- 详细的偏好反是将人工智能与人类意图更好地协调的关键.
- 拟议的方法为复杂任务中的人类-人工智能协作提供了有希望的进步.
相关概念视频
Role of Shaping in Operant Conditioning
267
Shaping is a technique used in operant conditioning to train complex behaviors by rewarding successive approximations toward the target behavior. This method is necessary because organisms are unlikely to perform complex behaviors spontaneously. Instead, shaping breaks down the desired behavior into small, manageable steps.
The steps involved in shaping begin with reinforcing any response that resembles the desired behavior. For example, parents might praise a child for picking up one toy. As...
The steps involved in shaping begin with reinforcing any response that resembles the desired behavior. For example, parents might praise a child for picking up one toy. As...
267
Reinforcement
180
Positive and negative reinforcement are key concepts in operant conditioning, a learning process where the consequences of a behavior affect the likelihood of that behavior being repeated.
Positive reinforcement occurs when a behavior is followed by the presentation of a rewarding stimulus, increasing the frequency of that behavior. For example:
Positive reinforcement occurs when a behavior is followed by the presentation of a rewarding stimulus, increasing the frequency of that behavior. For example:
180
Reinforcement Schedules
132
Positive reinforcement is a powerful method for teaching new behaviors to both animals and humans. B.F. Skinner demonstrated this with his experiments using rats in a Skinner box. When a rat pressed a lever, it received a food pellet. This immediate reward encouraged the rat to repeat the behavior. This method, where a reward follows every instance of the behavior, is known as continuous reinforcement. It is highly effective for establishing new behaviors quickly.
Once a behavior is learned,...
Once a behavior is learned,...
132
Law of Effect
1.3K
B.F. Skinner, a prominent figure in behavioral psychology, introduced operant conditioning by emphasizing the role of consequences in shaping behavior. This theory builds upon the law of effect proposed by Edward Thorndike, which posits that behaviors followed by satisfying outcomes are likely to be repeated. In contrast, those followed by unsatisfying outcomes are less likely to recur.
Edward Thorndike's foundational work involved studying learning in animals, particularly using puzzle...
Edward Thorndike's foundational work involved studying learning in animals, particularly using puzzle...
1.3K
Operant Conditioning
1.5K
Operant conditioning, a key concept in behavioral psychology, involves using reinforcement and punishment to alter the likelihood of a behavior being repeated. B.F. introduced this type of conditioning. Skinner focused on voluntary behaviors and the consequences that follow them, influencing whether these behaviors will be strengthened or diminished.
Reinforcement in operant conditioning can be positive or negative, both of which serve to increase the likelihood of a behavior. Positive...
Reinforcement in operant conditioning can be positive or negative, both of which serve to increase the likelihood of a behavior. Positive...
1.5K
Timing and Consequences on Behavior
82
In operant conditioning, the timing of reinforcement is crucial. For animals like rats and cats, immediate reinforcement (within a few seconds) is much more effective than delayed reinforcement. For example, a food reward for a rat needs to follow within 30 seconds of pressing a bar to be effective.
Humans, however, can respond to delayed reinforcers. We often make decisions between immediate small rewards and delayed larger rewards. This ability to delay gratification is a significant...
Humans, however, can respond to delayed reinforcers. We often make decisions between immediate small rewards and delayed larger rewards. This ability to delay gratification is a significant...
82


