Population Based Training (PBT) is a recent approach that jointly optimizes neural network weights and hyperparameters which periodically copies weights of the best performers and mutates hyperparameters during training. Previous PBT implementations have been synchronized glass-box systems. We propose a general, black-box PBT framework that distributes many asynchronous "trials" (a small number of training steps with warm-starting) across a cluster, coordinated by the PBT controller. The black-box design does not make assumptions on model architectures, loss functions or training procedures. Our system supports dynamic hyperparameter schedules to optimize both differentiable and non-differentiable metrics. We apply our system to train a state-of-the-art WaveNet generative model for human voice synthesis. We show that our PBT system achieves better accuracy, less sensitivity and faster convergence compared to existing methods, given the same computational resource.
Training populations of agents has demonstrated great promise in Reinforcement Learning for stabilizing training, improving exploration and asymptotic performance, and generating a diverse set of solutions. However, population-based training is often not considered by practitioners as it is perceived to be either prohibitively slow (when implemented sequentially), or computationally expensive (if agents are trained in parallel on independent accelerators). In this work, we compare implementations and revisit previous studies to show that the judicious use of compilation and vectorization allow…
University of California, Berkeley（加州大学伯克利分校）；Technical University of Munich（慕尼黑工业大学）
Self-play reinforcement learning has achieved state-of-the-art, and often superhuman, performance in a variety of zero-sum games. Yet prior work has found that policies that are highly capable against regular opponents can fail catastrophically against adversarial policies: an opponent trained explicitly against the victim. Prior defenses using adversarial training were able to make the victim robust to a specific adversary, but the victim remained vulnerable to new ones. We conjecture this limitation was due to insufficient diversity of adversaries seen during training. We analyze a defense u…
Tencent AI Lab（腾讯人工智能实验室）；Tsinghua University（清华大学）
We study the problem of training a Reinforcement Learning (RL) agent that is collaborative with humans without using any human data. Although such agents can be obtained through self-play training, they can suffer significantly from distributional shift when paired with unencountered partners, such as humans. To mitigate this distributional shift, we propose Maximum Entropy Population-based training (MEP). In MEP, agents in the population are trained with our derived Population Entropy bonus to promote both pairwise diversity between agents and individual diversity of agents themselves, and a …
The adoption of electronic health records (EHR) has become universal during the past decade, which has afforded in-depth data-based research. By learning from the large amount of healthcare data, various data-driven models have been built to predict future events for different medical tasks, such as auto diagnosis and heart-attack prediction. Although EHR is abundant, the population that satisfies specific criteria for learning population-specific tasks is scarce, making it challenging to train data-hungry deep learning models. This study presents the Claim Pre-Training (Claim-PT) framework, a…
Isfahan University of Technology（伊斯法罕理工大学）；RISE Research Institutes of Sweden（瑞典RISE研究院）；Mälardalen University（梅拉达伦大学）
Plagiarism is one of the leading problems in academic and industrial environments, which its goal is to find the similar items in a typical document or source code. This paper proposes an architecture based on a Long Short-Term Memory (LSTM) and attention mechanism called LSTM-AM-ABC boosted by a population-based approach for parameter initialization. Gradient-based optimization algorithms such as back-propagation (BP) are widely used in the literature for learning process in LSTM, attention mechanism, and feed-forward neural network, while they suffer from some problems such as getting stuck …
Reinforcement learning (RL) offers the potential for training generally capable agents that can interact autonomously in the real world. However, one key limitation is the brittleness of RL algorithms to core hyperparameters and network architecture choice. Furthermore, non-stationarities such as evolving training data and increased agent complexity mean that different hyperparameters and architectures may be optimal at different points of training. This motivates AutoRL, a class of methods seeking to automate these design choices. One prominent class of AutoRL methods is Population-Based Trai…
Strategic diversity is often essential in games: in multi-player games, for example, evaluating a player against a diverse set of strategies will yield a more accurate estimate of its performance. Furthermore, in games with non-transitivities diversity allows a player to cover several winning strategies. However, despite the significance of strategic diversity, training agents that exhibit diverse behaviour remains a challenge. In this paper we study how to construct diverse populations of agents by carefully structuring how individuals within a population interact. Our approach is based on in…
Generalized Contrastive Divergence: Joint Training of Energy-Based Model and Diffusion Model through Inverse Reinforcement Learning
Korea Institute for Advanced Study（韩国高等研究院）；University of Seoul（首尔市立大学）；Seoul National University（首尔国立大学）；Hanyang University（汉阳大学）；Saige Research（赛奇研究院）
Generalized zero-shot learning (GZSL) is one of the most realistic but challenging problems due to the partiality of the classifier to supervised classes, especially under the class-inductive instance-inductive (CIII) training setting, where testing data are not available. Instance-borrowing methods and synthesizing methods solve it to some extent with the help of testing semantics, but therefore neither can be used under CIII. Besides, the latter require the training process of a classifier after generating examples. In contrast, a novel non-transductive regularization under CIII called Seman…
Shanghai Jiao Tong University（上海交通大学）；University College London（伦敦大学学院）
Population-based multi-agent reinforcement learning (PB-MARL) refers to the series of methods nested with reinforcement learning (RL) algorithms, which produces a self-generated sequence of tasks arising from the coupled population dynamics. By leveraging auto-curricula to induce a population of distinct emergent strategies, PB-MARL has achieved impressive success in tackling multi-agent tasks. Despite remarkable prior arts of distributed RL frameworks, PB-MARL poses new challenges for parallelizing the training frameworks due to the additional complexity of multiple nested workloads between s…
Generalization is a major challenge for multi-agent reinforcement learning. How well does an agent perform when placed in novel environments and in interactions with new co-players? In this paper, we investigate and quantify the relationship between generalization and diversity in the multi-agent domain. Across the range of multi-agent environments considered here, procedurally generating training levels significantly improves agent performance on held-out levels. However, agent performance on the specific levels used in training sometimes declines as a result. To better understand the effects…
Multi-Agent Training for Pommerman: Curriculum Learning and Population-based Self-Play Approach
Agoda (Booking Holdings Inc.)（Agoda（Booking Holdings Inc.））；Ming Chi University of Technology（明志科技大学）；Can Tho University（芹苴大学）；National Yang Ming Chiao Tung University（国立阳明交通大学）
针对 Pommerman 中延迟效果、稀疏奖励和对手匹配难题，提出融合课程学习、种群自博弈、自适应稠密奖励退火与 Elo 匹配的训练系统，并训练出无需通信即可超越顶尖学习型智能体的队伍。
