Policy Transfer using Reward Shaping

Domain- and task-specific transfer learning for medical segmentation tasks

What matters in a transferable neural network model for relation classification in the biomedical domain?

Transfer learning in reinforcement learning has been an active area of research over the past decade. In transfer learning, training on a source task is leveraged to speed up or otherwise improve learning on a target task. This paper presents the more ambitious problem of curriculum learning in reinforcement learning, in which the goal is to design a sequence of source tasks for an agent to train on, such that final performance or learning speed is improved. We take the position that each stage of such a curriculum should be tailored to the current ability of the agent in order to promote learning new behaviors. Thus, as a first step towards creating a curriculum, the trainer must be able to create novel, agent-specific source tasks. We explore how such a space of useful tasks can be created using a parameterized model of the domain and observed trajectories on the target task. We experimentally show that these methods can be used to form components of a curriculum and that such a curriculum can be used successfully for transfer learning in 2 challenging multiagent reinforcement learning domains.

Transfer on Reinforcement Learning (RL) is a promising method on learning new skills and adapting new situations for humanoid robots as tasks or environments change. Within the normal process of Transfer Learning in RL, the inter-task mapping is manually defined, which lacks generalization ability. Therefore, how to automatically learning intertask relations becomes a hot topic. Considering the limited computational resource of a physical humanoid robot, high learning efficiency regarding to both fast speed algorithm and low sample complexity should be emphasized in skills transfer. According to this view, in this research, the inter-task relations are modelled using a three-way Restricted Boltzmann Machine (RBM), which is turned out to be a powerful model in capturing the similarity between samples from source task and target task. Since standard Contrastive Divergence (CD) algorithm commonly used for RBM learning suffers from the inputindependent problem and may lead the learning process timeconsuming or inapplicable, a Cyclic Contrastive Divergence (CCD) learning algorithm is employed. In order to evaluate the performance, experiment that transfer the skill of walking on flat surface to the skill walking on slope surface is conducted on our physical robot platform, PKU-HR5.1, and the result indicates that the method is feasible and efficient.

A key objective of transfer learning is to improve and speed-up learning on a target task after training on a different, but related, source task. This study presents a neuro-evolution method that transfers evolved policies within multi-agent tasks of varying degrees of complexity. The method incorporates behavioral diversity (novelty) search as a means to boost the task performance of transferred policies (multi-agent behaviors). Results indicate that transferred evolved multi-agent behaviors are significantly improved in more complex tasks when adapted using behavioral diversity. Comparatively, behaviors that do not use behavioral diversity to further adapt transferred behaviors, perform relatively poorly in terms of adaptation times and quality of solutions in target tasks. Also, in support of previous work, both policy transfer methods (with and without behavioral diversity adaptation), out-perform behaviors evolved in target tasks without transfer learning.

Transfer learning enables machine learning system to use previously acquired knowledge from source task on target task. Source task and target task should be correlated. Successful transfer learning would improve system performance on target task especially when target data is small. In this paper, the source task is Indonesian part-of-speech (POS) tagging system and the target task is Indonesian named entity recognition (NER) tagging system. A bidirectional gated recurrent unit and conditional random field (GRU-CRF) sequence labeller is crafted as the model for the feature representation transfer to be applied. We compared F1 evaluation results between system that used transfer and system that did not use transfer and saw an improvement around 0.02 on the transferred model when 100% training data was used and around 0.15 when 1% training data was used.

Transfer learning is a method for improving generalization performance by training a model for a different task first and then additionally training the pre-learned weights for the target task. However, transferability—the ease with which source task can be effectively transferred to which target task—is often unknown. Existing works proposed methods of measuring the transferability between classification tasks using images and discrete labels, but it cannot be applied to regression tasks. In this work, we investigate transferability among classification and regression tasks, and propose a method for predicting transferability by extending the optimal transport theory. Our transferability prediction model also can be applied to subjective tasks (e.g., aesthetics and memorability), which are usually regression tasks. We show that the appropriate source (pre-training) tasks can be predicted for the chosen target task without conducting actual pre-training and transferring trials. Experimental results demonstrated high prediction accuracy (correlation coefficient of ρ=0.791\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\rho =0.791$$\\end{document}) and a speed improvement of approximately 300 times compared with the above-mentioned greedy approach.

The SEAPoT algorithm [9] is a knowledge transfer mechanism in model-based reinforcement learning. By constructing subspaces around the changed regions, and selectively and efficiently exploring the target task, the transfer is most effective when the source and target tasks share similar objectives but differ in the transition dynamics. In this work, we identify the similarity between tasks using a new light-weight metric, based on the Jensen-Shannon distance, and show how the degree of similarity affects the transfer efficacy. We also empirically show that SEAPoT performs better in terms of jump starts and average rewards, as compared to the state-of-the-art policy reuse methods.

The recent development of connected and automated vehicle (CAV) technologies has spurred investigations to optimize dense urban traffic, maximizing vehicle speed and throughput. This paper explores advisory autonomy, in which real-time driving advisories are issued to human drivers, thus achieving near-term performance of automated vehicles. Due to the complexity of traffic systems, recent studies of coordinating CAVs have leveraged deep reinforcement learning (RL). Coarse-grained advisory is formalized as zero-order holds, and we consider a range of hold durations from 0.1 to 40 seconds. However, despite the similarity of the higher-frequency tasks for CAVs, a direct application of deep RL fails to generalize to advisory autonomy tasks. To overcome this, we employ zero-shot transfer, training policies on a set of source tasks-specific traffic scenarios with designated hold durations-and then evaluating the efficacy of these policies on different target tasks. We introduce Temporal Transfer Learning (TTL) algorithms to select source tasks for zero-shot transfer, systematically leveraging the temporal structure to solve the full range of tasks. TTL selects the most suitable source tasks to maximize the performance of the range of tasks. We validate our algorithms on diverse mixed-traffic scenarios, demonstrating that TTL more reliably solves the tasks than baselines. This paper underscores the potential of coarse-grained advisory autonomy with TTL in traffic flow optimization.

Recent advancements in reinforcement learning confirm that reinforcement learning techniques can solve large scale problems leading to high quality autonomous decision making. It is a matter of time until we will see large scale applications of reinforcement learning in various sectors, such as healthcare and cyber-security, among others. However, reinforcement learning can be time-consuming because the learning algorithms have to determine the long term consequences of their actions using delayed feedback or rewards. Reward shaping is a method of incorporating domain knowledge into reinforcement learning so that the algorithms are guided faster towards more promising solutions. Under an overarching theme of episodic reinforcement learning, this paper shows a unifying analysis of potential-based reward shaping which leads to new theoretical insights into reward shaping in both model-free and model-based algorithms, as well as in multi-agent reinforcement learning.

Interpretable policies for reinforcement learning by genetic programming

Transfer learning refers to the transfer of knowledge or information from a relevant source task to a target task. However, most existing works assume both tasks are sampled from a stationary task distribution, thereby leading to the sub-optimal performance for dynamic tasks drawn from a non-stationary task distribution in real scenarios. To bridge this gap, in this paper, we study a more realistic and challenging transfer learning setting with dynamic tasks, i.e., source and target tasks are continuously evolving over time. We theoretically show that the expected error on the dynamic target task can be tightly bounded in terms of source knowledge and consecutive distribution discrepancy across tasks. This result motivates us to propose a generic meta-learning framework L2E for modeling the knowledge transferability on dynamic tasks. It is centered around a task-guided meta-learning problem with a group of meta-pairs of tasks, based on which we are able to learn the prior model initialization for fast adaptation on the newest target task. L2E enjoys the following properties: (1) effective knowledge transferability across dynamic tasks; (2) fast adaptation to the new target task; (3) mitigation of catastrophic forgetting on historical target tasks; and (4) flexibility in incorporating any existing static transfer learning algorithms. Extensive experiments on various image data sets demonstrate the effectiveness of the proposed L2E framework.

We consider the problem of building a model to predict protein-protein interactions (PPIs) between the bacterial species Salmonella Typhimurium and the plant host Arabidopsis thaliana which is a host-pathogen pair for which no known PPIs are available. To achieve this, we present approaches, which use homology and statistical learning methods called “transfer learning.” In the transfer learning setting, the task of predicting PPIs between Arabidopsis and its pathogen S. Typhimurium is called the “target task.” The presented approaches utilize labeled data i.e., known PPIs of other host-pathogen pairs (we call these PPIs the “source tasks”). The homology based approaches use heuristics based on biological intuition to predict PPIs. The transfer learning methods use the similarity of the PPIs from the source tasks to the target task to build a model. For a quantitative evaluation we consider Salmonella-mouse PPI prediction and some other host-pathogen tasks where known PPIs exist. We use metrics such as precision and recall and our results show that our methods perform well on the target task in various transfer settings. We present a brief qualitative analysis of the Arabidopsis-Salmonella predicted interactions. We filter the predictions from all approaches using Gene Ontology term enrichment and only those interactions involving Salmonella effectors. Thereby we observe that Arabidopsis proteins involved e.g., in transcriptional regulation, hormone mediated signaling and defense response may be affected by Salmonella.

The success of transfer to improve learning on a target task is highly dependent on the selected source data. Instance-based transfer methods reuse data from the source tasks to augment the training data for the target task. If poorly chosen, this source data may inhibit learning, resulting in negative transfer. The current best performing algorithm for instance-based transfer, TrAdaBoost, performs poorly when given irrelevant source data. We present a novel set-based boosting technique for instance-based transfer. The proposed algorithm, TransferBoost, boosts both individual instances and collective sets of instances from each source task. In effect, TransferBoost boosts each source task, assigning higher weight to those source tasks which show positive transferability to the target task, and then adjusts the weights of the instances within each source task via AdaBoost. The results demonstrate that TransferBoost significantly improves transfer performance over existing instance-based algorithms when given a mix of relevant and irrelevant source data.

Parallel Multi-Environment Shaping Algorithm for Complex Multi-step Task