Since model bias and associated initialization shock are serious shortcomings that reduce prediction skills in state-of-the-art decadal climate prediction efforts, we pursue a complementary machine-learning-based approach to climate prediction. The example problem setting we consider consists of predicting natural variability of the North Atlantic sea surface temperature on the interannual timescale in the pre-industrial control simulation of the Community Earth System Model (CESM2). While previous works have considered the use of recurrent networks such as convolutional LSTMs and reservoir computing networks in this and other similar problem settings, we currently focus on the use of feedforward convolutional networks. In particular, we find that a feedforward convolutional network with a Densenet architecture is able to outperform a convolutional LSTM in terms of predictive skill. Next, we go on to consider a probabilistic formulation of the same network based on Stein variational gradient descent and find that in addition to providing useful measures of predictive uncertainty, the probabilistic (Bayesian) version improves on its deterministic counterpart in terms of predictive skill. Finally, we characterize the reliability of the ensemble of ML models obtained in the probabilistic setting by using analysis tools developed in the context of ensemble numerical weather prediction.
While natural gradients have been widely studied from both theoretical and empirical perspectives, we argue that a fundamental theoretical issue regarding the existence of gradients in infinite dimensional function spaces remains underexplored. We therefore study the natural gradient induced by Sobolev metrics and develop several rigorous results. Our results also establish new connections between natural gradients and RKHS theory, and specifically to the Neural Tangent Kernel (NTK). We develop computational techniques for the efficient approximation of the proposed Sobolev Natural Gradient. Preliminary experimental results reveal the potential of this new natural gradient variant.
Standard model-free reinforcement learning algorithms optimize a policy that generates the action to be taken in the current time step in order to maximize expected future return. While flexible, it faces difficulties arising from the inefficient exploration due to its single step nature. In this work, we present Generative Planning method (GPM), which can generate actions not only for the current step, but also for a number of future steps (thus termed as generative planning). This brings several benefits to GPM. Firstly, since GPM is trained by maximizing value, the plans generated from it can be regarded as intentional action sequences for reaching high value regions. GPM can therefore leverage its generated multi-step plans for temporally coordinated exploration towards high value regions, which is potentially more effective than a sequence of actions generated by perturbing each action at single step level, whose consistent movement decays exponentially with the number of exploration steps. Secondly, starting from a crude initial plan generator, GPM can refine it to be adaptive to the task, which, in return, benefits future explorations. This is potentially more effective than commonly used action-repeat strategy, which is non-adaptive in its form of plans. Additionally, since the multi-step plan can be interpreted as the intent of the agent from now to a span of time period into the future, it offers a more informative and intuitive signal for interpretation. Experiments are conducted on several benchmark environments and the results demonstrated its effectiveness compared with several baseline methods.
Maximum entropy (MaxEnt) RL maximizes a combination of the original task reward and an entropy reward. It is believed that the regularization imposed by entropy, on both policy improvement and policy evaluation, together contributes to good exploration, training convergence, and robustness of learned policies. This paper takes a closer look at entropy as an intrinsic reward, by conducting various ablation studies on soft actor-critic (SAC), a popular representative of MaxEnt RL. Our findings reveal that in general, entropy rewards should be applied with caution to policy evaluation. On one hand, the entropy reward, like any other intrinsic reward, could obscure the main task reward if it is not properly managed. We identify some failure cases of the entropy reward especially in episodic Markov decision processes (MDPs), where it could cause the policy to be overly optimistic or pessimistic. On the other hand, our large-scale empirical study shows that using entropy regularization alone in policy improvement, leads to comparable or even better performance and robustness than using it in both policy improvement and policy evaluation. Based on these observations, we recommend either normalizing the entropy reward to a zero mean (SACZero), or simply removing it from policy evaluation (SACLite) for better practical results.
We consider the safe reinforcement learning (RL) problem of maximizing utility while satisfying provided constraints. Since we do not assume any prior knowledge or pre-training of the safety concept, we are interested in asymptotic constraint satisfaction. A popular approach in this line of research is to combine the Lagrangian method with a model-free RL algorithm to adjust the weight of the constraint reward dynamically. It relies on a single policy to handle the conflict between utility and constraint rewards, which is often challenging. Inspired by the safety layer design (Dalal et al., 2018), we propose to separately learn a safety editor policy that transforms potentially unsafe actions output by a utility maximizer policy into safe ones. The safety editor is trained to maximize the constraint reward while minimizing a hinge loss of the utility Q values of actions before and after the edit. On 12 custom Safety Gym (Ray et al., 2019) tasks and 2 safe racing tasks with very harsh constraint thresholds, our approach demonstrates outstanding utility performance while complying with the constraints. Ablation studies reveal that our two-policy design is critical. Simply doubling the model capacity of typical single-policy approaches will not lead to comparable results. The Q hinge loss is also important in certain circumstances, and replacing it with the usual L2 distance could fail badly.
Starting from the design philosophy of "user-centered design", this paper analyzes the human factors characteristics of intelligent human-computer interaction (iHCI) and proposes a concept of "user-oriented iHCI". It further proposes a new human factors framework for iHCI based on the theories of joint cognitive systems, situation awareness, and intelligent agents. With the help of the new concept and framework, the paper analyzes the human factors issues in the ecosystem of autonomous vehicle co-driving and layouts future research agenda. Finally, the paper analyzes the two important research areas in iHCI (i.e., user intention recognition, human-computer collaboration) and points out the focus of human factors research in the future.
To enable robots to instruct humans in collaborations, we identify several aspects of language processing that are not commonly studied in this context. These include location, planning, and generation. We suggest evaluations for each task, offer baselines for simple methods, and close by discussing challenges and opportunities in studying language for collaboration.
An important task in NLP applications such as sentence simplification is the ability to take a long, complex sentence and split it into shorter sentences, rephrasing as necessary. We introduce a novel dataset and a new model for this `split and rephrase' task. Our BiSECT training data consists of 1 million long English sentences paired with shorter, meaning-equivalent English sentences. We obtain these by extracting 1-2 sentence alignments in bilingual parallel corpora and then using machine translation to convert both sides of the corpus into the same language. BiSECT contains higher quality training examples than previous Split and Rephrase corpora, with sentence splits that require more significant modifications. We categorize examples in our corpus, and use these categories in a novel model that allows us to target specific regions of the input sentence to be split and edited. Moreover, we show that models trained on BiSECT can perform a wider variety of split operations and improve upon previous state-of-the-art approaches in automatic and human evaluations.