People with Alzheimer's disease and related dementias (ADRD) often show the problem of repetitive questioning, which brings a great burden on persons with ADRD (PwDs) and their caregivers. Conversational robots hold promise of coping with this problem and hence alleviating the burdens on caregivers. In this paper, we proposed a partially observable markov decision process (POMDP) model for the PwD-robot interaction in the context of repetitive questioning, and used Q-learning to learn an adaptive conversation strategy (i.e., rate of follow-up question and difficulty of follow-up question) towards PwDs with different cognitive capabilities and different engagement levels. The results indicated that Q-learning was helpful for action selection for the robot. This may be a useful step towards the application of conversational social robots to cope with repetitive questioning in PwDs.

Value functions are central to Dynamic Programming and Reinforcement Learning but their exact estimation suffers from the curse of dimensionality, challenging the development of practical value-function (VF) estimation algorithms. Several approaches have been proposed to overcome this issue, from non-parametric schemes that aggregate states or actions to parametric approximations of state and action VFs via, e.g., linear estimators or deep neural networks. Relevantly, several high-dimensional state problems can be well-approximated by an intrinsic low-rank structure. Motivated by this and leveraging results from low-rank optimization, this paper proposes different stochastic algorithms to estimate a low-rank factorization of the $Q(s, a)$ matrix. This is a non-parametric alternative to VF approximation that dramatically reduces the computational and sample complexities relative to classical $Q$-learning methods that estimate $Q(s,a)$ separately for each state-action pair.

In this paper, we propose actor-critic approaches by introducing an actor policy on QMIX [9], which can remove the monotonicity constraint of QMIX and implement a non-monotonic value function factorization for joint action-value. We evaluate our actor-critic methods on StarCraft II micromanagement tasks, and show that it has a stronger performance on maps with heterogeneous agent types.

In model-based reinforcement learning (MBRL), Wan et al. (2019) showed conditions under which the environment model could produce the expectation of the next feature vector rather than the full distribution, or a sample thereof, with no loss in planning performance. Such expectation models are of interest when the environment is stochastic and non-stationary, and the model is approximate, such as when it is learned using function approximation. In these cases a full distribution model may be impractical and a sample model may be either more expensive computationally or of high variance. Wan et al. considered only planning for prediction to evaluate a fixed policy. In this paper, we treat the control case - planning to improve and find a good approximate policy. We prove that planning with an expectation model must update a state-value function, not an action-value function as previously suggested (e.g., Sorg & Singh, 2010). This opens the question of how planning influences action selections. We consider three strategies for this and present general MBRL algorithms for each. We identify the strengths and weaknesses of these algorithms in computational experiments. Our algorithms and experiments are the first to treat MBRL with expectation models in a general setting.

In this work we explore an auxiliary loss useful for reinforcement learning in environments where strong performing agents are required to be able to navigate a spatial environment. The auxiliary loss proposed is to minimize the classification error of a neural network classifier that predicts whether or not a pair of states sampled from the agents current episode trajectory are in order. The classifier takes as input a pair of states as well as the agent's memory. The motivation for this auxiliary loss is that there is a strong correlation with which of a pair of states is more recent in the agents episode trajectory and which of the two states is spatially closer to the agent. Our hypothesis is that learning features to answer this question encourages the agent to learn and internalize in memory representations of states that facilitate spatial reasoning. We tested this auxiliary loss on a navigation task in a gridworld and achieved 9.6% increase in accumulative episode reward compared to a strong baseline approach.

Action advising is a peer-to-peer knowledge exchange technique built on the teacher-student paradigm to alleviate the sample inefficiency problem in deep reinforcement learning. Recently proposed student-initiated approaches have obtained promising results. However, due to being in the early stages of development, these also have some substantial shortcomings. One of the abilities that are absent in the current methods is further utilising advice by reusing, which is especially crucial in the practical settings considering the budget and cost constraints in peer-to-peer. In this study, we present an approach to enable the student agent to imitate previously acquired advice to reuse them directly in its exploration policy, without any interventions in the learning mechanism itself. In particular, we employ a behavioural cloning module to imitate the teacher policy and use dropout regularisation to have a notion of epistemic uncertainty to keep track of which state-advice pairs are actually collected. As the results of experiments we conducted in three Atari games show, advice reusing via generalisation is indeed a feasible option in deep RL and our approach can successfully achieve this while significantly improving the learning performance, even when paired with a simple early advising heuristic.

Deep Reinforcement Learning (RL) techniques can benefit greatly from leveraging prior experience, which can be either self-generated or acquired from other entities. Action advising is a framework that provides a flexible way to transfer such knowledge in the form of actions between teacher-student peers. However, due to the realistic concerns, the number of these interactions is limited with a budget; therefore, it is crucial to perform these in the most appropriate moments. There have been several promising studies recently that address this problem setting especially from the student's perspective. Despite their success, they have some shortcomings when it comes to the practical applicability and integrity as an overall solution to the learning from advice challenge. In this paper, we extend the idea of advice reusing via teacher imitation to construct a unified approach that addresses both advice collection and advice utilisation problems. Furthermore, we also propose a method to automatically determine the relevant hyperparameters of these components on-the-fly to make it able to adapt to any task with minimal human intervention. The experiments we performed in 5 different Atari games verify that our algorithm can outperform its competitors by achieving state-of-the-art performance, and its components themselves also provides significant advantages individually.

Training with Reinforcement Learning requires a reward function that is used to guide the agent towards achieving its objective. However, designing smooth and well-behaved rewards is in general not trivial and requires significant human engineering efforts. Generating rewards in self-supervised way, by inspiring the agent with an intrinsic desire to learn and explore the environment, might induce more general behaviours. In this work, we propose a curiosity-based bonus as intrinsic reward for Reinforcement Learning, computed as the Bayesian surprise with respect to a latent state variable, learnt by reconstructing fixed random features. We extensively evaluate our model by measuring the agent's performance in terms of environment exploration, for continuous tasks, and looking at the game scores achieved, for video games. Our model is computationally cheap and empirically shows state-of-the-art performance on several problems. Furthermore, experimenting on an environment with stochastic actions, our approach emerged to be the most resilient to simple stochasticity. Further visualization is available on the project webpage.(https://lbsexploration.github.io/)

Estimating dynamic treatment regimes (DTRs) from retrospective observational data is challenging as some degree of unmeasured confounding is often expected. In this work, we develop a framework of estimating properly defined "optimal" DTRs with a time-varying instrumental variable (IV) when unmeasured covariates confound the treatment and outcome, rendering the potential outcome distributions only partially identified. We derive a novel Bellman equation under partial identification, use it to define a generic class of estimands (termed IV-optimal DTRs), and study the associated estimation problem. We then extend the IV-optimality framework to tackle the policy improvement problem, delivering IV-improved DTRs that are guaranteed to perform no worse and potentially better than a pre-specified baseline DTR. Importantly, our IV-improvement framework opens up the possibility of strictly improving upon DTRs that are optimal under the no unmeasured confounding assumption (NUCA). We demonstrate via extensive simulations the superior performance of IV-optimal and IV-improved DTRs over the DTRs that are optimal only under the NUCA. In a real data example, we embed retrospective observational registry data into a natural, two-stage experiment with noncompliance using a time-varying IV and estimate useful IV-optimal DTRs that assign mothers to high-level or low-level neonatal intensive care units based on their prognostic variables.

Using insight from numerical approximation of ODEs and the problem formulation and solution methodology of TD learning through a Galerkin relaxation, I propose a new class of TD learning algorithms. After applying the improved numerical methods, the parameter being approximated has a guaranteed order of magnitude reduction in the Taylor Series error of the solution to the ODE for the parameter $\theta(t)$ that is used in constructing the linearly parameterized value function. Predictor-Corrector Temporal Difference (PCTD) is what I call the translated discrete time Reinforcement Learning(RL) algorithm from the continuous time ODE using the theory of Stochastic Approximation(SA). Both causal and non-causal implementations of the algorithm are provided, and simulation results are listed for an infinite horizon task to compare the original TD(0) algorithm against both versions of PCTD(0).