Monte-Carlo planning and Reinforcement Learning (RL) are essential to sequential decision making. The recent AlphaGo and AlphaZero algorithms have shown how to successfully combine these two paradigms in order to solve large scale sequential decision problems. These methodologies exploit a variant of the well-known UCT algorithm to trade off exploitation of good actions and exploration of unvisited states, but their empirical success comes at the cost of poor sample-efficiency and high computation time. In this paper, we overcome these limitations by considering convex regularization in Monte-Carlo Tree Search (MCTS), which has been successfully used in RL to efficiently drive exploration. First, we introduce a unifying theory on the use of generic convex regularizers in MCTS, deriving the regret analysis and providing guarantees of exponential convergence rate. Second, we exploit our theoretical framework to introduce novel regularized backup operators for MCTS, based on the relative entropy of the policy update, and on the Tsallis entropy of the policy. Finally, we empirically evaluate the proposed operators in AlphaGo and AlphaZero on problems of increasing dimensionality and branching factor, from a toy problem to several Atari games, showing their superiority w.r.t. representative baselines.

Intelligent Tutoring Systems (ITSs) have been developed to provide students with personalized learning experiences by adaptively generating learning paths optimized for each individual. Within the vast scope of ITS, score prediction stands out as an area of study that enables students to construct individually realistic goals based on their current position. Via the expected score provided by the ITS, a student can instantaneously compare one's expected score to one's actual score, which directly corresponds to the reliability that the ITS can instill. In other words, refining the precision of predicted scores strictly correlates to the level of confidence that a student may have with an ITS, which will evidently ensue improved student engagement. However, previous studies have solely concentrated on improving the performance of a prediction model, largely lacking focus on the benefits generated by its practical application. In this paper, we demonstrate that the accuracy of the score prediction model deployed in a real-world setting significantly impacts user engagement by providing empirical evidence. To that end, we apply a state-of-the-art deep attentive neural network-based score prediction model to Santa, a multi-platform English ITS with approximately 780K users in South Korea that exclusively focuses on the TOEIC (Test of English for International Communications) standardized examinations. We run a controlled A/B test on the ITS with two models, respectively based on collaborative filtering and deep attentive neural networks, to verify whether the more accurate model engenders any student engagement. The results conclude that the attentive model not only induces high student morale (e.g. higher diagnostic test completion ratio, number of questions answered, etc.) but also encourages active engagement (e.g. higher purchase rate, improved total profit, etc.) on Santa.

Answering questions that involve multi-step reasoning requires decomposing them and using the answers of intermediate steps to reach the final answer. However, state-of-the-art models in grounded question answering often do not explicitly perform decomposition, leading to difficulties in generalization to out-of-distribution examples. In this work, we propose a model that computes a representation and denotation for all question spans in a bottom-up, compositional manner using a CKY-style parser. Our model effectively induces latent trees, driven by end-to-end (the answer) supervision only. We show that this inductive bias towards tree structures dramatically improves systematic generalization to out-of-distribution examples compared to strong baselines on an arithmetic expressions benchmark as well as on CLOSURE, a dataset that focuses on systematic generalization of models for grounded question answering. On this challenging dataset, our model reaches an accuracy of 92.8%, significantly higher than prior models that almost perfectly solve the task on a random, in-distribution split.

Research in developmental psychology consistently shows that children explore the world thoroughly and efficiently and that this exploration allows them to learn. In turn, this early learning supports more robust generalization and intelligent behavior later in life. While much work has gone into developing methods for exploration in machine learning, artificial agents have not yet reached the high standard set by their human counterparts. In this work we propose using DeepMind Lab (Beattie et al., 2016) as a platform to directly compare child and agent behaviors and to develop new exploration techniques. We outline two ongoing experiments to demonstrate the effectiveness of a direct comparison, and outline a number of open research questions that we believe can be tested using this methodology.

In the vision-and-language navigation (VLN) task, an agent follows natural language instructions and navigate in visual environments. Compared to the indoor navigation task that has been broadly studied, navigation in real-life outdoor environments remains a significant challenge with its complicated visual inputs and an insufficient amount of instructions that illustrate the intricate urban scenes. In this paper, we introduce a Multimodal Text Style Transfer (MTST) learning approach to mitigate the problem of data scarcity in outdoor navigation tasks by effectively leveraging external multimodal resources. We first enrich the navigation data by transferring the style of the instructions generated by Google Maps API, then pre-train the navigator with the augmented external outdoor navigation dataset. Experimental results show that our MTST learning approach is model-agnostic, and our MTST approach significantly outperforms the baseline models on the outdoor VLN task, improving task completion rate by 22\% relatively on the test set and achieving new state-of-the-art performance.

Influence diagrams (IDs) are well-known formalisms extending Bayesian networks to model decision situations under uncertainty. Although they are convenient as a decision theoretic tool, their knowledge representation ability is limited in capturing other crucial notions such as logical consistency. We complement IDs with the light-weight description logic (DL) EL to overcome such limitations. We consider a setup where DL axioms hold in some contexts, yet the actual context is uncertain. The framework benefits from the convenience of using DL as a domain knowledge representation language and the modelling strength of IDs to deal with decisions over contexts in the presence of contextual uncertainty. We define related reasoning problems and study their computational complexity.

We propose a Deep Reinforcement Learning (Deep RL) algorithm for solving the online 3D bin packing problem for an arbitrary number of bins and any bin size. The focus is on producing decisions that can be physically implemented by a robotic loading arm, a laboratory prototype used for testing the concept. The problem considered in this paper is novel in two ways. First, unlike the traditional 3D bin packing problem, we assume that the entire set of objects to be packed is not known a priori. Instead, a fixed number of upcoming objects is visible to the loading system, and they must be loaded in the order of arrival. Second, the goal is not to move objects from one point to another via a feasible path, but to find a location and orientation for each object that maximises the overall packing efficiency of the bin(s). Finally, the learnt model is designed to work with problem instances of arbitrary size without retraining. Simulation results show that the RL-based method outperforms state-of-the-art online bin packing heuristics in terms of empirical competitive ratio and volume efficiency.

Autonomous driving has achieved significant progress in recent years, but autonomous cars are still unable to tackle high-risk situations where a potential accident is likely. In such near-accident scenarios, even a minor change in the vehicle's actions may result in drastically different consequences. To avoid unsafe actions in near-accident scenarios, we need to fully explore the environment. However, reinforcement learning (RL) and imitation learning (IL), two widely-used policy learning methods, cannot model rapid phase transitions and are not scalable to fully cover all the states. To address driving in near-accident scenarios, we propose a hierarchical reinforcement and imitation learning (H-ReIL) approach that consists of low-level policies learned by IL for discrete driving modes, and a high-level policy learned by RL that switches between different driving modes. Our approach exploits the advantages of both IL and RL by integrating them into a unified learning framework. Experimental results and user studies suggest our approach can achieve higher efficiency and safety compared to other methods. Analyses of the policies demonstrate our high-level policy appropriately switches between different low-level policies in near-accident driving situations.

Language representations are known to carry stereotypical biases and, as a result, lead to biased predictions in downstream tasks. While existing methods are effective at mitigating biases by linear projection, such methods are too aggressive: they not only remove bias, but also erase valuable information from word embeddings. We develop new measures for evaluating specific information retention that demonstrate the tradeoff between bias removal and information retention. To address this challenge, we propose OSCaR (Orthogonal Subspace Correction and Rectification), a bias-mitigating method that focuses on disentangling biased associations between concepts instead of removing concepts wholesale. Our experiments on gender biases show that OSCaR is a well-balanced approach that ensures that semantic information is retained in the embeddings and bias is also effectively mitigated.

Many systems have been developed in recent years to mine logical rules from large-scale Knowledge Graphs (KGs), on the grounds that representing regularities as rules enables both the interpretable inference of new facts, and the explanation of known facts. Among these systems, the walk-based methods that generate the instantiated rules containing constants by abstracting sampled paths in KGs demonstrate strong predictive performance and expressivity. However, due to the large volume of possible rules, these systems do not scale well where computational resources are often wasted on generating and evaluating unpromising rules. In this work, we address such scalability issues by proposing new methods for pruning unpromising rules using rule hierarchies. The approach consists of two phases. Firstly, since rule hierarchies are not readily available in walk-based methods, we have built a Rule Hierarchy Framework (RHF), which leverages a collection of subsumption frameworks to build a proper rule hierarchy from a set of learned rules. And secondly, we adapt RHF to an existing rule learner where we design and implement two methods for Hierarchical Pruning (HPMs), which utilize the generated hierarchies to remove irrelevant and redundant rules. Through experiments over four public benchmark datasets, we show that the application of HPMs is effective in removing unpromising rules, which leads to significant reductions in the runtime as well as in the number of learned rules, without compromising the predictive performance.