In standard Reinforcement Learning settings, agents typically assume immediate feedback about the effects of their actions after taking them. However, in practice, this assumption may not hold true due to physical constraints and can significantly impact the performance of RL algorithms. In this paper, we focus on addressing observation delays in partially observable environments. We propose leveraging world models, which have shown success in integrating past observations and learning dynamics, to handle observation delays. By reducing delayed POMDPs to delayed MDPs with world models, our methods can effectively handle partial observability, where existing approaches achieve sub-optimal performance or even degrade quickly as observability decreases. Experiments suggest that one of our methods can outperform a naive model-based approach by up to %30. Moreover, we evaluate our methods on visual input based delayed environment, for the first time showcasing delay-aware reinforcement learning on visual observations.
Large language models (LLMs) are being applied as actors for sequential decision making tasks in domains such as robotics and games, utilizing their general world knowledge and planning abilities. However, previous work does little to explore what environment state information is provided to LLM actors via language. Exhaustively describing high-dimensional states can impair performance and raise inference costs for LLM actors. Previous LLM actors avoid the issue by relying on hand-engineered, task-specific protocols to determine which features to communicate about a state and which to leave out. In this work, we propose Brief Language INputs for DEcision-making Responses (BLINDER), a method for automatically selecting concise state descriptions by learning a value function for task-conditioned state descriptions. We evaluate BLINDER on the challenging video game NetHack and a robotic manipulation task. Our method improves task success rate, reduces input size and compute costs, and generalizes between LLM actors.
Question Answering (QA) is a task that entails reasoning over natural language contexts, and many relevant works augment language models (LMs) with graph neural networks (GNNs) to encode the Knowledge Graph (KG) information. However, most existing GNN-based modules for QA do not take advantage of rich relational information of KGs and depend on limited information interaction between the LM and the KG. To address these issues, we propose Question Answering Transformer (QAT), which is designed to jointly reason over language and graphs with respect to entity relations in a unified manner. Specifically, QAT constructs Meta-Path tokens, which learn relation-centric embeddings based on diverse structural and semantic relations. Then, our Relation-Aware Self-Attention module comprehensively integrates different modalities via the Cross-Modal Relative Position Bias, which guides information exchange between relevant entities of different modalities. We validate the effectiveness of QAT on commonsense question answering datasets like CommonsenseQA and OpenBookQA, and on a medical question answering dataset, MedQA-USMLE. On all the datasets, our method achieves state-of-the-art performance. Our code is available at http://github.com/mlvlab/QAT.
Existing message passing neural networks for heterogeneous graphs rely on the concepts of meta-paths or meta-graphs due to the intrinsic nature of heterogeneous graphs. However, the meta-paths and meta-graphs need to be pre-configured before learning and are highly dependent on expert knowledge to construct them. To tackle this challenge, we propose a novel concept of meta-node for message passing that can learn enriched relational knowledge from complex heterogeneous graphs without any meta-paths and meta-graphs by explicitly modeling the relations among the same type of nodes. Unlike meta-paths and meta-graphs, meta-nodes do not require any pre-processing steps that require expert knowledge. Going one step further, we propose a meta-node message passing scheme and apply our method to a contrastive learning model. In the experiments on node clustering and classification tasks, the proposed meta-node message passing method outperforms state-of-the-arts that depend on meta-paths. Our results demonstrate that effective heterogeneous graph learning is possible without the need for meta-paths that are frequently used in this field.
Generic Event Boundary Detection (GEBD) is a newly introduced task that aims to detect "general" event boundaries that correspond to natural human perception. In this paper, we introduce a novel contrastive learning based approach to deal with the GEBD. Our intuition is that the feature similarity of the video snippet would significantly vary near the event boundaries, while remaining relatively the same in the remaining part of the video. In our model, Temporal Self-similarity Matrix (TSM) is utilized as an intermediate representation which takes on a role as an information bottleneck. With our model, we achieved significant performance boost compared to the given baselines. Our code is available at https://github.com/hello-jinwoo/LOVEU-CVPR2021.
Multi-armed bandit methods have been used for dynamic experiments particularly in online services. Among the methods, thompson sampling is widely used because it is simple but shows desirable performance. Many thompson sampling methods for binary rewards use logistic model that is written in a specific parameterization. In this study, we reparameterize logistic model with odds ratio parameters. This shows that thompson sampling can be used with subset of parameters. Based on this finding, we propose a novel method, "Odds-ratio thompson sampling", which is expected to work robust to time-varying effect. Use of the proposed method in continuous experiment is described with discussing a desirable property of the method. In simulation studies, the novel method works robust to temporal background effect, while the loss of performance was only marginal in case with no such effect. Finally, using dataset from real service, we showed that the novel method would gain greater rewards in practical environment.