Search

Reinforcement learning (RL) techniques have been increasingly investigated for dynamic HVAC control in buildings. However, most studies focus on exploring solutions in online or off-policy scenarios without discussing in detail the implementation feasibility or effectiveness of dealing with purely offline datasets or trajectories. The lack of these works limits the real-world deployment of RL-based HVAC controllers, especially considering the abundance of historical data. To this end, this paper comprehensively evaluates the strengths and limitations of state-of-the-art offline RL algorithms by conducting analytical and numerical studies. The analysis is conducted from two perspectives: algorithms and dataset characteristics. As a prerequisite, the necessity of applying offline RL algorithms is first confirmed in two building environments. The ability of observation history modeling to reduce violations and enhance performance is subsequently studied. Next, the performance of RL-based controllers under datasets with different qualitative and quantitative conditions is investigated, including constraint satisfaction and power consumption. Finally, the sensitivity of certain hyperparameters is also evaluated. The results indicate that datasets of a certain suboptimality level and relatively small scale can be utilized to effectively train a well-performed RL-based HVAC controller. Specifically, such controllers can reduce at most 28.5% violation ratios of indoor temperatures and achieve at most 12.1% power savings compared to the baseline controller. In summary, this paper presents our well-structured investigations and new findings when applying offline reinforcement learning to building HVAC systems.

Rehearsal, retraining on a stored small data subset of old tasks, has been proven effective in solving catastrophic forgetting in continual learning. However, due to the sampled data may have a large bias towards the original dataset, retraining them is susceptible to driving continual domain drift of old tasks in feature space, resulting in forgetting. In this paper, we focus on tackling the continual domain drift problem with centroid distance distillation. First, we propose a centroid caching mechanism for sampling data points based on constructed centroids to reduce the sample bias in rehearsal. Then, we present a centroid distance distillation that only stores the centroid distance to reduce the continual domain drift. The experiments on four continual learning datasets show the superiority of the proposed method, and the continual domain drift can be reduced.

Existing class-incremental lifelong learning studies only the data is with single-label, which limits its adaptation to multi-label data. This paper studies Lifelong Multi-Label (LML) classification, which builds an online class-incremental classifier in a sequential multi-label classification data stream. Training on the data with Partial Labels in LML classification may result in more serious Catastrophic Forgetting in old classes. To solve the problem, the study proposes an Augmented Graph Convolutional Network (AGCN) with a built Augmented Correlation Matrix (ACM) across sequential partial-label tasks. The results of two benchmarks show that the method is effective for LML classification and reducing forgetting., Comment: arXiv admin note: substantial text overlap with arXiv:2203.05534

Sentence-based Image Editing (SIE) aims to deploy natural language to edit an image. Offering potentials to reduce expensive manual editing, SIE has attracted much interest recently. However, existing methods can hardly produce accurate editing and even lead to failures in attribute editing when the query sentence is with multiple editable attributes. To cope with this problem, by focusing on enhancing the difference between attributes, this paper proposes a novel model called Contrastive Attention Generative Adversarial Network (CA-GAN), which is inspired from contrastive training. Specifically, we first design a novel contrastive attention module to enlarge the editing difference between random combinations of attributes which are formed during training. We then construct an attribute discriminator to ensure effective editing on each attribute. A series of experiments show that our method can generate very encouraging results in sentence-based image editing with multiple attributes on CUB and COCO dataset. Our code is available at https://github.com/Zlq2021/CA-GAN, Comment: Accepted by BMVC 2021