Learning to answer multi-step complex questions requires machines to perform like a human to think and reason step by step, which is one of the core abilities of a question answering system. Recent advancements have revealed that large language models exhibit remarkable reasoning capabilities by generating intermediate chain-of-thought rationales. However, the completeness of their rationales lacks assurance as they are susceptible to omitting steps and making factual errors. In this paper, drawing inspiration from human-like reasoning processes in answering multi-step questions, we explicitly plan the rationales to ensure their completeness. We propose a two-stage Decomposition-Evaluation (Dec-Eval) framework including a step decomposition stage and a rationale generation stage. Specifically, in the first stage, we decompose the complex question into simpler sub-ones and simulate a human's ability to grasp logical clues to ensure the integrity of step planning. Then, in the second stage, based on the sub-questions, we generate and evaluate rationales step by step. Both stages work together organically, improving the completeness of rationales and the accuracy of the answer. To further control the question answering process, we propose a novel knowledge injection mechanism that incorporates external knowledge to guide both stages. Extensive experiments on three challenging multi-step QA datasets demonstrate that Dec-Eval can explicitly generate more logical rationales, and significantly improve the reasoning performances of different backbone models.

Knowledge-based question answering (KBQA) is a key task in natural language processing research, and also an approach to access the web data and knowledge, which requires exploiting knowledge graphs (KGs) for reasoning. In the literature, one promising solution for KBQA is to incorporate the pretrained language model (LM) with KGs by generating KG-centered pretraining corpus, which has shown its superiority. However, these methods often depend on specific techniques and resources to work, which may not always be available and restrict its application. Moreover, existing methods focus more on improving language understanding with KGs, while neglect the more important human-like complex reasoning. To this end, in this paper, we propose a general K nowledge-I njected C urriculum P retraining framework (KICP) to achieve comprehensive KG learning and exploitation for KBQA tasks, which is composed of knowledge injection (KI), knowledge adaptation (KA) and curriculum reasoning (CR). Specifically, the KI module first injects knowledge into the LM by generating KG-centered pretraining corpus, and generalizes the process into three key steps that could work with different implementations for flexible application. Next, the KA module learns knowledge from the generated corpus with LM equipped with an adapter as well as keeps its original natural language understanding ability to reduce the negative impacts of the difference between the generated and natural corpus. Last, to enable the LM with complex reasoning, the CR module follows human reasoning patterns to construct three corpora with increasing difficulties of reasoning, and further trains the LM from easy to hard in a curriculum manner to promote model learning. We provide an implementation of the general framework, and evaluate the proposed KICP on four real-word datasets. The results demonstrate that our framework can achieve higher performances, and have good generalization ability to other QA tasks.

Training GNNs over large graphs faces a severe data processing bottleneck, involving both sampling and feature loading. To tackle this issue, we introduce F 2 CGT, a fast GNN training system incorporating feature compression. To avoid potential accuracy degradation, we propose a two-level, hybrid feature compression approach that applies different compression methods to various graph nodes. This differentiated choice strikes a balance between rounding errors, compression ratios, model accuracy loss, and preprocessing costs. Our theoretical analysis proves that this approach offers convergence and comparable model accuracy as the conventional training without feature compression. Additionally, we also co-design the on-GPU cache sub-system with compression-enabled training within F 2 CGT. The new cache sub-system, driven by a cost model, runs new cache policies to carefully choose graph nodes with high access frequencies, and well partitions the spare GPU memory for various types of graph data, for improving cache hit rates. Finally, extensive evaluation of F 2 CGT on two popular GNN models and four datasets, including three large public datasets, demonstrates that F 2 CGT achieves a compression ratio of up to 128 and provides GNN training speedups of 1.23-2.56× and 3.58--71.46× for single-machine and distributed training, respectively, with up to 32 GPUs and marginal accuracy loss.

While highly influential in deep learning, especially in natural language processing, the Transformer model has not exhibited competitive performance in unsupervised graph representation learning (UGRL). Conventional approaches, which focus on local substructures on the graph, offer simplicity but often fall short in encapsulating comprehensive structural information of the graph. This deficiency leads to suboptimal generalization performance. To address this, we proposed the GraKerformer model, a variant of the standard Transformer architecture, to mitigate the shortfall in structural information representation and enhance the performance in UGRL. By leveraging the shortest-path graph kernel (SPGK) to weight attention scores and combining graph neural networks, the GraKerformer effectively encodes the nuanced structural information of graphs. We conducted evaluations on the benchmark datasets for graph classification to validate the superior performance of our approach.

Erasure coding is increasingly deployed in modern clustered storage systems to provide low-cost reliable storage. In particular, Locally Repairable Codes (LRCs) are a popular family of repair-efficient erasure codes that receive wide deployment in practice. In this paper, we analyze the storage process formulated as a data partitioning phase plus a node selection phase for LRCs in clustered storage systems. We show that the conventional flat partitioning and random partitioning incur significant cross-cluster repair traffic, while the random node selection causes storage and network imbalance. To this end, we design a new storage scheme composed of an optimal partitioning strategy and an enhanced node selection strategy for LRCs. Our partitioning strategy minimizes the cross-cluster repair traffic by dividing each group of blocks into the minimum number of clusters and further compactly placing the blocks. Our node selection strategy improves load balance by choosing less-loaded clusters and nodes to store blocks with potential higher access frequency at higher priority. We implement our storage scheme on a key-value store prototype atop Memcached. Evaluation on a LAN testbed shows that our scheme greatly improves the repair performance and load balance ratio compared to the baseline.

Both point cloud-based and view-based deep learning methods for 3D shape recognition have achieved relatively remarkable results in recent years. However, there are few methods to jointly represent 3D shapes from both point cloud and multi-view modal data. Therefore, we propose a guidance enhancement network (GENet) for 3D shape recognition based on multimodal data. On the one hand, the point cloud is encoded with features from both explicit and implicit aspects, and on the other hand, all views are encoded and constructed as a graph. In the multilayer guidance enhancement module, graph convolutional neural network (GCN) enhances each view feature, and then temporary high-level features (initially point cloud global feature) guide multiple low-level view features to obtain correlation coefficients, through which the views with higher importance are filtered as inputs for the next layer of the structure and the view features in the current layer are weighted and aggregated. The aggregated view features are then connected to the high-level features with residuals to form the enhanced high-level features. The 3D shape descriptor is finally obtained after several guidance and enhancements. The proposed GENet achieves state-of-the-art results on the 3D benchmark dataset ModelNet.

Both point cloud-based and multi-view-based methods have achieved remarkable results in 3D shape recognition, yet there are few methods that combine the two types of data. In this paper, a novel Global-Local Complementary Network (GLCNet) based on multimodal data is proposed. The network obtains more powerful shape descriptors by stacking multiple layers of Global-Local Complementary Module (GLC Module). More specifically, the Global-Local Relation Score Module is first used to obtain the relationship between view features and global feature. The relationship is then utilized to facilitate the aggregation of view features and to filter out the more important ones. Finally, the aggregated view features are fused with the global features to form a stronger global feature. GLCNet enables the characteristics of various data to be fully utilized and achieves a true sense of complementarity of strengths and weaknesses. Extensive experiments on the benchmark dataset ModelNet show that GLCNet achieves state-of-the-art results in 3D shape classification and retrieval.

This paper mainly studies graph representation learning in unsupervised scenarios combined with Transformer models. Transformer network models have been widely used in many fields of machine learning and deep learning, and the application of transformer architectures to graph data has been very popular recently. For graph data, the field of graph representation learning has recently attracted a lot of attention. Graph-level representation is widely used in the real world, such as drug molecule design and disease classification in biochemistry. Traditional graph kernel methods, which design different graph kernels for different substructures, are simple but have poor generalization performance. Recently methods based on language models, such as graph2vec, use a particular substructure as the graph representation, which is also similar to the hand-crafted approach and also leads to poor generalization ability. In this paper, we propose the UGTransformer model, which builds on the standard Transformer architecture. We introduce several simple and effective structural encoding methods in order to encode the structural information of the graph into the model efficiently. The unsupervised representation of graphs is learned through a multi-headed attention mechanism and by using powerful aggregation functions. We conducted experiments on a benchmark date set for graph classification, and the experimental results validate the effectiveness of our proposed model.

Vector retrieval focuses on finding the k-nearest neighbors from a bunch of data points, and is widely used in a diverse set of areas such as information retrieval and recommender system. The current state-of-the-art methods represented by HNSW usually generate indexes with a big memory footprint, restricting the scale of data they can handle, except resorting to a hybrid index with external storage. The space-partitioning learned indexes, which only occupy a small memory, have made great breakthroughs in recent years. However, these methods rely on a large amount of labeled data for supervised learning, so model complexity affects the generalization.

Math formulas (e.g., "distance = speed X time'') serve as one of the fundamental commonsense knowledge in human cognition, where humans naturally acquire and manipulate them in logical thinking for mathematical reasoning problems. However, existing reasoning models mainly focus on learning heuristic linguistics or patterns to generate answers, but do not pay enough attention on learning with such formula knowledge. Thus, they are not transparent (thus uninterpretable) in terms of understanding and grasping basic mathematical logic. In this paper, to promote a step forward in the domain, we first construct two datasets (Math23K-F and MAWPS-F) with precise annotations of formula usage in each reasoning step for math word problems. Especially, our datasets are refined on the benchmark datasets, and thus ensure the generality and comparability for relevant research. Then, we propose a novel Formula-mastered Solver (FOMAS) with the guidance of mastering formula knowledge to solve the problems. Specifically, we establish FOMAS with two systems drawing insight from the dual process theory, including a Knowledge System and a Reasoning System, to learn and apply formula knowledge, respectively. The Knowledge System accumulates the math formulas, where we propose a novel pretraining manner to mimic how humans grasp the mathematical logic behind them. Then, in the Reasoning System, we develop elaborate formula-guided symbol prediction and goal generation methods that retrieve the necessary formula knowledge from Knowledge System to improve both reasoning accuracy and interpretability. It organically simulates how humans conduct complex reasoning under the explicit instruction of math formulas. Experimental results prove that FOMAS has a stronger reasoning ability and achieves a more interpretable reasoning process, which verifies the necessity of introducing formula knowledge transparently.