Many interesting problems in machine learning are being revisited with new deep tools. For graph-based semi-supervised learning, a recent important development is graph convolutional networks (GCNs), which nicely integrate local vertex features and topology the layers. Although GCN model compares favorably other state-of-the-art methods, its mechanisms not clear it still requires considerable amount of labeled data for validation selection. In this paper, we develop deeper insights into...

Many interesting problems in machine learning are being revisited with new deep tools. For graph-based semisupervised learning, a recent important development is graph convolutional networks (GCNs), which nicely integrate local vertex features and topology the layers. Although GCN model compares favorably other state-of-the-art methods, its mechanisms not clear it still requires considerable amount of labeled data for validation selection. In this paper, we develop deeper insights into...

Attributed graph clustering is challenging as it requires joint modelling of structures and node attributes. Recent progress on convolutional networks has proved that convolution effective in combining structural content information, several recent methods based have achieved promising performance some real attributed networks. However, there limited understanding how affects to properly use optimize for different graphs. Existing essentially a fixed low order only takes into account...

Graph-based methods have been demonstrated as one of the most effective approaches for semi-supervised learning, they can exploit connectivity patterns between labeled and unlabeled data samples to improve learning performance. However, existing graph-based either are limited in their ability jointly model graph structures features, such classical label propagation methods, or require a considerable amount training validation due high complexity, recent neural-network-based methods. In this...

User intent classification plays a vital role in dialogue systems. Since user may frequently change over time many realistic scenarios, unknown (new) detection has become an essential problem, where the study just begun. This paper proposes semantic-enhanced Gaussian mixture model (SEG) for detection. In particular, we utterance embeddings with distribution and inject dynamic class semantic information into means, which enables learning more class-concentrated that help to facilitate...

Han Liu, Xiaotong Zhang, Lu Fan, Xuandi Fu, Qimai Li, Xiao-Ming Wu, Albert Y.S. Lam. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and 9th International Joint (EMNLP-IJCNLP). 2019.

The key issue of few-shot learning is to generalize. This paper proposes a large margin principle improve the generalization capacity metric based methods for learning. To realize it, we develop unified framework learn more discriminative space by augmenting classification loss function with distance training. Extensive experiments on two state-of-the-art methods, graph neural networks and prototypical networks, show that our method can performance existing models substantially very little...

User preference modeling is a vital yet challenging problem in personalized product search. In recent years, latent space based methods have achieved state-of-the-art performance by jointly learning semantic representations of products, users, and text tokens. However, existing are limited their ability to model user preferences. They typically represent users the products they visited short span time using attentive models lack exploit relational information such as user-product...

Attributed graph clustering is a challenging task as it requires to jointly model structure and node attributes. Although recent advances in convolutional networks have shown the effectiveness of convolution combining structural content information, there limited understanding how properly apply for attributed clustering. Previous methods commonly use fixed low order convolution, which only aggregates information few-hop neighbours hence cannot fully capture cluster structures diverse...

Attributed graph clustering is challenging as it requires joint modelling of structures and node attributes. Recent progress on convolutional networks has proved that convolution effective in combining structural content information, several recent methods based have achieved promising performance some real attributed networks. However, there limited understanding how affects to properly use optimize for different graphs. Existing essentially a fixed low order only takes into account...

A fundamental challenge for multi-task learning is that different tasks may conflict with each other when they are solved jointly, and a cause of this phenomenon conflicting gradients during optimization. Recent works attempt to mitigate the influence by directly altering based on some criteria. However, our empirical study shows ``gradient surgery'' cannot effectively reduce occurrence gradients. In paper, we take approach from root. essence, investigate task w.r.t. shared network layer,...

Graph-based methods have been demonstrated as one of the most effective approaches for semi-supervised learning, they can exploit connectivity patterns between labeled and unlabeled data samples to improve learning performance. However, existing graph-based either are limited in their ability jointly model graph structures features, such classical label propagation methods, or require a considerable amount training validation due high complexity, recent neural-network-based methods. In this...

Graph convolutional neural networks (GCN) have been the model of choice for graph representation learning, which is mainly due to effective design convolution that computes a node by aggregating those its neighbors. However, existing GCN variants commonly use 1-D solely operates on object link without exploring informative relational information among attributes. This significantly limits their modeling capability and may lead inferior performance noisy sparse real-world networks. In this...

Using reinforcement learning with human feedback (RLHF) has shown significant promise in fine-tuning diffusion models. Previous methods start by training a reward model that aligns preferences, then leverage RL techniques to fine-tune the underlying However, crafting an efficient demands extensive datasets, optimal architecture, and manual hyperparameter tuning, making process both time cost-intensive. The direct preference optimization (DPO) method, effective large language models,...

In recent years, Multi-Agent Path Finding (MAPF) has attracted attention from the fields of both Operations Research (OR) and Reinforcement Learning (RL). However, in 2021 Flatland3 Challenge, a competition on MAPF, best RL method scored only 27.9, far less than OR method. This paper proposes new solution to which scores 125.3, several times higher before. We creatively apply novel network architecture, TreeLSTM, MAPF our solution. Together with other techniques, including reward shaping,...

Clustering uncertain data is an essential task in mining for the internet of things. Possible world based algorithms seem promising clustering data. However, there are two issues existing possible algorithms: (1) They rely on all worlds and treat them equally, but some marginal may cause negative effects. (2) do not well utilize consistency among worlds, since they conduct or construct affinity matrix each independently. In this paper, we propose a representative consistent (RPC) algorithm...

Linearized Graph Neural Networks (GNNs) have attracted great attention in recent years for graph representation learning. Compared with nonlinear Network (GNN) models, linearized GNNs are much more time-efficient and can achieve comparable performances on typical downstream tasks such as node classification. Although some GNN variants purposely crafted to mitigate ``over-smoothing", empirical studies demonstrate that they still somehow suffer from this issue. In paper, we instead relate...

Decision-based methods have shown to be effective in black-box adversarial attacks, as they can obtain satisfactory performance and only require access the final model prediction. Gradient estimation is a critical step it will directly affect query efficiency. Recent works attempted utilize gradient priors facilitate score-based better results. However, these still suffer from edge discrepancy issue successive iteration direction issue, thus are difficult simply extend decision-based...

Decision-based methods have shown to be effective in black-box adversarial attacks, as they can obtain satisfactory performance and only require access the final model prediction. Gradient estimation is a critical step it will directly affect query efficiency. Recent works attempted utilize gradient priors facilitate score-based better results. However, these still suffer from edge discrepancy issue successive iteration direction issue, thus are difficult simply extend decision-based...