Terrain Generation

Generate Terrain using a single 128*128 image.

Equivariant Hypergraph Diffusion Neural Operators

In this work, we are inspired by hypergraph diffusion algorithms and design a novel HNN architecture that holds provable expressiveness while keeping efficiency.

Neural Higher-order Pattern (Motif) Prediction in Temporal Networks

Dynamic systems that consist of a set of interacting elements can be abstracted as temporal networks. Recently, higher-order patterns that involve multiple interacting nodes have been found crucial to indicate domain-specific laws of different temporal networks. Here, we propose the first model, named HIT, for higher-order pattern prediction in temporal hypergraphs. Particularly, we focus on predicting three types of common but important interaction patterns involving three interacting elements in temporal networks, which could be extended to even higher-order patterns.

Inductive Representation Learning in Temporal Networks via Causal Anonymous Walks

We introduce Causal Anonymous Walks (CAW) for inductive representation learning in temporal networks. CAWs are extracted by temporal random walks and work as automatic retrieval of temporal network motifs to represent network dynamics, while avoiding the time-consuming selection and counting of those motifs.

Generative-View-Correlation-Adaptation-for-Semi-Supervised-Multi-View-Learning

Multi-view learning (MVL) explores the data extracted from multiple resources. It assumes that the complementary information between different views could be revealed to further improve the learning performance. There are two challenges. First, it is difficult to effectively combine the different view data while still fully preserve the view-specific information. Second, multi-view datasets are usually small, which means the model can be easily overfitted. To address the challenges, we propose a novel View-Correlation Adaptation (VCA) framework in semisupervised fashion. A semi-supervised data augmentation me-thod is designed to generate extra features and labels based on both labeled and unlabeled samples. In addition, a cross-view adversarial training strategy is proposed to explore the structural information from one view and help the representation learning of the other view. Moreover, an effective and simple fusion network is proposed for the late fusion stage. In our model, all networks are jointly trained in an end-to-end fashion. Extensive experiments demonstrate that our approach is effective and stable compared with other state-of-the-art methods

Semi-supervised Multi-label Learning

We proposed a method which jointly explores the inter-instance feature distribution and intra-instance label relation simultaneously. We also applied co-training with pseudo labels to further improve learning performance. A relation network is further proposed to explore the cross-label relation knowledge.

Generative Multi View Human Action Recognition

We implemented a generative model to fully utilize the pair-wise information in different views. Our model can handle complete view, partial view and missing view at the same time. We also proposed a simple yet effective View Correlation Discovery Network to do the label-level fusion.

Action Recognition based on EMG signal

EMG stands for electromyography. It is the study of muscle electrical signals. Since EMG signal always contains noise, conventional methods use Root Mean Square(RMS) to reduce the noise and extract feature. However the RMS misses a lot of useful information. To reduce the dimension and avoid shifting in the signal, LDA and PCA are utilized. Traditional machine learning methods are implemented for the classification, including Random Forest, KNN, SVM. These methods lack a unified and principled objective function. Therefore, we proposed a method using Fast Fourier Transform(FFT) to extract the features. And then applied a LSTM model to do the classification. Our result is better than the previous traditional methods.

Password Analysis and Dictionary Attack

We analyzed several leaked password datasets and used machine learning to create a more general password dictionary.

Acedemic Network

I analyzed the relationship of topics and authors using some algorithms.

DTU with Bluetooth

Traditional DTUs do not have bluetooth. To make it easier to use, we implemented a bluetooth part in it.

Evil Twin

We proposed several methods to defend Evil Twin.