Time | Topic | Contents | Presenter |
01/14 | Introduction | (1) Motivation (2) Syllabus and grading policy (3) Random graphs (4) Paper presentations | Jianzhu Ma |
01/16 | Network Visualization | (1) Cytoscape (2) HiView (3) NDEx | Jianzhu Ma |
01/21 | Introduction to Deep Learning | (1) Basic concepts, MLP (2) CNN (3) RNN (4) LSTM (5) Resnet (6) GNN | Jianzhu Ma |
01/23 | Network Motifs | (1) Network motifs in biological networks (2) G-tries algorithm | Jianzhu Ma |
01/28 | PageRank | (1) PageRank algorithm (2) Personalized PageRank and Random Walk algorithm (3) Random Walk in Cancer | Jianzhu Ma |
01/30 | Network Community Detection I | (1) Bron-Kerbosch algorithm (2) Kernighan-Lin algorithm (3) Louvain algorithm | Jianzhu Ma |
02/04 | Network Community Detection II | (1) Spectual clustering (2) Spectual modularity maximization (3) Hierarchical graph clustering Optional reading: Awesome community detection algorithms [github] | Jianzhu Ma |
02/06 | Network Alignment | (1) PathBLAST (2) IsoRank (3) Representation-based network alignments Optional Reading: (1) REGAL: Representation Learning-based Graph Alignment [pdf] (2) Deep Adversarial Network Alignment [pdf] | Jianzhu Ma |
02/11 | Structural Roles in Network | (1) Roles and Communities (2) ReFeX (3) RolX Optional Reading: (1) From Community to Role-based Graph Embeddings [pdf] (2) Role Discovery in Networks [pdf] (3) Introduction to social network methods Chapter 14 [url] | Jianzhu Ma |
02/13 | Graph Summarization | (1) Graph Dedensification (2) Vocabulary-based Summarization (3) SlashBurn algorithm (4) TimeCrunch algorithm Reading: (1) VOG: Summarizing and Understanding Large Graphs [pdf] (2) SlashBurn: Graph Compression and Mining beyond Caveman Communities [pdf] (3) Latent Network Summarization: Bridging Network Embedding and Summarization [pdf] | Jianzhu Ma |
02/18 | Adversarial Attacks and Defenses | (1) Fast Gradient Sign Method (2) Jacobian-based Saliency Map Attack (3) DeepFool (4) Universal Adversarial Perturbations Optional reading: 1. The Limitations of Deep Learning in Adversarial Settings [pdf] 2. Why deep-learning AIs are so easy to fool [nature article] | Jianzhu Ma |
02/20 | Graphical Models I | (1) General concepts (2) Exact inference (3) Sum-product algorithm (4) Max-product algorithm (5) Conditional Random Fields Optional textbooks: (1) “Probabilistic Graphical Models” by By Daphne Koller and Nir Friedman (2) “Graphical Models, Exponential Families, and Variational Inference” by Martin J. Wainwright and Michael I. Jordan (3) Chapter 8 of “Pattern Recognition and Machine Learning” by Christopher M. Bishop | Jianzhu Ma |
02/25 | Graphical Models II | (1) Structure learning (2) Gaussian Graphical Model (3) Pseudo-likelihood approximation (4) Protein contact prediction (5) Deep-Learning-based structure learning Optional reading: (1) Sparse Inverse Covariance Estimation with the Graphical Lasso [pdf] (2) High-Dimensional Graphs and Variable Selection with the Lasso [pdf] (3) Protein 3D Structure Computed from Evolutionary Sequence Variation [pdf] (4) AlphaFold: Using AI for scientific discovery [BLOG POST] (5) Foldit computer game [link] | Jianzhu Ma |
02/27 | Large-scale Graphs | Reading: (1) Distributed GraphLab: A Framework for Machine Learning and Data Mining in the Cloud [pdf] (2) GraphLab: A New Framework For Parallel Machine Learning [pdf] (3) GraphChi: Large-Scale Graph Computation on Just a PC [pdf] (4) Software demo [download] | Rajat Verma, Jiqian Dong |
03/03 | Network Embedding I | Reading: (1) DeepWalk: Online Learning of Social Representations [pdf] (2) Don't Walk, Skip! Online Learning of Multi-scale Network Embeddings [pdf] (3) LINE: Large-scale Information Network Embedding [pdf] Optional reading: Efficient Estimation of Word Representations in Vector Space [pdf] | Juan Shu, Qian Zhang, Jiajun Liang |
03/05 | Network Embedding II | Reading: (1) node2vec: Scalable Feature Learning for Networks [pdf] (2) struc2vec: Learning Node Representations from Structural Identity [pdf] (3) Inductive Representation Learning on Large Graphs [pdf] Optional reading: Learning Role-based Graph Embeddings [pdf] | Vinith Budde, Susheel Suresh |
03/10 | Learning on Knowledge Graphs | Reading: (1) TransE paper: Translating Embeddings for Modeling Multi-relational Data [pdf] (2) TransH paper: Knowledge Graph Embedding by Translating on Hyperplanes [pdf] (3) TransR paper: Learning Entity and Relation Embeddings for Knowledge Graph Completion [pdf] Optional reading: Mining Knowledge Graphs from Text [Tutorial] | Xiaonan Jing, Abida Sanjana, Amira Mamoun |
03/12 | Deep Learning on Graphs | (1) General concepts (2) The graph neural network model (3) Popular datasets (4) Gated Graph Sequence Neural Networks Optional reading: (1) The graph neural network model [pdf] (2) Gated Graph Sequence Neural Networks [https:arxiv.orgpdf1511.05493.pdf [pdf] | Jianzhu Ma |
03/24 | Deep Reinforcement Learning | (1) Basic reinforcement learning (2) Applications (3) Q-learning (4) Policy Gradient (5) Actor-Critic Algorithm Optional reading: (1) Yuyi Li: Deep reinforcement learning [pdf] | Jianzhu Ma |
03/26 | Graph Convolutional Networks I | Reading: (1) Spectral networks and locally connected networks on graphs [pdf] (2) Deep Convolutional Networks on Graph-Structured Data [pdf] | Matthew Muhoberac, Adam Johnston, Connor Beveridge |
03/31 | Graph Convolutional Networks II | Reading: (1) Semi-Supervised Classification with Graph Convolutional Networks [pdf] (2) Modeling Relational Data with Graph Convolutional Networks [pdf] (3) Column Networks for Collective Classification [pdf] | Jiang Nan, Pang Qiyuan, Liang Senwei, Xue Jiawei |
04/02 | Generative Models of Graphs | Reading: (1) Learning Deep Generative Models of Graphs [pdf] (2) GraphRNN: Generating Realistic Graphs with Deep Auto-regressive Models [pdf] (3) Junction tree variational autoencoder for molecular graph generation [pdf] | Omar Eldaghar, Evzenie Coupkova |
04/07 | Adversarial Attack on Graphs | Reading: (1) Adversarial Attacks on Neural Networks for Graph Data [pdf] (2) Investigating Robustness and Interpretability of Link Prediction via Adversarial Modifications [pdf] Optional reading: (1) Adversarial Attack on Graph Structured Data [pdf] (2) Adversarial Attacks on Node Embeddings via Graph Poisoning [pdf] (3) Attacking Graph Convolutional Networks via Rewiring [pdf] | Jungeum Kim, Jiacheng Li |
04/09 | Higher-order Networks | Reading: (1) Structural deep embedding for hypernetworks [pdf] (2) Hyper2vec: Biased random walk for hyper-network embedding [pdf] (3) Hyper-SAGNN: a self-attention based graph neural network for hypergraphs [pdf] | Xiao Wang, Maria Pacheco, Sean T Flannery |
04/14 | Deep Reinforcement Learning on Graphs I | Reading: (1) NerveNet: Learning Structured Policy with Graph Neural Networks [pdf] (2) Playing Text-Adventure Games with Graph-Based Deep Reinforcement Learning [pdf] | Zhi Huang, Xiaoyu Xiang |
04/16 | Deep Reinforcement Learning on Graphs II | Reading: (1) Graph Convolutional Policy Network for Goal-Directed Molecular Graph Generation [pdf] (2) MolGAN: An implicit generative model for small molecular graphs [pdf] (3) Learning Multimodal Graph-to-Graph Translation for Molecular Optimization [pdf] | Kendal Graham Norman, Arvind Sundaram |
04/21 | Project presentations | TBD | Students |
04/23 | Project presentations | TBD | Students |
04/28 | Project presentations | TBD | Students |
04/30 | Project presentations | TBD | Students |
|