Yao Lu

I am a Postdoc at Peking University.

I got my PhD from Australian National University and Master from University of Helsinki, both in computer science.

My research interest is neural networks, including their theory, architectures, learning algorithms and dynamics.

Email  /  Google Scholar  /  Twitter  /  Github

profile photo
Publications
Learning Sequence Attractors in Hopfield Networks with Hidden Neurons
Yao Lu, Si Wu
NeurIPS Workshop on Associative Memory & Hopfield Networks, 2023
[paper] [code]

We pointed out the problem of classical Hopfield networks in generating sequences and solved it by adding hidden neurons. A new algorithm with theoretical guarantees is proposed to learn sequences as attractors for the networks with hidden neurons.
Bidirectionally Self-Normalizing Neural Networks
Yao Lu, Stephen Gould, Thalaiyasingam Ajanthan
Neural Networks, 2023
[paper] [code] [slides]

We theoretically solve the vanishing/exploding gradients problem in neural networks. Key idea: constraining signal propagation in both directions via a new class of activation functions and orthogonal weight matrices based on high-dimensional probability theory.
Learning to Estimate Hidden Motions with Global Motion Aggregation
Shihao Jiang, Dylan Campbell, Yao Lu, Hongdong Li, Richard Hartley
IEEE International Conference on Computer Vision (ICCV), 2021
[paper] [code]

We obtained the state-of-art optical flow estimation by integrating global motion features to handle occlusion.
Learning Optical Flow from a Few Matches
Shihao Jiang, Yao Lu, Hongdong Li, Richard Hartley
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2021
[paper] [code]

We found high performance optical flow estimation can be achieved with a few candidate correspondences.
Devon: Deformable Volume Network for Learning Optical Flow
Yao Lu, Jack Valmadre, Heng Wang, Juho Kannala, Mehrtash Harandi, Philip Torr
IEEE Winter Conference on Applications of Computer Vision (WACV), 2020
[paper] [code] [slides]

We pointed out the problem of image warping in estimating optical flow and proposed the deformable cost volume to solve the problem.
Doubly Stochastic Neighbor Embedding on Spheres
Yao Lu, Jukka Corander, Zhirong Yang
Pattern Recognition Letters, 2019
[paper] [code] [demo]

We proposed a new method for data visualization, making it on spheres!
Learning Image Relations with Contrast Association Networks
Yao Lu, Zhirong Yang, Juho Kannala, Samuel Kaski
International Joint Conference on Neural Networks (IJCNN), 2019
[paper] [code] [slides]

We proposed a new neural network module, Contrast Association Units, to model the relations between two sets of input variables.
Block Mean Approximation for Efficient Second Order Optimization
Yao Lu, Mehrtash Harandi, Richard Hartley, Razvan Pascanu
ICML workshop on Modern Trends in Nonconvex Optimization for Machine Learning, 2018
[paper] [code] [slides]

We proposed a new matrix approximation method which allows efficient matrix inversion. We then applied the method to second order optimization algorithms for training neural networks.
Oblivious Neural Network Predictions via MiniONN Transformations
Jian Liu, Mika Juuti, Yao Lu, N. Asokan
ACM Conference on Computer and Communications Security (CCS), 2017
[paper] [code]

We proposed a new method for privacy-preserving predictions with trained neural networks.
Unsupervised Learning on Neural Network Outputs: With Application in Zero-shot Learning
Yao Lu
International Joint Conference on Artificial Intelligence (IJCAI), 2016
[paper] [code]

We found visual attributes of object classes can be unsuperisedly learned by applying Independent Component Analysis on the softmax outputs of a trained ConvNet. We showed such attributes can be useful for object recognition by performing zero-shot learning experiments on the ImageNet dataset of over 20,000 object classes.
A Fast Projected Fixed-Point Algorithm for Large Graph Matching
Yao Lu, Kaizhu Huang, Cheng-Lin Liu
Pattern Recognition, 2016
[paper] [code]

We designed a fast graph matching algorithm with time complexity O(n^3) per iteration, where n is the size of a graph. We proved its convergence rate. It takes within 10 seconds to match two graphs of 1,000 nodes on a PC.
An Algorithm for Maximum Common Subgraph of Planar Triangulation Graphs
Yao Lu, Horst Bunke, Cheng-Lin Liu
Graph-Based Representations in Pattern Recognition (GbR), 2013
[paper] [code]

We designed a fast Maximum Common Subgraph (MCS) algorithm for Planar Triangulation Graphs. Its time complexity is O(mnk), where n is the size of one graph, m is the size of the other graph and k is the size of their MCS.
Traveling Bumps and Their Collisions in a Two-Dimensional Neural Field
Yao Lu, Yuzuru Sato, Shun-ichi Amari
Neural Computation, 2011
[paper] [code]

We found a unique traveling bump solution, which was unknown to exist before, in a set of two-dimensional neural network equations.
Projects
4D convolution and cost volume in CUDA
Yao Lu
[code]
A Simple ConvNet in 200 MATLAB lines
Yao Lu
[code]
Twitter Sentiment Analysis using ConvNet
Han Xiao, Yao Lu
[code]

Site template from Jon Barron