Will Y. Zou

Welcome to my homepage. With interest in deep learning, scalable production ML systems, and computer vision, my resarch highlights include action recognition with Independent Subspace Analysis, and object detection with convolutional networks. With natural languages, we took the word-embedding idea to machine translation. I also worked simulation of cognition of approximate visual number sense and heterogeneous causal learning for contextual bandits.

Obtained B.A. and M.Eng. from University of Cambridge, Ph.D. at Stanford University. During PhD, I had the fortune to work with the Stanford AI Lab, NLP Group, Stanford Psychology, and NEC Labs America Media Analytics Group on deep learning applications to computer vision. In industry, I have worked with many wonderful people and teams over the years.

Publications

Publish with Will Y. Zou

Aggregating Counterfactual Queries with Neural Architectures to Learn Complex Action Policies

Heterogeneous Causal Learning for Effectiveness Optimization in User Marketing.

arxiv paper

RapidScorer: Fast Tree Ensemble Evaluation by Maximizing Compactness in Data Level Parallelization.

ACM SIGKDD (KDD) 2018

Paper

Invariance for Perceptual Recognition through Deep Learning.

Ph.D. Thesis, Stanford University.

Initial Competence and Development of a Sense of Number in a Deep Neural Network.

Submitted to Cognition , International Journal of Cognitive Science.

Unsupervised learning with neural networks in the formation of abstract number features in deep layers.

Time-course of Development and Factors Affecting Sensitivity to Visual Numbers Through Unsupervised Learning.

NCPW, 2014.

[Abstract]

[Stoianov & Zorzi]

Generic Object Detection with Dense Neural Patterns and Regionlets.

British Machine Vision Conference (BMVC), 2014.

arXiv preprint 1404.4316

Competitive and very fast generic object detection (64x Caffe), combining cascaded boosting classifiers with convolutional nets.

ImageNet Detection Results

Bilingual Word Embeddings for Phrase-based Machine Translation.

Empirical Methods in Natural Language Processing (EMNLP)

[PDF]

[bib]

[Poster]

[Bilingual Embeddings]

[Project Page]

[Slides]

[Word Similarity Dataset]

The first application of deep learning word embeddings in machine translation.

Progressive Development of the Number Sense in a Deep Neural Network.

Annual Conference of the Cognitive Science Society (CogSci)

[Poster]

[bib]

[Link]

Deep Learning of Invariant Features via Simulated Fixations in Video.

Neural Information Processing Systems (NIPS)

[Invariance visualizations]

Illustration of applying slowness in neural networks to learn invariant features on still images.

Unsupervised Learning of Visual Invariance with Temporal Coherence.

Neural Information Processing Systems (NIPS) Workshop on Deep Learning and Unsupervised Feature Learning

[PDF]

[bib]

Haptic Belt with Pedestrian Detection [Tech Demo].

Neural Information Processing Systems

[Poster]

[Video]

Real-time, 360 degree pedestrian detection on a GPU-equipped back-pack with 'third-sense' haptic vibration belt.

Real-time People Detection with Convolutional Neural Networks in a Haptic Device.

TEDx Stanford

Learning Hierarchical Spatio-temporal Features for Action Recognition with Independent Subspace Analysis.
Quoc V. Le, Will Y. Zou, Serena Y. Yeung, Andrew Y. Ng.
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2011. (Oral presentation)
[PDF] [Appendix] [Multi-resolution Code (~53%AP)] [Simple Code (~50.5%AP)] [Notes] [Link to Hollywood2].
State-of-the-art action recognition with unsupervised feature learning.
[Example Actions on Youtube]
[Features] already trained on Hollywood2
[Here] is some code to extract cool ISA features on images instead of videos, e.g. this paper.