Yicun Liu

I am a Research Engineer at Google DeepMind, working on conversation agents that renovate some of the largest Google products.

Previously, I was an ML Research Engineer at Twitter Cortex, where I transformed research ideas like language embeddings and continuous learning to its ranking systems. I also spent a year on vision proptotypes at SenseTime in its early days.

I received M.S. from Columbia University, advised by Professor John R. Kender. I received my B.Eng from The Chinese University of Hong Kong, where I was advised by Professor Loy Change Chen and Dr. Jimmy Ren.

GitHub  /  LinkedIn  /  Scholar  /  Photos

Research

I am genuinely interested in applied research, when it meets production scale and generalization challenges in the real-world.

Model Size Reduction Using Frequency Based Double Hashing for Recommender Systems
C. Zhang*, Y. Liu*, Y. Xie, S. Ktena, A. Tejani, A. Gupta, P. Myana, D. Dilipkumar, S. Paul, I. Ihara, P. Upadhyaya, F. Huszar, W. Shi

ACM Conference on Recommender Systems (RecSys), 2020
pdf / video / bibtex

We propose a hybrid hashing method to combine frequency hashing and double hashing techniques for model size reduction of large-scale recommender systems.

Self-Guided Novel View Synthesis via Elastic Displacement Network
Yicun Liu, Jiawei Zhang, Ye Ma, Jimmy Ren

IEEE Winter Conference on Applications of Computer Vision (WACV), 2020
pdf / video / bibtex

To generate high-quality novel view, we design a non-discrete scene representation for 3D transformation and use the edge info in the input image for spatial filtering.

Visually Imbalanced Stereo Matching
Yicun Liu*, Jimmy Ren*, Jiawei Zhang, Jianbo Liu, Mude Lin

IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2020
pdf / video / code / bibtex

Resemble to the human visual system (HVS), stereo machines collapse under imbalanced stereo inputs. We show that guided by the rough object contour, the corrupted view can be restored, and stereopsis can be regenerated.

Learning to Deblur Face Images via Sketch Synthesis
Songnan Lin, Jiawei Zhang, Jinshan Pan, Yicun Liu, Yongtian Wang, Jing Chen, Jimmy Ren

AAAI Conference on Artificial Intelligence (AAAI), 2020
pdf / bibtex

As most face images share some common global structures which can be modeled well by sketch information, we propose to learn face sketches first to help the motion blur estimation.

Learning Selfie-Friendly Abstraction from Artistic Style Images
Yicun Liu, Jimmy Ren, Jianbo Liu, Jiawei Zhang, Xiaohao Chen

Asian Conference on Machine Learning (ACML), 2018   (Long Oral Presentation)
pdf / video / code / bibtex

By exploiting the properties of the gradient domain, we establish a selfie-friendly stylization framework that preserved natural skin color and facial structure.

Projects

Improving Mobile Phone's Image Quality by Deep Learning
Cen Huang*, Yicun Liu*, supervised by Prof. Loy Change Chen

Final Year Thesis (Undergraduate), 2018
pdf / slides

We bridge the gap of image quality in mobile phones and DSLRs by designing a multi-domain image translation framework, which learns the comprehensive enhancement transformation from heterogeneous real-world datasets.

Understanding MOBA Player Experience: What Can We Know from Social Big Data?
J. Fan, C. Huang, Y. Liu, X. Lyu, Z. Wang, supervised by Prof. Rosanna Yuen-Yan Chan

Technical Report, 2017
pdf / dataset

Based on over 400K posts crawled from League of Legends NA forum, we establish a framework that assesses the player experience from the sentiment polarity of the posts.

Forward Private Dynamic Searchable Symmetric Encryption: A Review
Yicun Liu, supervised by Prof. Sherman S. M. Chow

Technical Report, 2016
pdf / slides

Dynamic Searchable Symmetric Encryption (DSSE) aims at making possible queries and updates over an encrypted database on an untrusted server, with minimum exposure about user data to the server.

Service
Reviewer: ICCV 2022, AAAI 2023-2024, WACV 2020-2021

Teaching Assistant: COMS 4731 Computer Vision, 19 Fall

This website's template credits to :)