CLeAR

Generative Models

Generative models capture underlying data distributions. In contrast to discriminative models, generative models can be used to sample / generate new data and tell us the likelihood of a given sample. Famous generative models include Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs). At CLeAR, we have developed new deep generative models (see below) and are particularly interested in structured temporal models that can capture complex relationships, yet are interpretable and useful for planning and decision-making.

★ CLeAR Group Members

Deep Explicit Duration Switching Models for Time Series

C21-5Deep Explicit Duration Switching Models for Time Series [PDF]
Abdul Fatir Ansari★, Konstantinos Benidis, Richard Kurle, Ali Caner Turkmen, Harold Soh★, Alex Smola, Bernie Wang, Tim Januschowski
Neural Information Processing Systems (NeurIPS), 2021
TL;DR: We propose a deep switching state space model that can capture both state-dependent and time-dependent switching patterns in time series data.

    More Links: [BibTeX] 
    @inproceedings{ansari2021deep,
author    = {Abdul Fatir Ansari and Konstantinos Benidis and Richard Kurle and Ali Caner Turkmen and Harold Soh and Alex Smola and Bernie Wang and Tim Januschowski},
title     = {Deep Explicit Duration Switching Models for Time Series},
year      = {2021},
booktitle = {Neural Information Processing Systems (NeurIPS)}}
Refining Deep Generative Models via Discriminator Gradient Flow

C21-1Refining Deep Generative Models via Discriminator Gradient Flow [PDF] [Blog]
Abdul Fatir Ansari ★, Ming Liang Ang ★, and Harold Soh ★
International Conference on Learning Representations (ICLR), 2021
TL;DR: We propose new gradient flow methods for improving samples from deep generative models.

    More Links: [Github] [BibTeX] 
    @inproceedings{fatir2021refining,
title={Refining Deep Generative Models via Discriminator Gradient Flow},
author={Ansari, Abdul Fatir and Ang, Ming Liang and Soh, Harold},
booktitle={International Conference on Learning Representations (ICLR)},
year={2021}}
A Characteristic Function Approach to Deep Implicit Generative Modeling

C20-2OralA Characteristic Function Approach to Deep Implicit Generative Modeling [PDF]
Abdul Fatir Ansari★, Jonathan Scarlett, and Harold Soh★
The Conference on Computer Vision and Pattern Recognition (CVPR), 2020
TL;DR: We present a new loss for training generative models based on characteristic functions.

    More Links: [Github] [Video] [BibTeX] 
    @misc{ansari2019characteristic,
title={A Characteristic Function Approach to Deep Implicit Generative Modeling},
author={Abdul Fatir Ansari and Jonathan Scarlett and Harold Soh},
year={2019},
eprint={1909.07425},
archivePrefix={arXiv}}
Factorized Inference in Deep Markov Models for Incomplete Multimodal Time Series

C20-1OralFactorized Inference in Deep Markov Models for Incomplete Multimodal Time Series [PDF]
Zhi-Xuan Tan, Harold Soh ★, Desmond Ong
Thirty-Fourth AAAI Conference on Artificial Intelligence (AAAI), 2020
TL;DR: We present a new factorized inference scheme for learing time series with deep Markov models.

    More Links: [Github] [BibTeX] 
    @inproceedings{tan2020mdmm,
title={Factorized Inference in Deep Markov Models for Incomplete Multimodal Time Series},
author={Tan, Zhi-Xuan and Soh, Harold and Ong, Desmond},
booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
volume={34}}
Hyperprior Induced Unsupervised Disentanglement of Latent Representations

C19-5Hyperprior Induced Unsupervised Disentanglement of Latent Representations [PDF]
Abdul Fatir Ansari★ and Harold Soh ★
Thirty-Third AAAI Conference on Artificial Intelligence (AAAI), 2019
TL;DR: We show how a simple hierarchical Bayesian VAE leads to controllable disentanglement and how the Bayesian approach is connected to other disentanglement models.

    More Links: [Github] [BibTeX] 
    @inproceedings{ansari2019hyperprior,
title={Hyperprior induced unsupervised disentanglement of latent representations},
author={Ansari, Abdul Fatir and Soh, Harold},
booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
volume={33},
pages={3175--3182},
year={2019}}
Applying Probabilistic Programming to Affective Computing

J19-1Applying Probabilistic Programming to Affective Computing [PDF]
Desmond Ong, Harold Soh ★, Jamil Zaki and Noah Goodman
IEEE Transactions on Affective Computing, 2019
TL;DR: We show how probabilitic programming can be used to enable machines to infer human emotions.

    More Links: [BibTeX] 
    @article{ong2019applying,
title={Applying Probabilistic Programming to Affective Computing},
author={Ong, Desmond and Soh, Harold and Zaki, Jamil and Goodman, Noah},
journal={IEEE Transactions on Affective Computing},
year={2019},
publisher={IEEE}}
Generation Meets Recommendation: Proposing Novel Items for Groups of Users

C18-4BEST PAPER FINALISTGeneration Meets Recommendation: Proposing Novel Items for Groups of Users [PDF]
Vo Vinh Thanh★ and Harold Soh ★
ACM Recommender Systems Conference (RecSys), 2018
TL;DR: Recommendation for creators rather than consumers. We take a first stab at answering "what should I make and who will it appeal to?"

    More Links: [Publisher Link] [Poster] [Slides] [BibTeX] 
    @inproceedings{vo2018generation,
title={Generation meets recommendation: proposing novel items for groups of users},
author={Vo, Thanh Vinh and Soh, Harold},
booktitle={Proceedings of the 12th ACM Conference on Recommender Systems},
pages={145--153},
year={2018},
organization={ACM}}