no code implementations • 1 May 2024 • Shikhar Tuli, Chi-Heng Lin, Yen-Chang Hsu, Niraj K. Jha, Yilin Shen, Hongxia Jin
We also propose systematic qualitative and quantitative methods to rigorously test the quality of generated text for non-autoregressive generation.
no code implementations • 2 Dec 2023 • Minchul Kim, Shangqian Gao, Yen-Chang Hsu, Yilin Shen, Hongxia Jin
In this paper, we introduce "Token Fusion" (ToFu), a method that amalgamates the benefits of both token pruning and token merging.
no code implementations • 30 Nov 2023 • James Seale Smith, Yen-Chang Hsu, Zsolt Kira, Yilin Shen, Hongxia Jin
We show that STAMINA outperforms the prior SOTA for the setting of text-to-image continual customization on a 50-concept benchmark composed of landmarks and human faces, with no stored replay data.
no code implementations • 12 Apr 2023 • James Seale Smith, Yen-Chang Hsu, Lingyu Zhang, Ting Hua, Zsolt Kira, Yilin Shen, Hongxia Jin
We show that C-LoRA not only outperforms several baselines for our proposed setting of text-to-image continual customization, which we refer to as Continual Diffusion, but that we achieve a new state-of-the-art in the well-established rehearsal-free continual learning setting for image classification.
no code implementations • 2 Nov 2022 • Ting Hua, Yen-Chang Hsu, Felicity Wang, Qian Lou, Yilin Shen, Hongxia Jin
However, standard SVD treats the parameters within the matrix with equal importance, which is a simple but unrealistic assumption.
no code implementations • ICLR 2022 • Yen-Chang Hsu, Ting Hua, SungEn Chang, Qian Lou, Yilin Shen, Hongxia Jin
In other words, the optimization objective of SVD is not aligned with the trained model's task accuracy.
no code implementations • 31 Mar 2022 • James Seale Smith, Junjiao Tian, Shaunak Halbe, Yen-Chang Hsu, Zsolt Kira
Next, we explore how to leverage knowledge from a pre-trained model in rehearsal-free continual learning and find that vanilla L2 parameter regularization outperforms EWC parameter regularization and feature distillation.
no code implementations • 18 Mar 2022 • Yen-Chang Hsu, James Smith, Yilin Shen, Zsolt Kira, Hongxia Jin
Knowledge distillation (KD) is a substantial strategy for transferring learned knowledge from one neural network model to another.
no code implementations • NAACL 2021 • Ting Hua, Yilin Shen, Changsheng Zhao, Yen-Chang Hsu, Hongxia Jin
Most existing continual learning approaches suffer from low accuracy and performance fluctuation, especially when the distributions of old and new data are significantly different.
no code implementations • CVPR 2022 • Qian Lou, Yen-Chang Hsu, Burak Uzkent, Ting Hua, Yilin Shen, Hongxia Jin
The key primitive is that Dictionary-Lookup-Transformormations (DLT) is proposed to replace Linear Transformation (LT) in multi-modal detectors where each weight in Linear Transformation (LT) is approximately factorized into a smaller dictionary, index, and coefficient.
1 code implementation • NeurIPS 2021 • Junjiao Tian, Dylan Yung, Yen-Chang Hsu, Zsolt Kira
It is well known that vision classification models suffer from poor calibration in the face of data distribution shifts.
no code implementations • ICLR 2022 • Qian Lou, Ting Hua, Yen-Chang Hsu, Yilin Shen, Hongxia Jin
DictFormer significantly reduces the redundancy in the transformer's parameters by replacing the prior transformer's parameters with compact, shared dictionary, a few unshared coefficients, and indices.
no code implementations • 29 Sep 2021 • Junjiao Tian, Yen-Chang Hsu, Yilin Shen, Hongxia Jin, Zsolt Kira
To this end, we theoretically derive two score functions for OOD detection, the covariate shift score and concept shift score, based on the decomposition of KL-divergence for both scores, and propose a geometrically-inspired method (Geometric ODIN) to improve OOD detection under both shifts with only in-distribution data.
no code implementations • ACL 2021 • Yilin Shen, Yen-Chang Hsu, Avik Ray, Hongxia Jin
This paper proposes to train a model with only IND data while supporting both IND intent classification and OOD detection.
2 code implementations • ICCV 2021 • James Smith, Yen-Chang Hsu, Jonathan Balloch, Yilin Shen, Hongxia Jin, Zsolt Kira
Modern computer vision applications suffer from catastrophic forgetting when incrementally learning new concepts over time.
Ranked #5 on Class Incremental Learning on cifar100
1 code implementation • 23 Jan 2021 • James Smith, Jonathan Balloch, Yen-Chang Hsu, Zsolt Kira
Our work investigates whether we can significantly reduce this memory budget by leveraging unlabeled data from an agent's environment in a realistic and challenging continual learning paradigm.
no code implementations • NeurIPS 2020 • Junjiao Tian, Yen-Cheng Liu, Nathan Glaser, Yen-Chang Hsu, Zsolt Kira
Neural Networks can perform poorly when the training label distribution is heavily imbalanced, as well as when the testing data differs from the training distribution.
Ranked #31 on Long-tail Learning on CIFAR-100-LT (ρ=10)
2 code implementations • CVPR 2020 • Yen-Chang Hsu, Yilin Shen, Hongxia Jin, Zsolt Kira
Deep neural networks have attained remarkable performance when applied to data that comes from the same distribution as that of the training set, but can significantly degrade otherwise.
Out-of-Distribution Detection Out of Distribution (OOD) Detection
1 code implementation • ICLR 2019 • Yen-Chang Hsu, Zhaoyang Lv, Joel Schlosser, Phillip Odom, Zsolt Kira
This work presents a new strategy for multi-class classification that requires no class-specific labels, but instead leverages pairwise similarity between examples, which is a weaker form of annotation.
3 code implementations • 30 Oct 2018 • Yen-Chang Hsu, Yen-Cheng Liu, Anita Ramasamy, Zsolt Kira
Continual learning has received a great deal of attention recently with several approaches being proposed.
no code implementations • 28 Jun 2018 • Yen-Chang Hsu, Zhaoyang Lv, Joel Schlosser, Phillip Odom, Zsolt Kira
The proposed objective directly minimizes the negative log-likelihood of cluster assignment with respect to the pairwise constraints, has no hyper-parameters, and demonstrates improved scalability and performance on both supervised learning and unsupervised transfer learning.
Ranked #1 on Ecg Risk Stratification on ngm
1 code implementation • 17 Mar 2018 • Yen-Chang Hsu, Zheng Xu, Zsolt Kira, Jiawei Huang
We utilize the most fundamental property of instance labeling -- the pairwise relationship between pixels -- as the supervision to formulate the learning objective, then apply it to train a fully convolutional network (FCN) for learning to perform pixel-wise clustering.
Ranked #13 on Lane Detection on TuSimple
1 code implementation • ICLR 2018 • Yen-Chang Hsu, Zhaoyang Lv, Zsolt Kira
The key insight is that, in addition to features, we can transfer similarity information and this is sufficient to learn a similarity function and clustering network to perform both domain adaptation and cross-task transfer learning.
no code implementations • 2 Sep 2017 • Zheng Xu, Yen-Chang Hsu, Jiawei Huang
There is an increasing interest on accelerating neural networks for real-time applications.
no code implementations • 5 Dec 2016 • Yen-Chang Hsu, Zhaoyang Lv, Zsolt Kira
We propose that this network can be learned with contrastive loss which is only based on weak binary pair-wise constraints.
2 code implementations • 19 Nov 2015 • Yen-Chang Hsu, Zsolt Kira
Robustness analysis also shows that the method is largely insensitive to the number of clusters.