MUSEDA: Multilingual Unsupervised and Supervised Embedding for Domain Adaption

Article Computer Science, Software Engineering

Efficient Estimation of Word Representations using Constrained Positional Weighting

Vit Novotny et al.

Summary: This paper presents a positional language model and its improved version, which adapts a sparse attention mechanism to improve speed. The models are evaluated on qualitative criteria and language modeling, demonstrating their interpretability and outperformance compared to other shallow models. The constrained model also shows better performance and faster training speed than the positional model.

JOURNAL OF UNIVERSAL COMPUTER SCIENCE (2022)

添加到收藏夹

Article Computer Science, Artificial Intelligence

Unsupervised deep clustering via contractive feature representation and focal loss

Jinyu Cai et al.

Summary: This study introduces a new joint optimization clustering framework that improves the label assignment mechanism in feature learning and clustering layer through the introduction of contractive representation and utilization of focal loss. Experimental results demonstrate the effectiveness of the proposed method on several clustering tasks.

PATTERN RECOGNITION (2022)

添加到收藏夹

Article Computer Science, Artificial Intelligence

Unsupervised embedded feature learning for deep clustering with stacked sparse auto-encoder

Jinyu Cai et al.

Summary: The paper proposes a deep stacked sparse embedded clustering method that considers both local structure preservation and input sparsity. The deep learning approach jointly learns clustering-oriented features and optimizes cluster label assignments by minimizing both the reconstruction and clustering loss. Comprehensive experiments validate the effectiveness of introducing sparsity and preserving local structure in the proposed method.

EXPERT SYSTEMS WITH APPLICATIONS (2021)

添加到收藏夹

Article Engineering, Multidisciplinary

Gradient Shielding: Towards Understanding Vulnerability of Deep Neural Networks

Zhaoquan Gu et al.

Summary: This paper proposes a new method of attacking deep neural networks by gradient shielding, which ignores insensitive information in images and fools image classification DNNs by adding perturbations. Experimental results demonstrate the effectiveness of this method, leading to a first place win in a competition.

IEEE TRANSACTIONS ON NETWORK SCIENCE AND ENGINEERING (2021)

添加到收藏夹

Article Computer Science, Artificial Intelligence

Self-Training With Progressive Representation Enhancement for Unsupervised Cross-Domain Person Re-Identification

Hang Zhang et al.

Summary: This paper introduces a Self-Training with Progressive Representation Enhancement (PREST) framework, which utilizes multi-scale self-training and view-invariant representation learning to unsupervisedly improve person re-identification performance in the target domain.

IEEE TRANSACTIONS ON IMAGE PROCESSING (2021)

添加到收藏夹

Article Computer Science, Artificial Intelligence