Kai Yi

ORCID: 0000-0003-0415-3584
Publications
Citations
Views
---
Saved
---
About
Contact & Profiles
Research Areas
  • Domain Adaptation and Few-Shot Learning
  • Multimodal Machine Learning Applications
  • Stochastic Gradient Optimization Techniques
  • RNA and protein synthesis mechanisms
  • Topic Modeling
  • COVID-19 diagnosis using AI
  • Privacy-Preserving Technologies in Data
  • Advanced Steganography and Watermarking Techniques
  • Genomics and Phylogenetic Studies
  • Digital Media Forensic Detection
  • Bioinformatics and Genomic Networks
  • Advanced Neural Network Applications
  • Advanced Graph Neural Networks
  • Probability and Risk Models
  • Chaos-based Image/Signal Encryption
  • Neural Networks and Applications
  • Computational Physics and Python Applications
  • Advanced Image and Video Retrieval Techniques
  • AI in cancer detection
  • Gene expression and cancer classification
  • High-Velocity Impact and Material Behavior
  • Bayesian Methods and Mixture Models
  • Bacterial Genetics and Biotechnology
  • Mathematical Approximation and Integration
  • Generative Adversarial Networks and Image Synthesis

UNSW Sydney
2020-2024

Anhui Normal University
2023

Beijing Institute of Technology
2023

King Abdullah University of Science and Technology
2021-2023

Kootenay Association for Science & Technology
2023

University of Technology Malaysia
2023

Shanghai Police College
2023

MediaTek (China)
2020

Xi'an Jiaotong University
2018

Stanford University
2014

The limited availability of annotated data often hinders real-world applications machine learning. To efficiently learn from small quantities multimodal data, we leverage the linguistic knowledge a large pre-trained language model (PLM) and quickly adapt it to new domains image captioning. effectively utilize pretrained model, is critical balance visual input prior pretraining. We propose VisualGPT, which employs novel self-resurrecting encoder-decoder attention mechanism PLM with amount...

10.1109/cvpr52688.2022.01750 article EN 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2022-06-01

Protein engineering faces challenges in finding optimal mutants from a massive pool of candidate mutants. In this study, we introduce deep-learning-based data-efficient fitness prediction tool to steer protein engineering. Our methodology establishes lightweight graph neural network scheme for structures, which efficiently analyzes the microenvironment amino acids wild-type proteins and reconstructs distribution acid sequences that are more likely pass natural selection. This serves as...

10.1021/acs.jcim.4c00036 article EN Journal of Chemical Information and Modeling 2024-04-17

Popular post-training pruning methods such as Wanda and RIA are known for their simple, yet effective, designs that have shown exceptional empirical performance. optimizes performance through calibrated activations during pruning, while emphasizes the relative, rather than absolute, importance of weight elements. Despite practical success, a thorough theoretical foundation explaining these outcomes has been lacking. This paper introduces new insights redefine standard minimization objective...

10.48550/arxiv.2501.18980 preprint EN arXiv (Cornell University) 2025-01-31

Inverse protein folding is challenging due to its inherent one-to-many mapping characteristic, where numerous possible amino acid sequences can fold into a single, identical backbone. This task involves not only identifying viable but also representing the sheer diversity of potential solutions. However, existing discriminative models, such as transformer-based auto-regressive struggle encapsulate diverse range plausible In contrast, diffusion probabilistic an emerging genre generative...

10.48550/arxiv.2306.16819 preprint EN cc-by arXiv (Cornell University) 2023-01-01

10.1109/icme57554.2024.10687685 article EN 2022 IEEE International Conference on Multimedia and Expo (ICME) 2024-07-15

The transparent cornea is the window of eye, facilitating entry light rays and controlling focusing movement within eye. critical, contributing to 75% refractive power Keratoconus a progressive multifactorial corneal degenerative disease affecting 1 in 2000 individuals worldwide. Currently, there no cure for keratoconus other than transplantation advanced stage or cross-linking, which can only halt KC progression. ability accurately identify subtle progression vital clinical significance. To...

10.1109/ijcnn48605.2020.9206694 article EN 2022 International Joint Conference on Neural Networks (IJCNN) 2020-07-01

Abstract Deep learning-based methods for generating functional proteins address the growing need novel biocatalysts, allowing precise tailoring of functionalities to meet specific requirements. This emergence leads creation highly efficient and specialized with wide-ranging applications in scientific, technological, biomedical domains. study establishes a pipeline protein sequence generation conditional diffusion model, namely CPDiffusion, deliver diverse sequences enhanced functions....

10.1101/2023.08.10.552783 preprint EN bioRxiv (Cold Spring Harbor Laboratory) 2023-08-14

Abstract Protein engineering faces challenges in finding optimal mutants from the massive pool of candidate mutants. In this study, we introduce a deep learning-based data-efficient fitness prediction tool to steer protein engineering. Our methodology establishes lightweight graph neural network scheme for structures, which efficiently analyzes microenvironment amino acids wild-type proteins and reconstructs distribution acid sequences that are more likely pass natural selection. This serves...

10.1101/2023.11.05.565665 preprint EN bioRxiv (Cold Spring Harbor Laboratory) 2023-11-05

Learning novel concepts, remembering previous knowledge, and adapting it to future tasks occur simultaneously throughout a human's lifetime. To model such comprehensive abilities, continual zero-shot learning (CZSL) has recently been introduced. However, most existing methods overused unseen semantic information that may not be continually accessible in realistic settings. In this paper, we address the challenge of where is provided during training, by leveraging generative modeling. The...

10.1109/iccv51070.2023.01063 article EN 2021 IEEE/CVF International Conference on Computer Vision (ICCV) 2023-10-01

Cosmic microwave background radiation (CMB) is critical to the understanding of early universe and precise estimation cosmological constants. Due contamination thermal dust noise in galaxy, CMB map that an image on two-dimensional sphere has missing observations, mainly concentrated equatorial region. The a significant impact precision for parameters. Inpainting can effectively reduce uncertainty parametric estimation. In this paper, we propose deep learning-based variational autoencoder -...

10.1109/ijcnn48605.2020.9207123 article EN 2022 International Joint Conference on Neural Networks (IJCNN) 2020-07-01

Neural message passing is a basic feature extraction unit for graph-structured data considering neighboring node features in network propagation from one layer to the next. We model such process by an interacting particle system with attractive and repulsive forces Allen-Cahn force arising modeling of phase transition. The dynamics reaction-diffusion which can separate particles without blowing up. This induces (ACMP) graph neural networks where numerical iteration solution constitutes...

10.48550/arxiv.2206.05437 preprint EN other-oa arXiv (Cornell University) 2022-01-01

At present, the performance of deep neural network in general object detection is comparable to or even surpasses that human beings. However, due limitations learning itself, small proportion feature pixels, and occurence blur occlusion, objects complex scenes still an open question. But we can not deny real-time accurate fundamental automatic perception subsequent perception-based decision-making planning tasks autonomous driving. Considering characteristics driving scene, proposed a novel...

10.48550/arxiv.1803.05263 preprint EN other-oa arXiv (Cornell University) 2018-01-01

Receiver Operating Characteristic (ROC) surfaces have been studied in the literature essentially during last decade and are considered as a natural generalization of ROC curves three-class problems. The volume under surface (VUS) is useful for evaluating performance trichotomous diagnostic system or classifier's overall accuracy when possible disease condition sample belongs to one three ordered categories. In areas medical studies machine learning, VUS new statistical model typically...

10.1109/access.2020.3011159 article EN cc-by IEEE Access 2020-01-01

The ability to quickly learn from a small quantity oftraining data widens the range of machine learning applications. In this paper, we propose data-efficient image captioning model, VisualGPT, which leverages linguistic knowledge large pretrained language model(LM). A crucial challenge is balance between use visual information in and prior acquired pretraining. We designed novel self-resurrecting encoder-decoder attention mechanism adapt LM as decoder ona amount in-domain training data....

10.48550/arxiv.2102.10407 preprint EN cc-by-nc-sa arXiv (Cornell University) 2021-01-01

Federated Learning (FL) has garnered increasing attention due to its unique characteristic of allowing heterogeneous clients process their private data locally and interact with a central server, while being respectful privacy. A critical bottleneck in FL is the communication cost. pivotal strategy mitigate this burden \emph{Local Training}, which involves running multiple local stochastic gradient descent iterations between phases. Our work inspired by innovative \emph{Scaffnew} algorithm,...

10.48550/arxiv.2403.09904 preprint EN arXiv (Cornell University) 2024-03-14

Tabular data plays a crucial role in various domains but often suffers from missing values, thereby curtailing its potential utility. Traditional imputation techniques frequently yield suboptimal results and impose substantial computational burdens, leading to inaccuracies subsequent modeling tasks. To address these challenges, we propose DiffImpute, novel Denoising Diffusion Probabilistic Model (DDPM). Specifically, DiffImpute is trained on complete tabular datasets, ensuring that it can...

10.48550/arxiv.2403.13863 preprint EN arXiv (Cornell University) 2024-03-20

The interest in federated learning has surged recent research due to its unique ability train a global model using privacy-secured information held locally on each client. This paper pays particular attention the issue of client-side heterogeneity, pervasive challenge practical implementation FL that escalates complexity. Assuming scenario where client possesses varied memory storage, processing capabilities and network bandwidth - phenomenon referred as system heterogeneity there is...

10.48550/arxiv.2404.09816 preprint EN arXiv (Cornell University) 2024-04-15
Coming Soon ...