The Problem I EncounteredSo I have updated the Linux Kernel of a server of our lab yesterday, and hence many packages need to be upgraded to the compatible version as well. There are a few dependency problems during the package updating process, but wi...

---

Published on December 08, 2023 | 3 min read

Recently I am working on a new Multiple Instance Learning (MIL) framework that can couple the patch representation learning with the bag-level classification. The process has been pretty hard, since I also have to deal with other business at the same time. However, I am still very exciting at ...

---

Published on May 05, 2023 | 3 min read

As the third year of this COVID-19 pandemic (and seems like the last year), 2022 is an ordinary but somewhat also special year for me. I got rejected for three times on a paper, helped an undergraduated student, upgraded several machines, and started the very beginning of a visiting tour to Ho...

---

Published on March 20, 2023 | 3 min read

I have to admit that pathology data are really hard to deal with.Whole Slide ImagesA pathology dataset contains many Whole Slide Images (WSI). WSIs are designed to be like microscope images. Therefore, it also includes images of different magnificati...

---

Published on November 07, 2022 | 4 min read

Currently I am working on reseaches related to pathology dataset, and as you may know these images are usually extremely large. For example, a .tif whole slide image from a Phillips machine can cost me about 2.3G disk space and remember, remember that it is just ONE SINGLE image. For my pathol...

---

Published on September 20, 2022 | 3 min read

Previously I have spent some time researching on pure MLP-based segmentation methods. I did find some amazing work that play all kinds of tricks on MLP, but still, i give up on segmentation MLPs in the end. Generally speaking, MLPs are not suitable for dense prediction tasks unless some ground...

---

Published on September 05, 2022 | 3 min read

I was thinking about using probablity density function to describe a parameter matrix. I dont know if it is feasible but theoretically speaking it is at least “runnable”.Why we need model compressionThere are many reasons to employ model c...

---

Published on July 20, 2022 | 1 min read

Convolution has become dominant in visual tasks for a very long time. Recent years, people argue that convolution lacks the ability to model long-range dependencies and propose to use Non-Lcoal operations to make up this flaw. For example, in CVPR2018, Xiaolong Wang et al. propose to use Self-...

---

Published on April 06, 2022 | 2 min read

Recently I am working on designing 3D Vision Transformers for medical images. I have found many problems in my experiments, and i will list some of them in this blog.Challenges 1: Computational CostComputational cost can be mainly devide...

---

Published on January 10, 2022 | 3 min read

Recently one of my reviewers asks me to conduct t-test on my model to verify if it is statistically different from other sota methods. In this blog, I will briefly introduce t-test and tell you how to use it to evaluate your deep learning models.What is T-Test?</h...

---

Published on January 05, 2022 | 3 min read

As the second year of this COVID-19 pandemic, 2021 is actually not so bad. Even though our movement was restricted, there are still some unique experience that worth remembering.1. Attending an International ConferenceThis year s...

---

Published on December 29, 2021 | 2 min read

Decomposition, as the most intuitive method for reducing memory cost, has been forgotten by people since they usually leads to performance degration as well. This article reviews decomposition in deep learning and also introduces some great work in this field.In deep learning era, GPU...

---

Published on December 20, 2021 | 2 min read

Hello, my name is Hongyi and this website is used as my personal page to post new information or blogs. You can find my related publications in the header-&gt;more-&gt;publications. You can ask any questions about my works through any channel I provided publicly (such as leave a comment under ...

---

Published on December 17, 2021 | 0 min read