It took me a long while (~ 4 days?) to understand the theories of Variation autoencoder (VAE) and how to actually implement it. And it’s not entirely my fault, because:

  • The original paper (Auto-Encoding Variational Bayes) explains
Read more...

Techy

Introduction

I spent the last week of 2018 and the first week of 2019 preparing CityGML data for a machine learning project. In particular, I need to extract 3D point cloud representation of individual buildings in New York (and Berlin and Zurich)… Read more...

Techy

This post describes how to implement stochastic gradient in a distributed fashion with MPI. It will cover the following topics in a high-level fashion, as it is challenging to cover every details in a single post. I will point to other resources… Read more...

Techy

这篇文章的 key idea 是,把关于文档结构的层级结构信息加入模型,有助于生成更好的文本表征。

这里的文档结构主要是说文章由句子组成,是层级结构。之前的方法是把所有句子连成一起输入一个 RNN 模型,这样其实丢失了段落这样的层级结构。

相应地,另一种方法是把每个句子里的词先分步输入一个 RNN 模型,生成句子表征;再将所有的句子表征输入另一个RNN模型,生成… Read more...

Random_old

This post describes how to write CUDA C code to perform 2D convolution on GPU with tiling technique. It is very brief, only covers basic concepts but with links to code examples. We are not going to use cuDNN, only the bare bones of CUDA. But hereRead more...

Techy

This post follows this post last year about vanilla Recurrent Neural Network structures.

One of the ultimate goal of a recurrent neural network is to summarize a sequence of vectors into a single vector representing all the information.

In… Read more...

Techy

这几天(上周)在看 pointnet 系列的论文,具体研究问题是如何处理点云数据。大的研究领域是机器学习如何处理3D视觉问题,在无人车和机器人领域有应用,因为 depth camera,传感器,激光雷达传回来的图片有些是 point cloud。

点云数据是用点来代表三维图像的一种数据格式,例如一张椅子的三维图。从椅子平面上抽取2400个点,每个点是 (x, y, z) 的一个三维向量,那么把… Read more...

Techy

So I set myself another task the other day: writing tests for an open-source project —- a baby step towards software engineering! This post describes the tools and practices I’ve learnt.

The project I’ve committed to is… Read more...

Techy

For the past two weeks I’ve been playing around with ROS. All was fine until I tried to embed tensorflow and “deep learning” stuff into ROS with conda.

Things went south immediately. Pipelines broke down, what used to work… Read more...

Techy

For the past two months I spent many hours solving algorithmic problems. It’s fun, but very time-consuming. I think the majority of time is spent on trying to find accurate solutions and systematic ways for problem solving. I also spent… Read more...

Techy